import FIREFOX_52_6_0esr_RELEASE from mozilla-esr52 hg repo

dom/media/webspeech/moz.build

@@ -0,0 +1,9 @@
+# vim: set filetype=python:
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+DIRS = ['synth']
+
+if CONFIG['MOZ_WEBSPEECH']:
+    DIRS += ['recognition']

dom/media/webspeech/recognition/PocketSphinxSpeechRecognitionService.cpp

@@ -0,0 +1,357 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsThreadUtils.h"
+#include "nsXPCOMCIDInternal.h"
+#include "PocketSphinxSpeechRecognitionService.h"
+#include "nsIFile.h"
+#include "SpeechGrammar.h"
+#include "SpeechRecognition.h"
+#include "SpeechRecognitionAlternative.h"
+#include "SpeechRecognitionResult.h"
+#include "SpeechRecognitionResultList.h"
+#include "nsIObserverService.h"
+#include "MediaPrefs.h"
+#include "mozilla/Services.h"
+#include "nsDirectoryServiceDefs.h"
+#include "nsDirectoryServiceUtils.h"
+#include "nsMemory.h"
+
+extern "C" {
+#include "pocketsphinx/pocketsphinx.h"
+#include "sphinxbase/logmath.h"
+#include "sphinxbase/sphinx_config.h"
+#include "sphinxbase/jsgf.h"
+}
+
+namespace mozilla {
+
+using namespace dom;
+
+class DecodeResultTask : public Runnable
+{
+public:
+  DecodeResultTask(const nsString& hypstring,
+                   float64 confidence,
+                   WeakPtr<dom::SpeechRecognition> recognition)
+      : mResult(hypstring),
+        mConfidence(confidence),
+        mRecognition(recognition),
+        mWorkerThread(do_GetCurrentThread())
+  {
+    MOZ_ASSERT(
+      !NS_IsMainThread()); // This should be running on the worker thread
+  }
+
+  NS_IMETHOD
+  Run() override
+  {
+    MOZ_ASSERT(NS_IsMainThread()); // This method is supposed to run on the main
+                                   // thread!
+
+    // Declare javascript result events
+    RefPtr<SpeechEvent> event = new SpeechEvent(
+      mRecognition, SpeechRecognition::EVENT_RECOGNITIONSERVICE_FINAL_RESULT);
+    SpeechRecognitionResultList* resultList =
+      new SpeechRecognitionResultList(mRecognition);
+    SpeechRecognitionResult* result = new SpeechRecognitionResult(mRecognition);
+    if (0 < mRecognition->MaxAlternatives()) {
+      SpeechRecognitionAlternative* alternative =
+        new SpeechRecognitionAlternative(mRecognition);
+
+      alternative->mTranscript = mResult;
+      alternative->mConfidence = mConfidence;
+
+      result->mItems.AppendElement(alternative);
+    }
+    resultList->mItems.AppendElement(result);
+
+    event->mRecognitionResultList = resultList;
+    NS_DispatchToMainThread(event);
+
+    // If we don't destroy the thread when we're done with it, it will hang
+    // around forever... bad!
+    // But thread->Shutdown must be called from the main thread, not from the
+    // thread itself.
+    return mWorkerThread->Shutdown();
+  }
+
+private:
+  nsString mResult;
+  float64 mConfidence;
+  WeakPtr<dom::SpeechRecognition> mRecognition;
+  nsCOMPtr<nsIThread> mWorkerThread;
+};
+
+class DecodeTask : public Runnable
+{
+public:
+  DecodeTask(WeakPtr<dom::SpeechRecognition> recogntion,
+             const nsTArray<int16_t>& audiovector, ps_decoder_t* ps)
+      : mRecognition(recogntion), mAudiovector(audiovector), mPs(ps)
+  {
+  }
+
+  NS_IMETHOD
+  Run() override
+  {
+    char const* hyp;
+    int rv;
+    int32 final;
+    int32 logprob;
+    float64 confidence;
+    nsAutoCString hypoValue;
+
+    rv = ps_start_utt(mPs);
+    rv = ps_process_raw(mPs, &mAudiovector[0], mAudiovector.Length(), FALSE,
+                        FALSE);
+
+    rv = ps_end_utt(mPs);
+    confidence = 0;
+    if (rv >= 0) {
+      hyp = ps_get_hyp_final(mPs, &final);
+      if (hyp && final) {
+        logprob = ps_get_prob(mPs);
+        confidence = logmath_exp(ps_get_logmath(mPs), logprob);
+        hypoValue.Assign(hyp);
+      }
+    }
+
+    nsCOMPtr<nsIRunnable> resultrunnable =
+      new DecodeResultTask(NS_ConvertUTF8toUTF16(hypoValue), confidence, mRecognition);
+    return NS_DispatchToMainThread(resultrunnable);
+  }
+
+private:
+  WeakPtr<dom::SpeechRecognition> mRecognition;
+  nsTArray<int16_t> mAudiovector;
+  ps_decoder_t* mPs;
+};
+
+NS_IMPL_ISUPPORTS(PocketSphinxSpeechRecognitionService,
+                  nsISpeechRecognitionService, nsIObserver)
+
+PocketSphinxSpeechRecognitionService::PocketSphinxSpeechRecognitionService()
+{
+  mSpeexState = nullptr;
+
+  // get root folder
+  nsCOMPtr<nsIFile> tmpFile;
+  nsAutoString aStringAMPath;   // am folder
+  nsAutoString aStringDictPath; // dict folder
+
+  NS_GetSpecialDirectory(NS_GRE_DIR, getter_AddRefs(tmpFile));
+#if defined(XP_WIN) // for some reason, on windows NS_GRE_DIR is not bin root,
+                    // but bin/browser
+  tmpFile->AppendRelativePath(NS_LITERAL_STRING(".."));
+#endif
+  tmpFile->AppendRelativePath(NS_LITERAL_STRING("models"));
+  tmpFile->AppendRelativePath(NS_LITERAL_STRING("en-US"));
+  tmpFile->GetPath(aStringAMPath);
+
+  NS_GetSpecialDirectory(NS_GRE_DIR, getter_AddRefs(tmpFile));
+#if defined(XP_WIN) // for some reason, on windows NS_GRE_DIR is not bin root,
+                    // but bin/browser
+  tmpFile->AppendRelativePath(NS_LITERAL_STRING(".."));
+#endif
+  tmpFile->AppendRelativePath(NS_LITERAL_STRING("models"));     //
+  tmpFile->AppendRelativePath(NS_LITERAL_STRING("dict"));       //
+  tmpFile->AppendRelativePath(NS_LITERAL_STRING("en-US.dic")); //
+  tmpFile->GetPath(aStringDictPath);
+
+  // FOR B2G PATHS HARDCODED (APPEND /DATA ON THE BEGINING, FOR DESKTOP, ONLY
+  // MODELS/ RELATIVE TO ROOT
+  mPSConfig = cmd_ln_init(nullptr, ps_args(), TRUE, "-bestpath", "yes", "-hmm",
+                          ToNewUTF8String(aStringAMPath), // acoustic model
+                          "-dict", ToNewUTF8String(aStringDictPath), nullptr);
+  if (mPSConfig == nullptr) {
+    ISDecoderCreated = false;
+  } else {
+    mPSHandle = ps_init(mPSConfig);
+    if (mPSHandle == nullptr) {
+      ISDecoderCreated = false;
+    } else {
+      ISDecoderCreated = true;
+    }
+  }
+
+  ISGrammarCompiled = false;
+}
+
+PocketSphinxSpeechRecognitionService::~PocketSphinxSpeechRecognitionService()
+{
+  if (mPSConfig) {
+    free(mPSConfig);
+  }
+  if (mPSHandle) {
+    free(mPSHandle);
+  }
+
+  mSpeexState = nullptr;
+}
+
+// CALL START IN JS FALLS HERE
+NS_IMETHODIMP
+PocketSphinxSpeechRecognitionService::Initialize(
+    WeakPtr<SpeechRecognition> aSpeechRecognition)
+{
+  if (!ISDecoderCreated || !ISGrammarCompiled) {
+    return NS_ERROR_NOT_INITIALIZED;
+  } else {
+    mAudioVector.Clear();
+
+    if (mSpeexState) {
+      mSpeexState = nullptr;
+    }
+
+    mRecognition = aSpeechRecognition;
+    nsCOMPtr<nsIObserverService> obs = services::GetObserverService();
+    obs->AddObserver(this, SPEECH_RECOGNITION_TEST_EVENT_REQUEST_TOPIC, false);
+    obs->AddObserver(this, SPEECH_RECOGNITION_TEST_END_TOPIC, false);
+    return NS_OK;
+  }
+}
+
+NS_IMETHODIMP
+PocketSphinxSpeechRecognitionService::ProcessAudioSegment(
+  AudioSegment* aAudioSegment, int32_t aSampleRate)
+{
+  if (!mSpeexState) {
+    mSpeexState = speex_resampler_init(1, aSampleRate, 16000,
+                                       SPEEX_RESAMPLER_QUALITY_MAX, nullptr);
+  }
+  aAudioSegment->ResampleChunks(mSpeexState, aSampleRate, 16000);
+
+  AudioSegment::ChunkIterator iterator(*aAudioSegment);
+
+  while (!iterator.IsEnded()) {
+    mozilla::AudioChunk& chunk = *(iterator);
+    MOZ_ASSERT(chunk.mBuffer);
+    const int16_t* buf = static_cast<const int16_t*>(chunk.mChannelData[0]);
+
+    for (int i = 0; i < iterator->mDuration; i++) {
+      mAudioVector.AppendElement((int16_t)buf[i]);
+    }
+    iterator.Next();
+  }
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+PocketSphinxSpeechRecognitionService::SoundEnd()
+{
+  speex_resampler_destroy(mSpeexState);
+  mSpeexState = nullptr;
+
+  // To create a new thread, get the thread manager
+  nsCOMPtr<nsIThreadManager> tm = do_GetService(NS_THREADMANAGER_CONTRACTID);
+  nsCOMPtr<nsIThread> decodethread;
+  nsresult rv = tm->NewThread(0, 0, getter_AddRefs(decodethread));
+  if (NS_FAILED(rv)) {
+    // In case of failure, call back immediately with an empty string which
+    // indicates failure
+    return NS_OK;
+  }
+
+  nsCOMPtr<nsIRunnable> r =
+    new DecodeTask(mRecognition, mAudioVector, mPSHandle);
+  decodethread->Dispatch(r, nsIEventTarget::DISPATCH_NORMAL);
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+PocketSphinxSpeechRecognitionService::ValidateAndSetGrammarList(
+  SpeechGrammar* aSpeechGrammar,
+  nsISpeechGrammarCompilationCallback* aCallback)
+{
+  if (!ISDecoderCreated) {
+    ISGrammarCompiled = false;
+  } else if (aSpeechGrammar) {
+    nsAutoString grammar;
+    ErrorResult rv;
+    aSpeechGrammar->GetSrc(grammar, rv);
+
+    int result = ps_set_jsgf_string(mPSHandle, "name",
+                                    NS_ConvertUTF16toUTF8(grammar).get());
+
+    if (result != 0) {
+      ISGrammarCompiled = false;
+    } else {
+      ps_set_search(mPSHandle, "name");
+
+      ISGrammarCompiled = true;
+    }
+  } else {
+    ISGrammarCompiled = false;
+  }
+
+  return ISGrammarCompiled ? NS_OK : NS_ERROR_NOT_INITIALIZED;
+}
+
+NS_IMETHODIMP
+PocketSphinxSpeechRecognitionService::Abort()
+{
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+PocketSphinxSpeechRecognitionService::Observe(nsISupports* aSubject,
+                                              const char* aTopic,
+                                              const char16_t* aData)
+{
+  MOZ_ASSERT(MediaPrefs::WebSpeechFakeRecognitionService(),
+             "Got request to fake recognition service event, "
+             "but " TEST_PREFERENCE_FAKE_RECOGNITION_SERVICE " is not set");
+
+  if (!strcmp(aTopic, SPEECH_RECOGNITION_TEST_END_TOPIC)) {
+    nsCOMPtr<nsIObserverService> obs = services::GetObserverService();
+    obs->RemoveObserver(this, SPEECH_RECOGNITION_TEST_EVENT_REQUEST_TOPIC);
+    obs->RemoveObserver(this, SPEECH_RECOGNITION_TEST_END_TOPIC);
+
+    return NS_OK;
+  }
+
+  const nsDependentString eventName = nsDependentString(aData);
+
+  if (eventName.EqualsLiteral("EVENT_RECOGNITIONSERVICE_ERROR")) {
+    mRecognition->DispatchError(
+      SpeechRecognition::EVENT_RECOGNITIONSERVICE_ERROR,
+      SpeechRecognitionErrorCode::Network, // TODO different codes?
+      NS_LITERAL_STRING("RECOGNITIONSERVICE_ERROR test event"));
+
+  } else if (eventName.EqualsLiteral("EVENT_RECOGNITIONSERVICE_FINAL_RESULT")) {
+    RefPtr<SpeechEvent> event = new SpeechEvent(
+      mRecognition, SpeechRecognition::EVENT_RECOGNITIONSERVICE_FINAL_RESULT);
+
+    event->mRecognitionResultList = BuildMockResultList();
+    NS_DispatchToMainThread(event);
+  }
+
+  return NS_OK;
+}
+
+SpeechRecognitionResultList*
+PocketSphinxSpeechRecognitionService::BuildMockResultList()
+{
+  SpeechRecognitionResultList* resultList =
+    new SpeechRecognitionResultList(mRecognition);
+  SpeechRecognitionResult* result = new SpeechRecognitionResult(mRecognition);
+  if (0 < mRecognition->MaxAlternatives()) {
+    SpeechRecognitionAlternative* alternative =
+      new SpeechRecognitionAlternative(mRecognition);
+
+    alternative->mTranscript = NS_LITERAL_STRING("Mock final result");
+    alternative->mConfidence = 0.0f;
+
+    result->mItems.AppendElement(alternative);
+  }
+  resultList->mItems.AppendElement(result);
+
+  return resultList;
+}
+
+} // namespace mozilla

dom/media/webspeech/recognition/PocketSphinxSpeechRecognitionService.h

@@ -0,0 +1,85 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_PocketSphinxRecognitionService_h
+#define mozilla_dom_PocketSphinxRecognitionService_h
+
+#include "nsCOMPtr.h"
+#include "nsTArray.h"
+#include "nsIObserver.h"
+#include "nsISpeechRecognitionService.h"
+#include "speex/speex_resampler.h"
+
+extern "C" {
+#include <pocketsphinx/pocketsphinx.h>
+#include <sphinxbase/sphinx_config.h>
+}
+
+#define NS_POCKETSPHINX_SPEECH_RECOGNITION_SERVICE_CID                         \
+  {                                                                            \
+    0x0ff5ce56, 0x5b09, 0x4db8, {                                              \
+      0xad, 0xc6, 0x82, 0x66, 0xaf, 0x95, 0xf8, 0x64                           \
+    }                                                                          \
+  };
+
+namespace mozilla {
+
+/**
+ * Pocketsphix implementation of the nsISpeechRecognitionService interface
+ */
+class PocketSphinxSpeechRecognitionService : public nsISpeechRecognitionService,
+                                             public nsIObserver
+{
+public:
+  // Add XPCOM glue code
+  NS_DECL_ISUPPORTS
+  NS_DECL_NSISPEECHRECOGNITIONSERVICE
+
+  // Add nsIObserver code
+  NS_DECL_NSIOBSERVER
+
+  /**
+   * Default constructs a PocketSphinxSpeechRecognitionService loading default
+   * files
+   */
+  PocketSphinxSpeechRecognitionService();
+
+private:
+  /**
+   * Private destructor to prevent bypassing of reference counting
+   */
+  virtual ~PocketSphinxSpeechRecognitionService();
+
+  /** The associated SpeechRecognition */
+  WeakPtr<dom::SpeechRecognition> mRecognition;
+
+  /**
+   * Builds a mock SpeechRecognitionResultList
+   */
+  dom::SpeechRecognitionResultList* BuildMockResultList();
+
+  /** Speex state */
+  SpeexResamplerState* mSpeexState;
+
+  /** Pocksphix decoder */
+  ps_decoder_t* mPSHandle;
+
+  /** Sphinxbase parsed command line arguments */
+  cmd_ln_t* mPSConfig;
+
+  /** Flag to verify if decoder was created */
+  bool ISDecoderCreated;
+
+  /** Flag to verify if grammar was compiled */
+  bool ISGrammarCompiled;
+
+  /** Audio data */
+  nsTArray<int16_t> mAudioVector;
+};
+
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/SpeechGrammar.cpp

@@ -0,0 +1,81 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechGrammar.h"
+
+#include "mozilla/dom/SpeechGrammarBinding.h"
+
+namespace mozilla {
+namespace dom {
+
+NS_IMPL_CYCLE_COLLECTION_WRAPPERCACHE(SpeechGrammar, mParent)
+NS_IMPL_CYCLE_COLLECTING_ADDREF(SpeechGrammar)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(SpeechGrammar)
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(SpeechGrammar)
+  NS_WRAPPERCACHE_INTERFACE_MAP_ENTRY
+  NS_INTERFACE_MAP_ENTRY(nsISupports)
+NS_INTERFACE_MAP_END
+
+SpeechGrammar::SpeechGrammar(nsISupports* aParent)
+  : mParent(aParent)
+{
+}
+
+SpeechGrammar::~SpeechGrammar()
+{
+}
+
+already_AddRefed<SpeechGrammar>
+SpeechGrammar::Constructor(const GlobalObject& aGlobal,
+                           ErrorResult& aRv)
+{
+  RefPtr<SpeechGrammar> speechGrammar =
+    new SpeechGrammar(aGlobal.GetAsSupports());
+  return speechGrammar.forget();
+}
+
+nsISupports*
+SpeechGrammar::GetParentObject() const
+{
+  return mParent;
+}
+
+JSObject*
+SpeechGrammar::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto)
+{
+  return SpeechGrammarBinding::Wrap(aCx, this, aGivenProto);
+}
+
+void
+SpeechGrammar::GetSrc(nsString& aRetVal, ErrorResult& aRv) const
+{
+  aRetVal = mSrc;
+  return;
+}
+
+void
+SpeechGrammar::SetSrc(const nsAString& aArg, ErrorResult& aRv)
+{
+  mSrc = aArg;
+  return;
+}
+
+float
+SpeechGrammar::GetWeight(ErrorResult& aRv) const
+{
+  aRv.Throw(NS_ERROR_NOT_IMPLEMENTED);
+  return 0;
+}
+
+void
+SpeechGrammar::SetWeight(float aArg, ErrorResult& aRv)
+{
+  aRv.Throw(NS_ERROR_NOT_IMPLEMENTED);
+  return;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/recognition/SpeechGrammar.h

@@ -0,0 +1,59 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechGrammar_h
+#define mozilla_dom_SpeechGrammar_h
+
+#include "nsCOMPtr.h"
+#include "nsCycleCollectionParticipant.h"
+#include "nsString.h"
+#include "nsWrapperCache.h"
+#include "js/TypeDecls.h"
+
+#include "mozilla/Attributes.h"
+#include "mozilla/ErrorResult.h"
+
+namespace mozilla {
+namespace dom {
+
+class GlobalObject;
+
+class SpeechGrammar final : public nsISupports,
+                            public nsWrapperCache
+{
+public:
+  explicit SpeechGrammar(nsISupports* aParent);
+
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_SCRIPT_HOLDER_CLASS(SpeechGrammar)
+
+  nsISupports* GetParentObject() const;
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  static already_AddRefed<SpeechGrammar>
+    Constructor(const GlobalObject& aGlobal, ErrorResult& aRv);
+
+  void GetSrc(nsString& aRetVal, ErrorResult& aRv) const;
+
+  void SetSrc(const nsAString& aArg, ErrorResult& aRv);
+
+  float GetWeight(ErrorResult& aRv) const;
+
+  void SetWeight(float aArg, ErrorResult& aRv);
+
+private:
+  ~SpeechGrammar();
+
+  nsCOMPtr<nsISupports> mParent;
+
+  nsString mSrc;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/SpeechGrammarList.cpp

@@ -0,0 +1,103 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechGrammarList.h"
+
+#include "mozilla/dom/SpeechGrammarListBinding.h"
+#include "mozilla/ErrorResult.h"
+#include "nsCOMPtr.h"
+#include "nsXPCOMStrings.h"
+#include "SpeechRecognition.h"
+
+namespace mozilla {
+namespace dom {
+
+NS_IMPL_CYCLE_COLLECTION_WRAPPERCACHE(SpeechGrammarList, mParent, mItems)
+NS_IMPL_CYCLE_COLLECTING_ADDREF(SpeechGrammarList)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(SpeechGrammarList)
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(SpeechGrammarList)
+  NS_WRAPPERCACHE_INTERFACE_MAP_ENTRY
+  NS_INTERFACE_MAP_ENTRY(nsISupports)
+NS_INTERFACE_MAP_END
+
+SpeechGrammarList::SpeechGrammarList(nsISupports* aParent)
+  : mParent(aParent)
+{
+}
+
+SpeechGrammarList::~SpeechGrammarList()
+{
+}
+
+already_AddRefed<SpeechGrammarList>
+SpeechGrammarList::Constructor(const GlobalObject& aGlobal,
+                               ErrorResult& aRv)
+{
+  RefPtr<SpeechGrammarList> speechGrammarList =
+    new SpeechGrammarList(aGlobal.GetAsSupports());
+  return speechGrammarList.forget();
+}
+
+JSObject*
+SpeechGrammarList::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto)
+{
+  return SpeechGrammarListBinding::Wrap(aCx, this, aGivenProto);
+}
+
+nsISupports*
+SpeechGrammarList::GetParentObject() const
+{
+  return mParent;
+}
+
+uint32_t
+SpeechGrammarList::Length() const
+{
+  return mItems.Length();
+}
+
+already_AddRefed<SpeechGrammar>
+SpeechGrammarList::Item(uint32_t aIndex, ErrorResult& aRv)
+{
+  RefPtr<SpeechGrammar> result = mItems.ElementAt(aIndex);
+  return result.forget();
+}
+
+void
+SpeechGrammarList::AddFromURI(const nsAString& aSrc,
+                              const Optional<float>& aWeight,
+                              ErrorResult& aRv)
+{
+  aRv.Throw(NS_ERROR_NOT_IMPLEMENTED);
+  return;
+}
+
+void
+SpeechGrammarList::AddFromString(const nsAString& aString,
+                                 const Optional<float>& aWeight,
+                                 ErrorResult& aRv)
+{
+  SpeechGrammar* speechGrammar = new SpeechGrammar(mParent);
+  speechGrammar->SetSrc(aString, aRv);
+  mItems.AppendElement(speechGrammar);
+  return;
+}
+
+already_AddRefed<SpeechGrammar>
+SpeechGrammarList::IndexedGetter(uint32_t aIndex, bool& aPresent,
+                                 ErrorResult& aRv)
+{
+  if (aIndex >= Length()) {
+    aPresent = false;
+    return nullptr;
+  }
+  ErrorResult rv;
+  aPresent = true;
+  return Item(aIndex, rv);
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/recognition/SpeechGrammarList.h

@@ -0,0 +1,63 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechGrammarList_h
+#define mozilla_dom_SpeechGrammarList_h
+
+#include "mozilla/Attributes.h"
+#include "nsCOMPtr.h"
+#include "nsCycleCollectionParticipant.h"
+#include "nsWrapperCache.h"
+
+struct JSContext;
+
+namespace mozilla {
+
+class ErrorResult;
+
+namespace dom {
+
+class GlobalObject;
+class SpeechGrammar;
+template<typename> class Optional;
+
+class SpeechGrammarList final : public nsISupports,
+                                public nsWrapperCache
+{
+public:
+  explicit SpeechGrammarList(nsISupports* aParent);
+
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_SCRIPT_HOLDER_CLASS(SpeechGrammarList)
+
+  static already_AddRefed<SpeechGrammarList> Constructor(const GlobalObject& aGlobal, ErrorResult& aRv);
+
+  nsISupports* GetParentObject() const;
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  uint32_t Length() const;
+
+  already_AddRefed<SpeechGrammar> Item(uint32_t aIndex, ErrorResult& aRv);
+
+  void AddFromURI(const nsAString& aSrc, const Optional<float>& aWeight, ErrorResult& aRv);
+
+  void AddFromString(const nsAString& aString, const Optional<float>& aWeight, ErrorResult& aRv);
+
+  already_AddRefed<SpeechGrammar> IndexedGetter(uint32_t aIndex, bool& aPresent, ErrorResult& aRv);
+
+private:
+  ~SpeechGrammarList();
+
+  nsCOMPtr<nsISupports> mParent;
+
+  nsTArray<RefPtr<SpeechGrammar>> mItems;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/SpeechRecognition.h

@@ -0,0 +1,296 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechRecognition_h
+#define mozilla_dom_SpeechRecognition_h
+
+#include "mozilla/Attributes.h"
+#include "mozilla/DOMEventTargetHelper.h"
+#include "nsCOMPtr.h"
+#include "nsString.h"
+#include "nsWrapperCache.h"
+#include "nsTArray.h"
+#include "js/TypeDecls.h"
+
+#include "nsIDOMNavigatorUserMedia.h"
+#include "nsITimer.h"
+#include "MediaEngine.h"
+#include "MediaStreamGraph.h"
+#include "AudioSegment.h"
+#include "mozilla/WeakPtr.h"
+
+#include "SpeechGrammarList.h"
+#include "SpeechRecognitionResultList.h"
+#include "SpeechStreamListener.h"
+#include "nsISpeechRecognitionService.h"
+#include "endpointer.h"
+
+#include "mozilla/dom/SpeechRecognitionError.h"
+
+namespace mozilla {
+
+namespace dom {
+
+#define SPEECH_RECOGNITION_TEST_EVENT_REQUEST_TOPIC "SpeechRecognitionTest:RequestEvent"
+#define SPEECH_RECOGNITION_TEST_END_TOPIC "SpeechRecognitionTest:End"
+
+class GlobalObject;
+class SpeechEvent;
+
+LogModule* GetSpeechRecognitionLog();
+#define SR_LOG(...) MOZ_LOG(GetSpeechRecognitionLog(), mozilla::LogLevel::Debug, (__VA_ARGS__))
+
+class SpeechRecognition final : public DOMEventTargetHelper,
+                                public nsIObserver,
+                                public SupportsWeakPtr<SpeechRecognition>
+{
+public:
+  MOZ_DECLARE_WEAKREFERENCE_TYPENAME(SpeechRecognition)
+  explicit SpeechRecognition(nsPIDOMWindowInner* aOwnerWindow);
+
+  NS_DECL_ISUPPORTS_INHERITED
+  NS_DECL_CYCLE_COLLECTION_CLASS_INHERITED(SpeechRecognition, DOMEventTargetHelper)
+
+  NS_DECL_NSIOBSERVER
+
+  nsISupports* GetParentObject() const;
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  static bool IsAuthorized(JSContext* aCx, JSObject* aGlobal);
+
+  static already_AddRefed<SpeechRecognition>
+  Constructor(const GlobalObject& aGlobal, ErrorResult& aRv);
+
+  already_AddRefed<SpeechGrammarList> Grammars() const;
+
+  void SetGrammars(mozilla::dom::SpeechGrammarList& aArg);
+
+  void GetLang(nsString& aRetVal) const;
+
+  void SetLang(const nsAString& aArg);
+
+  bool GetContinuous(ErrorResult& aRv) const;
+
+  void SetContinuous(bool aArg, ErrorResult& aRv);
+
+  bool InterimResults() const;
+
+  void SetInterimResults(bool aArg);
+
+  uint32_t MaxAlternatives() const;
+
+  void SetMaxAlternatives(uint32_t aArg);
+
+  void GetServiceURI(nsString& aRetVal, ErrorResult& aRv) const;
+
+  void SetServiceURI(const nsAString& aArg, ErrorResult& aRv);
+
+  void Start(const Optional<NonNull<DOMMediaStream>>& aStream, ErrorResult& aRv);
+
+  void Stop();
+
+  void Abort();
+
+  IMPL_EVENT_HANDLER(audiostart)
+  IMPL_EVENT_HANDLER(soundstart)
+  IMPL_EVENT_HANDLER(speechstart)
+  IMPL_EVENT_HANDLER(speechend)
+  IMPL_EVENT_HANDLER(soundend)
+  IMPL_EVENT_HANDLER(audioend)
+  IMPL_EVENT_HANDLER(result)
+  IMPL_EVENT_HANDLER(nomatch)
+  IMPL_EVENT_HANDLER(error)
+  IMPL_EVENT_HANDLER(start)
+  IMPL_EVENT_HANDLER(end)
+
+  enum EventType {
+    EVENT_START,
+    EVENT_STOP,
+    EVENT_ABORT,
+    EVENT_AUDIO_DATA,
+    EVENT_AUDIO_ERROR,
+    EVENT_RECOGNITIONSERVICE_INTERMEDIATE_RESULT,
+    EVENT_RECOGNITIONSERVICE_FINAL_RESULT,
+    EVENT_RECOGNITIONSERVICE_ERROR,
+    EVENT_COUNT
+  };
+
+  void DispatchError(EventType aErrorType, SpeechRecognitionErrorCode aErrorCode, const nsAString& aMessage);
+  uint32_t FillSamplesBuffer(const int16_t* aSamples, uint32_t aSampleCount);
+  uint32_t SplitSamplesBuffer(const int16_t* aSamplesBuffer, uint32_t aSampleCount, nsTArray<RefPtr<SharedBuffer>>& aResult);
+  AudioSegment* CreateAudioSegment(nsTArray<RefPtr<SharedBuffer>>& aChunks);
+  void FeedAudioData(already_AddRefed<SharedBuffer> aSamples, uint32_t aDuration, MediaStreamListener* aProvider, TrackRate aTrackRate);
+
+  friend class SpeechEvent;
+private:
+  virtual ~SpeechRecognition() {};
+
+  enum FSMState {
+    STATE_IDLE,
+    STATE_STARTING,
+    STATE_ESTIMATING,
+    STATE_WAITING_FOR_SPEECH,
+    STATE_RECOGNIZING,
+    STATE_WAITING_FOR_RESULT,
+    STATE_COUNT
+  };
+
+  void SetState(FSMState state);
+  bool StateBetween(FSMState begin, FSMState end);
+
+  bool SetRecognitionService(ErrorResult& aRv);
+  bool ValidateAndSetGrammarList(ErrorResult& aRv);
+
+  class GetUserMediaSuccessCallback : public nsIDOMGetUserMediaSuccessCallback
+  {
+  public:
+    NS_DECL_ISUPPORTS
+    NS_DECL_NSIDOMGETUSERMEDIASUCCESSCALLBACK
+
+    explicit GetUserMediaSuccessCallback(SpeechRecognition* aRecognition)
+      : mRecognition(aRecognition)
+    {}
+
+  private:
+    virtual ~GetUserMediaSuccessCallback() {}
+
+    RefPtr<SpeechRecognition> mRecognition;
+  };
+
+  class GetUserMediaErrorCallback : public nsIDOMGetUserMediaErrorCallback
+  {
+  public:
+    NS_DECL_ISUPPORTS
+    NS_DECL_NSIDOMGETUSERMEDIAERRORCALLBACK
+
+    explicit GetUserMediaErrorCallback(SpeechRecognition* aRecognition)
+      : mRecognition(aRecognition)
+    {}
+
+  private:
+    virtual ~GetUserMediaErrorCallback() {}
+
+    RefPtr<SpeechRecognition> mRecognition;
+  };
+
+  NS_IMETHOD StartRecording(DOMMediaStream* aDOMStream);
+  NS_IMETHOD StopRecording();
+
+  uint32_t ProcessAudioSegment(AudioSegment* aSegment, TrackRate aTrackRate);
+  void NotifyError(SpeechEvent* aEvent);
+
+  void ProcessEvent(SpeechEvent* aEvent);
+  void Transition(SpeechEvent* aEvent);
+
+  void Reset();
+  void ResetAndEnd();
+  void WaitForAudioData(SpeechEvent* aEvent);
+  void StartedAudioCapture(SpeechEvent* aEvent);
+  void StopRecordingAndRecognize(SpeechEvent* aEvent);
+  void WaitForEstimation(SpeechEvent* aEvent);
+  void DetectSpeech(SpeechEvent* aEvent);
+  void WaitForSpeechEnd(SpeechEvent* aEvent);
+  void NotifyFinalResult(SpeechEvent* aEvent);
+  void DoNothing(SpeechEvent* aEvent);
+  void AbortSilently(SpeechEvent* aEvent);
+  void AbortError(SpeechEvent* aEvent);
+
+  RefPtr<DOMMediaStream> mDOMStream;
+  RefPtr<SpeechStreamListener> mSpeechListener;
+  nsCOMPtr<nsISpeechRecognitionService> mRecognitionService;
+
+  FSMState mCurrentState;
+
+  Endpointer mEndpointer;
+  uint32_t mEstimationSamples;
+
+  uint32_t mAudioSamplesPerChunk;
+
+  // buffer holds one chunk of mAudioSamplesPerChunk
+  // samples before feeding it to mEndpointer
+  RefPtr<SharedBuffer> mAudioSamplesBuffer;
+  uint32_t mBufferedSamples;
+
+  nsCOMPtr<nsITimer> mSpeechDetectionTimer;
+  bool mAborted;
+
+  nsString mLang;
+
+  RefPtr<SpeechGrammarList> mSpeechGrammarList;
+
+  // WebSpeechAPI (http://bit.ly/1gIl7DC) states:
+  //
+  // 1. Default value MUST be false
+  // 2. If true, interim results SHOULD be returned
+  // 3. If false, interim results MUST NOT be returned
+  //
+  // Pocketsphinx does not return interm results; so, defaulting
+  // mInterimResults to false, then ignoring its subsequent value
+  // is a conforming implementation.
+  bool mInterimResults;
+
+  // WebSpeechAPI (http://bit.ly/1JAiqeo) states:
+  //
+  // 1. Default value is 1
+  // 2. Subsequent value is the "maximum number of SpeechRecognitionAlternatives per result"
+  //
+  // Pocketsphinx can only return at maximum a single SpeechRecognitionAlternative
+  // per SpeechRecognitionResult. So defaulting mMaxAlternatives to 1, for all non
+  // zero values ignoring mMaxAlternatives while for a 0 value returning no
+  // SpeechRecognitionAlternative per result is a conforming implementation.
+  uint32_t mMaxAlternatives;
+
+  void ProcessTestEventRequest(nsISupports* aSubject, const nsAString& aEventName);
+
+  const char* GetName(FSMState aId);
+  const char* GetName(SpeechEvent* aId);
+};
+
+class SpeechEvent : public Runnable
+{
+public:
+  SpeechEvent(SpeechRecognition* aRecognition, SpeechRecognition::EventType aType)
+  : mAudioSegment(0)
+  , mRecognitionResultList(nullptr)
+  , mError(nullptr)
+  , mRecognition(aRecognition)
+  , mType(aType)
+  , mTrackRate(0)
+  {
+  }
+
+  ~SpeechEvent();
+
+  NS_IMETHOD Run() override;
+  AudioSegment* mAudioSegment;
+  RefPtr<SpeechRecognitionResultList> mRecognitionResultList; // TODO: make this a session being passed which also has index and stuff
+  RefPtr<SpeechRecognitionError> mError;
+
+  friend class SpeechRecognition;
+private:
+  SpeechRecognition* mRecognition;
+
+  // for AUDIO_DATA events, keep a reference to the provider
+  // of the data (i.e., the SpeechStreamListener) to ensure it
+  // is kept alive (and keeps SpeechRecognition alive) until this
+  // event gets processed.
+  RefPtr<MediaStreamListener> mProvider;
+  SpeechRecognition::EventType mType;
+  TrackRate mTrackRate;
+};
+
+} // namespace dom
+
+inline nsISupports*
+ToSupports(dom::SpeechRecognition* aRec)
+{
+  return ToSupports(static_cast<DOMEventTargetHelper*>(aRec));
+}
+
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/SpeechRecognitionAlternative.cpp

@@ -0,0 +1,59 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechRecognitionAlternative.h"
+
+#include "mozilla/dom/SpeechRecognitionAlternativeBinding.h"
+
+#include "SpeechRecognition.h"
+
+namespace mozilla {
+namespace dom {
+
+NS_IMPL_CYCLE_COLLECTION_WRAPPERCACHE(SpeechRecognitionAlternative, mParent)
+NS_IMPL_CYCLE_COLLECTING_ADDREF(SpeechRecognitionAlternative)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(SpeechRecognitionAlternative)
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(SpeechRecognitionAlternative)
+  NS_WRAPPERCACHE_INTERFACE_MAP_ENTRY
+  NS_INTERFACE_MAP_ENTRY(nsISupports)
+NS_INTERFACE_MAP_END
+
+SpeechRecognitionAlternative::SpeechRecognitionAlternative(SpeechRecognition* aParent)
+  : mConfidence(0)
+  , mParent(aParent)
+{
+}
+
+SpeechRecognitionAlternative::~SpeechRecognitionAlternative()
+{
+}
+
+JSObject*
+SpeechRecognitionAlternative::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto)
+{
+  return SpeechRecognitionAlternativeBinding::Wrap(aCx, this, aGivenProto);
+}
+
+nsISupports*
+SpeechRecognitionAlternative::GetParentObject() const
+{
+  return static_cast<DOMEventTargetHelper*>(mParent.get());
+}
+
+void
+SpeechRecognitionAlternative::GetTranscript(nsString& aRetVal) const
+{
+  aRetVal = mTranscript;
+}
+
+float
+SpeechRecognitionAlternative::Confidence() const
+{
+  return mConfidence;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/recognition/SpeechRecognitionAlternative.h

@@ -0,0 +1,50 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechRecognitionAlternative_h
+#define mozilla_dom_SpeechRecognitionAlternative_h
+
+#include "nsCycleCollectionParticipant.h"
+#include "nsString.h"
+#include "nsWrapperCache.h"
+#include "js/TypeDecls.h"
+
+#include "mozilla/Attributes.h"
+
+namespace mozilla {
+namespace dom {
+
+class SpeechRecognition;
+
+class SpeechRecognitionAlternative final : public nsISupports,
+                                           public nsWrapperCache
+{
+public:
+  explicit SpeechRecognitionAlternative(SpeechRecognition* aParent);
+
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_SCRIPT_HOLDER_CLASS(SpeechRecognitionAlternative)
+
+  nsISupports* GetParentObject() const;
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  void GetTranscript(nsString& aRetVal) const;
+
+  float Confidence() const;
+
+  nsString mTranscript;
+  float mConfidence;
+private:
+  ~SpeechRecognitionAlternative();
+
+  RefPtr<SpeechRecognition> mParent;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/SpeechRecognitionResult.cpp

@@ -0,0 +1,76 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechRecognitionResult.h"
+#include "mozilla/dom/SpeechRecognitionResultBinding.h"
+
+#include "SpeechRecognition.h"
+
+namespace mozilla {
+namespace dom {
+
+NS_IMPL_CYCLE_COLLECTION_WRAPPERCACHE(SpeechRecognitionResult, mParent)
+NS_IMPL_CYCLE_COLLECTING_ADDREF(SpeechRecognitionResult)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(SpeechRecognitionResult)
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(SpeechRecognitionResult)
+  NS_WRAPPERCACHE_INTERFACE_MAP_ENTRY
+  NS_INTERFACE_MAP_ENTRY(nsISupports)
+NS_INTERFACE_MAP_END
+
+SpeechRecognitionResult::SpeechRecognitionResult(SpeechRecognition* aParent)
+  : mParent(aParent)
+{
+}
+
+SpeechRecognitionResult::~SpeechRecognitionResult()
+{
+}
+
+JSObject*
+SpeechRecognitionResult::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto)
+{
+  return SpeechRecognitionResultBinding::Wrap(aCx, this, aGivenProto);
+}
+
+nsISupports*
+SpeechRecognitionResult::GetParentObject() const
+{
+  return static_cast<DOMEventTargetHelper*>(mParent.get());
+}
+
+already_AddRefed<SpeechRecognitionAlternative>
+SpeechRecognitionResult::IndexedGetter(uint32_t aIndex, bool& aPresent)
+{
+  if (aIndex >= Length()) {
+    aPresent = false;
+    return nullptr;
+  }
+
+  aPresent = true;
+  return Item(aIndex);
+}
+
+uint32_t
+SpeechRecognitionResult::Length() const
+{
+  return mItems.Length();
+}
+
+already_AddRefed<SpeechRecognitionAlternative>
+SpeechRecognitionResult::Item(uint32_t aIndex)
+{
+  RefPtr<SpeechRecognitionAlternative> alternative = mItems.ElementAt(aIndex);
+  return alternative.forget();
+}
+
+bool
+SpeechRecognitionResult::IsFinal() const
+{
+  return true; // TODO
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/recognition/SpeechRecognitionResult.h

@@ -0,0 +1,55 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechRecognitionResult_h
+#define mozilla_dom_SpeechRecognitionResult_h
+
+#include "nsCOMPtr.h"
+#include "nsCycleCollectionParticipant.h"
+#include "nsWrapperCache.h"
+#include "nsTArray.h"
+#include "js/TypeDecls.h"
+
+#include "mozilla/Attributes.h"
+
+#include "SpeechRecognitionAlternative.h"
+
+namespace mozilla {
+namespace dom {
+
+class SpeechRecognitionResult final : public nsISupports,
+                                      public nsWrapperCache
+{
+public:
+  explicit SpeechRecognitionResult(SpeechRecognition* aParent);
+
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_SCRIPT_HOLDER_CLASS(SpeechRecognitionResult)
+
+  nsISupports* GetParentObject() const;
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  uint32_t Length() const;
+
+  already_AddRefed<SpeechRecognitionAlternative> Item(uint32_t aIndex);
+
+  bool IsFinal() const;
+
+  already_AddRefed<SpeechRecognitionAlternative> IndexedGetter(uint32_t aIndex, bool& aPresent);
+
+  nsTArray<RefPtr<SpeechRecognitionAlternative>> mItems;
+
+private:
+  ~SpeechRecognitionResult();
+
+  RefPtr<SpeechRecognition> mParent;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/SpeechRecognitionResultList.cpp

@@ -0,0 +1,71 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechRecognitionResultList.h"
+
+#include "mozilla/dom/SpeechRecognitionResultListBinding.h"
+
+#include "SpeechRecognition.h"
+
+namespace mozilla {
+namespace dom {
+
+NS_IMPL_CYCLE_COLLECTION_WRAPPERCACHE(SpeechRecognitionResultList, mParent, mItems)
+NS_IMPL_CYCLE_COLLECTING_ADDREF(SpeechRecognitionResultList)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(SpeechRecognitionResultList)
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(SpeechRecognitionResultList)
+  NS_WRAPPERCACHE_INTERFACE_MAP_ENTRY
+  NS_INTERFACE_MAP_ENTRY(nsISupports)
+NS_INTERFACE_MAP_END
+
+SpeechRecognitionResultList::SpeechRecognitionResultList(SpeechRecognition* aParent)
+  : mParent(aParent)
+{
+}
+
+SpeechRecognitionResultList::~SpeechRecognitionResultList()
+{
+}
+
+nsISupports*
+SpeechRecognitionResultList::GetParentObject() const
+{
+  return static_cast<DOMEventTargetHelper*>(mParent.get());
+}
+
+JSObject*
+SpeechRecognitionResultList::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto)
+{
+  return SpeechRecognitionResultListBinding::Wrap(aCx, this, aGivenProto);
+}
+
+already_AddRefed<SpeechRecognitionResult>
+SpeechRecognitionResultList::IndexedGetter(uint32_t aIndex, bool& aPresent)
+{
+  if (aIndex >= Length()) {
+    aPresent = false;
+    return nullptr;
+  }
+
+  aPresent = true;
+  return Item(aIndex);
+}
+
+uint32_t
+SpeechRecognitionResultList::Length() const
+{
+  return mItems.Length();
+}
+
+already_AddRefed<SpeechRecognitionResult>
+SpeechRecognitionResultList::Item(uint32_t aIndex)
+{
+  RefPtr<SpeechRecognitionResult> result = mItems.ElementAt(aIndex);
+  return result.forget();
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/recognition/SpeechRecognitionResultList.h

@@ -0,0 +1,53 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechRecognitionResultList_h
+#define mozilla_dom_SpeechRecognitionResultList_h
+
+#include "nsCycleCollectionParticipant.h"
+#include "nsWrapperCache.h"
+#include "nsTArray.h"
+#include "js/TypeDecls.h"
+
+#include "mozilla/Attributes.h"
+
+#include "SpeechRecognitionResult.h"
+
+namespace mozilla {
+namespace dom {
+
+class SpeechRecognition;
+
+class SpeechRecognitionResultList final : public nsISupports,
+                                          public nsWrapperCache
+{
+public:
+  explicit SpeechRecognitionResultList(SpeechRecognition* aParent);
+
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_SCRIPT_HOLDER_CLASS(SpeechRecognitionResultList)
+
+  nsISupports* GetParentObject() const;
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  uint32_t Length() const;
+
+  already_AddRefed<SpeechRecognitionResult> Item(uint32_t aIndex);
+
+  already_AddRefed<SpeechRecognitionResult> IndexedGetter(uint32_t aIndex, bool& aPresent);
+
+  nsTArray<RefPtr<SpeechRecognitionResult>> mItems;
+private:
+  ~SpeechRecognitionResultList();
+
+  RefPtr<SpeechRecognition> mParent;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/SpeechStreamListener.cpp

@@ -0,0 +1,94 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechStreamListener.h"
+
+#include "SpeechRecognition.h"
+#include "nsProxyRelease.h"
+
+namespace mozilla {
+namespace dom {
+
+SpeechStreamListener::SpeechStreamListener(SpeechRecognition* aRecognition)
+  : mRecognition(aRecognition)
+{
+}
+
+SpeechStreamListener::~SpeechStreamListener()
+{
+  nsCOMPtr<nsIThread> mainThread;
+  NS_GetMainThread(getter_AddRefs(mainThread));
+
+  NS_ProxyRelease(mainThread, mRecognition.forget());
+}
+
+void
+SpeechStreamListener::NotifyQueuedAudioData(MediaStreamGraph* aGraph, TrackID aID,
+                                            StreamTime aTrackOffset,
+                                            const AudioSegment& aQueuedMedia,
+                                            MediaStream* aInputStream,
+                                            TrackID aInputTrackID)
+{
+  AudioSegment* audio = const_cast<AudioSegment*>(
+    static_cast<const AudioSegment*>(&aQueuedMedia));
+
+  AudioSegment::ChunkIterator iterator(*audio);
+  while (!iterator.IsEnded()) {
+    // Skip over-large chunks so we don't crash!
+    if (iterator->GetDuration() > INT_MAX) {
+      continue;
+    }
+    int duration = int(iterator->GetDuration());
+
+    if (iterator->IsNull()) {
+      nsTArray<int16_t> nullData;
+      PodZero(nullData.AppendElements(duration), duration);
+      ConvertAndDispatchAudioChunk(duration, iterator->mVolume,
+                                   nullData.Elements(), aGraph->GraphRate());
+    } else {
+      AudioSampleFormat format = iterator->mBufferFormat;
+
+      MOZ_ASSERT(format == AUDIO_FORMAT_S16 || format == AUDIO_FORMAT_FLOAT32);
+
+      if (format == AUDIO_FORMAT_S16) {
+        ConvertAndDispatchAudioChunk(duration,iterator->mVolume,
+                                     static_cast<const int16_t*>(iterator->mChannelData[0]),
+                                     aGraph->GraphRate());
+      } else if (format == AUDIO_FORMAT_FLOAT32) {
+        ConvertAndDispatchAudioChunk(duration,iterator->mVolume,
+                                     static_cast<const float*>(iterator->mChannelData[0]),
+                                     aGraph->GraphRate());
+      }
+    }
+
+    iterator.Next();
+  }
+}
+
+template<typename SampleFormatType> void
+SpeechStreamListener::ConvertAndDispatchAudioChunk(int aDuration, float aVolume,
+                                                   SampleFormatType* aData,
+                                                   TrackRate aTrackRate)
+{
+  RefPtr<SharedBuffer> samples(SharedBuffer::Create(aDuration *
+                                                      1 * // channel
+                                                      sizeof(int16_t)));
+
+  int16_t* to = static_cast<int16_t*>(samples->Data());
+  ConvertAudioSamplesWithScale(aData, to, aDuration, aVolume);
+
+  mRecognition->FeedAudioData(samples.forget(), aDuration, this, aTrackRate);
+}
+
+void
+SpeechStreamListener::NotifyEvent(MediaStreamGraph* aGraph,
+                                  MediaStreamGraphEvent event)
+{
+  // TODO dispatch SpeechEnd event so services can be informed
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/recognition/SpeechStreamListener.h

@@ -0,0 +1,46 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechStreamListener_h
+#define mozilla_dom_SpeechStreamListener_h
+
+#include "MediaStreamGraph.h"
+#include "MediaStreamListener.h"
+#include "AudioSegment.h"
+
+namespace mozilla {
+
+class AudioSegment;
+
+namespace dom {
+
+class SpeechRecognition;
+
+class SpeechStreamListener : public MediaStreamListener
+{
+public:
+  explicit SpeechStreamListener(SpeechRecognition* aRecognition);
+  ~SpeechStreamListener();
+
+  void NotifyQueuedAudioData(MediaStreamGraph* aGraph, TrackID aID,
+                             StreamTime aTrackOffset,
+                             const AudioSegment& aQueuedMedia,
+                             MediaStream* aInputStream,
+                             TrackID aInputTrackID) override;
+
+  void NotifyEvent(MediaStreamGraph* aGraph,
+                   MediaStreamGraphEvent event) override;
+
+private:
+  template<typename SampleFormatType>
+  void ConvertAndDispatchAudioChunk(int aDuration, float aVolume, SampleFormatType* aData, TrackRate aTrackRate);
+  RefPtr<SpeechRecognition> mRecognition;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/endpointer.cc

@@ -0,0 +1,193 @@
+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//    * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//    * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//    * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include "endpointer.h"
+
+#include "AudioSegment.h"
+
+namespace {
+const int kFrameRate = 200;  // 1 frame = 5ms of audio.
+}
+
+namespace mozilla {
+
+Endpointer::Endpointer(int sample_rate)
+    : speech_input_possibly_complete_silence_length_us_(-1),
+      speech_input_complete_silence_length_us_(-1),
+      audio_frame_time_us_(0),
+      sample_rate_(sample_rate),
+      frame_size_(0) {
+  Reset();
+
+  frame_size_ = static_cast<int>(sample_rate / static_cast<float>(kFrameRate));
+
+  speech_input_minimum_length_us_ =
+      static_cast<int64_t>(1.7 * 1000000);
+  speech_input_complete_silence_length_us_ =
+      static_cast<int64_t>(0.5 * 1000000);
+  long_speech_input_complete_silence_length_us_ = -1;
+  long_speech_length_us_ = -1;
+  speech_input_possibly_complete_silence_length_us_ =
+      1 * 1000000;
+
+  // Set the default configuration for Push To Talk mode.
+  EnergyEndpointerParams ep_config;
+  ep_config.set_frame_period(1.0f / static_cast<float>(kFrameRate));
+  ep_config.set_frame_duration(1.0f / static_cast<float>(kFrameRate));
+  ep_config.set_endpoint_margin(0.2f);
+  ep_config.set_onset_window(0.15f);
+  ep_config.set_speech_on_window(0.4f);
+  ep_config.set_offset_window(0.15f);
+  ep_config.set_onset_detect_dur(0.09f);
+  ep_config.set_onset_confirm_dur(0.075f);
+  ep_config.set_on_maintain_dur(0.10f);
+  ep_config.set_offset_confirm_dur(0.12f);
+  ep_config.set_decision_threshold(1000.0f);
+  ep_config.set_min_decision_threshold(50.0f);
+  ep_config.set_fast_update_dur(0.2f);
+  ep_config.set_sample_rate(static_cast<float>(sample_rate));
+  ep_config.set_min_fundamental_frequency(57.143f);
+  ep_config.set_max_fundamental_frequency(400.0f);
+  ep_config.set_contamination_rejection_period(0.25f);
+  energy_endpointer_.Init(ep_config);
+}
+
+void Endpointer::Reset() {
+  old_ep_status_ = EP_PRE_SPEECH;
+  waiting_for_speech_possibly_complete_timeout_ = false;
+  waiting_for_speech_complete_timeout_ = false;
+  speech_previously_detected_ = false;
+  speech_input_complete_ = false;
+  audio_frame_time_us_ = 0; // Reset time for packets sent to endpointer.
+  speech_end_time_us_ = -1;
+  speech_start_time_us_ = -1;
+}
+
+void Endpointer::StartSession() {
+  Reset();
+  energy_endpointer_.StartSession();
+}
+
+void Endpointer::EndSession() {
+  energy_endpointer_.EndSession();
+}
+
+void Endpointer::SetEnvironmentEstimationMode() {
+  Reset();
+  energy_endpointer_.SetEnvironmentEstimationMode();
+}
+
+void Endpointer::SetUserInputMode() {
+  energy_endpointer_.SetUserInputMode();
+}
+
+EpStatus Endpointer::Status(int64_t *time) {
+  return energy_endpointer_.Status(time);
+}
+
+EpStatus Endpointer::ProcessAudio(const AudioChunk& raw_audio, float* rms_out) {
+  MOZ_ASSERT(raw_audio.mBufferFormat == AUDIO_FORMAT_S16, "Audio is not in 16 bit format");
+  const int16_t* audio_data = static_cast<const int16_t*>(raw_audio.mChannelData[0]);
+  const int num_samples = raw_audio.mDuration;
+  EpStatus ep_status = EP_PRE_SPEECH;
+
+  // Process the input data in blocks of frame_size_, dropping any incomplete
+  // frames at the end (which is ok since typically the caller will be recording
+  // audio in multiples of our frame size).
+  int sample_index = 0;
+  while (sample_index + frame_size_ <= num_samples) {
+    // Have the endpointer process the frame.
+    energy_endpointer_.ProcessAudioFrame(audio_frame_time_us_,
+                                         audio_data + sample_index,
+                                         frame_size_,
+                                         rms_out);
+    sample_index += frame_size_;
+    audio_frame_time_us_ += (frame_size_ * 1000000) /
+                         sample_rate_;
+
+    // Get the status of the endpointer.
+    int64_t ep_time;
+    ep_status = energy_endpointer_.Status(&ep_time);
+    if (old_ep_status_ != ep_status)
+        fprintf(stderr, "Status changed old= %d, new= %d\n", old_ep_status_, ep_status);
+
+    // Handle state changes.
+    if ((EP_SPEECH_PRESENT == ep_status) &&
+        (EP_POSSIBLE_ONSET == old_ep_status_)) {
+      speech_end_time_us_ = -1;
+      waiting_for_speech_possibly_complete_timeout_ = false;
+      waiting_for_speech_complete_timeout_ = false;
+      // Trigger SpeechInputDidStart event on first detection.
+      if (false == speech_previously_detected_) {
+        speech_previously_detected_ = true;
+        speech_start_time_us_ = ep_time;
+      }
+    }
+    if ((EP_PRE_SPEECH == ep_status) &&
+        (EP_POSSIBLE_OFFSET == old_ep_status_)) {
+      speech_end_time_us_ = ep_time;
+      waiting_for_speech_possibly_complete_timeout_ = true;
+      waiting_for_speech_complete_timeout_ = true;
+    }
+    if (ep_time > speech_input_minimum_length_us_) {
+      // Speech possibly complete timeout.
+      if ((waiting_for_speech_possibly_complete_timeout_) &&
+          (ep_time - speech_end_time_us_ >
+              speech_input_possibly_complete_silence_length_us_)) {
+        waiting_for_speech_possibly_complete_timeout_ = false;
+      }
+      if (waiting_for_speech_complete_timeout_) {
+        // The length of the silence timeout period can be held constant, or it
+        // can be changed after a fixed amount of time from the beginning of
+        // speech.
+        bool has_stepped_silence =
+            (long_speech_length_us_ > 0) &&
+            (long_speech_input_complete_silence_length_us_ > 0);
+        int64_t requested_silence_length;
+        if (has_stepped_silence &&
+            (ep_time - speech_start_time_us_) > long_speech_length_us_) {
+          requested_silence_length =
+              long_speech_input_complete_silence_length_us_;
+        } else {
+          requested_silence_length =
+              speech_input_complete_silence_length_us_;
+        }
+
+        // Speech complete timeout.
+        if ((ep_time - speech_end_time_us_) > requested_silence_length) {
+          waiting_for_speech_complete_timeout_ = false;
+          speech_input_complete_ = true;
+        }
+      }
+    }
+    old_ep_status_ = ep_status;
+  }
+  return ep_status;
+}
+
+}  // namespace mozilla

dom/media/webspeech/recognition/endpointer.h

@@ -0,0 +1,180 @@
+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//    * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//    * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//    * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#ifndef CONTENT_BROWSER_SPEECH_ENDPOINTER_ENDPOINTER_H_
+#define CONTENT_BROWSER_SPEECH_ENDPOINTER_ENDPOINTER_H_
+
+#include "energy_endpointer.h"
+
+namespace mozilla {
+
+struct AudioChunk;
+
+// A simple interface to the underlying energy-endpointer implementation, this
+// class lets callers provide audio as being recorded and let them poll to find
+// when the user has stopped speaking.
+//
+// There are two events that may trigger the end of speech:
+//
+// speechInputPossiblyComplete event:
+//
+// Signals that silence/noise has  been detected for a *short* amount of
+// time after some speech has been detected. It can be used for low latency
+// UI feedback. To disable it, set it to a large amount.
+//
+// speechInputComplete event:
+//
+// This event is intended to signal end of input and to stop recording.
+// The amount of time to wait after speech is set by
+// speech_input_complete_silence_length_ and optionally two other
+// parameters (see below).
+// This time can be held constant, or can change as more speech is detected.
+// In the latter case, the time changes after a set amount of time from the
+// *beginning* of speech.  This is motivated by the expectation that there
+// will be two distinct types of inputs: short search queries and longer
+// dictation style input.
+//
+// Three parameters are used to define the piecewise constant timeout function.
+// The timeout length is speech_input_complete_silence_length until
+// long_speech_length, when it changes to
+// long_speech_input_complete_silence_length.
+class Endpointer {
+ public:
+  explicit Endpointer(int sample_rate);
+
+  // Start the endpointer. This should be called at the beginning of a session.
+  void StartSession();
+
+  // Stop the endpointer.
+  void EndSession();
+
+  // Start environment estimation. Audio will be used for environment estimation
+  // i.e. noise level estimation.
+  void SetEnvironmentEstimationMode();
+
+  // Start user input. This should be called when the user indicates start of
+  // input, e.g. by pressing a button.
+  void SetUserInputMode();
+
+  // Process a segment of audio, which may be more than one frame.
+  // The status of the last frame will be returned.
+  EpStatus ProcessAudio(const AudioChunk& raw_audio, float* rms_out);
+
+  // Get the status of the endpointer.
+  EpStatus Status(int64_t *time_us);
+
+  // Get the expected frame size for audio chunks. Audio chunks are expected
+  // to contain a number of samples that is a multiple of this number, and extra
+  // samples will be dropped.
+  int32_t FrameSize() const {
+    return frame_size_;
+  }
+
+  // Returns true if the endpointer detected reasonable audio levels above
+  // background noise which could be user speech, false if not.
+  bool DidStartReceivingSpeech() const {
+    return speech_previously_detected_;
+  }
+
+  bool IsEstimatingEnvironment() const {
+    return energy_endpointer_.estimating_environment();
+  }
+
+  void set_speech_input_complete_silence_length(int64_t time_us) {
+    speech_input_complete_silence_length_us_ = time_us;
+  }
+
+  void set_long_speech_input_complete_silence_length(int64_t time_us) {
+    long_speech_input_complete_silence_length_us_ = time_us;
+  }
+
+  void set_speech_input_possibly_complete_silence_length(int64_t time_us) {
+    speech_input_possibly_complete_silence_length_us_ = time_us;
+  }
+
+  void set_long_speech_length(int64_t time_us) {
+    long_speech_length_us_ = time_us;
+  }
+
+  bool speech_input_complete() const {
+    return speech_input_complete_;
+  }
+
+  // RMS background noise level in dB.
+  float NoiseLevelDb() const { return energy_endpointer_.GetNoiseLevelDb(); }
+
+ private:
+  // Reset internal states. Helper method common to initial input utterance
+  // and following input utternaces.
+  void Reset();
+
+  // Minimum allowable length of speech input.
+  int64_t speech_input_minimum_length_us_;
+
+  // The speechInputPossiblyComplete event signals that silence/noise has been
+  // detected for a *short* amount of time after some speech has been detected.
+  // This proporty specifies the time period.
+  int64_t speech_input_possibly_complete_silence_length_us_;
+
+  // The speechInputComplete event signals that silence/noise has been
+  // detected for a *long* amount of time after some speech has been detected.
+  // This property specifies the time period.
+  int64_t speech_input_complete_silence_length_us_;
+
+  // Same as above, this specifies the required silence period after speech
+  // detection. This period is used instead of
+  // speech_input_complete_silence_length_ when the utterance is longer than
+  // long_speech_length_. This parameter is optional.
+  int64_t long_speech_input_complete_silence_length_us_;
+
+  // The period of time after which the endpointer should consider
+  // long_speech_input_complete_silence_length_ as a valid silence period
+  // instead of speech_input_complete_silence_length_. This parameter is
+  // optional.
+  int64_t long_speech_length_us_;
+
+  // First speech onset time, used in determination of speech complete timeout.
+  int64_t speech_start_time_us_;
+
+  // Most recent end time, used in determination of speech complete timeout.
+  int64_t speech_end_time_us_;
+
+  int64_t audio_frame_time_us_;
+  EpStatus old_ep_status_;
+  bool waiting_for_speech_possibly_complete_timeout_;
+  bool waiting_for_speech_complete_timeout_;
+  bool speech_previously_detected_;
+  bool speech_input_complete_;
+  EnergyEndpointer energy_endpointer_;
+  int sample_rate_;
+  int32_t frame_size_;
+};
+
+}  // namespace mozilla
+
+#endif  // CONTENT_BROWSER_SPEECH_ENDPOINTER_ENDPOINTER_H_

dom/media/webspeech/recognition/energy_endpointer.cc

@@ -0,0 +1,393 @@
+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//    * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//    * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//    * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include "energy_endpointer.h"
+
+#include <math.h>
+
+namespace {
+
+// Returns the RMS (quadratic mean) of the input signal.
+float RMS(const int16_t* samples, int num_samples) {
+  int64_t ssq_int64_t = 0;
+  int64_t sum_int64_t = 0;
+  for (int i = 0; i < num_samples; ++i) {
+    sum_int64_t += samples[i];
+    ssq_int64_t += samples[i] * samples[i];
+  }
+  // now convert to floats.
+  double sum = static_cast<double>(sum_int64_t);
+  sum /= num_samples;
+  double ssq = static_cast<double>(ssq_int64_t);
+  return static_cast<float>(sqrt((ssq / num_samples) - (sum * sum)));
+}
+
+int64_t Secs2Usecs(float seconds) {
+  return static_cast<int64_t>(0.5 + (1.0e6 * seconds));
+}
+
+float GetDecibel(float value) {
+  if (value > 1.0e-100)
+    return 20 * log10(value);
+  return -2000.0;
+}
+
+}  // namespace
+
+namespace mozilla {
+
+// Stores threshold-crossing histories for making decisions about the speech
+// state.
+class EnergyEndpointer::HistoryRing {
+ public:
+  HistoryRing() : insertion_index_(0) {}
+
+  // Resets the ring to |size| elements each with state |initial_state|
+  void SetRing(int size, bool initial_state);
+
+  // Inserts a new entry into the ring and drops the oldest entry.
+  void Insert(int64_t time_us, bool decision);
+
+  // Returns the time in microseconds of the most recently added entry.
+  int64_t EndTime() const;
+
+  // Returns the sum of all intervals during which 'decision' is true within
+  // the time in seconds specified by 'duration'. The returned interval is
+  // in seconds.
+  float RingSum(float duration_sec);
+
+ private:
+  struct DecisionPoint {
+    int64_t time_us;
+    bool decision;
+  };
+
+  std::vector<DecisionPoint> decision_points_;
+  int insertion_index_;  // Index at which the next item gets added/inserted.
+
+  HistoryRing(const HistoryRing&);
+  void operator=(const HistoryRing&);
+};
+
+void EnergyEndpointer::HistoryRing::SetRing(int size, bool initial_state) {
+  insertion_index_ = 0;
+  decision_points_.clear();
+  DecisionPoint init = { -1, initial_state };
+  decision_points_.resize(size, init);
+}
+
+void EnergyEndpointer::HistoryRing::Insert(int64_t time_us, bool decision) {
+  decision_points_[insertion_index_].time_us = time_us;
+  decision_points_[insertion_index_].decision = decision;
+  insertion_index_ = (insertion_index_ + 1) % decision_points_.size();
+}
+
+int64_t EnergyEndpointer::HistoryRing::EndTime() const {
+  int ind = insertion_index_ - 1;
+  if (ind < 0)
+    ind = decision_points_.size() - 1;
+  return decision_points_[ind].time_us;
+}
+
+float EnergyEndpointer::HistoryRing::RingSum(float duration_sec) {
+  if (!decision_points_.size())
+    return 0.0;
+
+  int64_t sum_us = 0;
+  int ind = insertion_index_ - 1;
+  if (ind < 0)
+    ind = decision_points_.size() - 1;
+  int64_t end_us = decision_points_[ind].time_us;
+  bool is_on = decision_points_[ind].decision;
+  int64_t start_us = end_us - static_cast<int64_t>(0.5 + (1.0e6 * duration_sec));
+  if (start_us < 0)
+    start_us = 0;
+  size_t n_summed = 1;  // n points ==> (n-1) intervals
+  while ((decision_points_[ind].time_us > start_us) &&
+         (n_summed < decision_points_.size())) {
+    --ind;
+    if (ind < 0)
+      ind = decision_points_.size() - 1;
+    if (is_on)
+      sum_us += end_us - decision_points_[ind].time_us;
+    is_on = decision_points_[ind].decision;
+    end_us = decision_points_[ind].time_us;
+    n_summed++;
+  }
+
+  return 1.0e-6f * sum_us;  //  Returns total time that was super threshold.
+}
+
+EnergyEndpointer::EnergyEndpointer()
+    : status_(EP_PRE_SPEECH),
+      offset_confirm_dur_sec_(0),
+      endpointer_time_us_(0),
+      fast_update_frames_(0),
+      frame_counter_(0),
+      max_window_dur_(4.0),
+      sample_rate_(0),
+      history_(new HistoryRing()),
+      decision_threshold_(0),
+      estimating_environment_(false),
+      noise_level_(0),
+      rms_adapt_(0),
+      start_lag_(0),
+      end_lag_(0),
+      user_input_start_time_us_(0) {
+}
+
+EnergyEndpointer::~EnergyEndpointer() {
+}
+
+int EnergyEndpointer::TimeToFrame(float time) const {
+  return static_cast<int32_t>(0.5 + (time / params_.frame_period()));
+}
+
+void EnergyEndpointer::Restart(bool reset_threshold) {
+  status_ = EP_PRE_SPEECH;
+  user_input_start_time_us_ = 0;
+
+  if (reset_threshold) {
+    decision_threshold_ = params_.decision_threshold();
+    rms_adapt_ = decision_threshold_;
+    noise_level_ = params_.decision_threshold() / 2.0f;
+    frame_counter_ = 0;  // Used for rapid initial update of levels.
+  }
+
+  // Set up the memories to hold the history windows.
+  history_->SetRing(TimeToFrame(max_window_dur_), false);
+
+  // Flag that indicates that current input should be used for
+  // estimating the environment. The user has not yet started input
+  // by e.g. pressed the push-to-talk button. By default, this is
+  // false for backward compatibility.
+  estimating_environment_ = false;
+}
+
+void EnergyEndpointer::Init(const EnergyEndpointerParams& params) {
+  params_ = params;
+
+  // Find the longest history interval to be used, and make the ring
+  // large enough to accommodate that number of frames.  NOTE: This
+  // depends upon ep_frame_period being set correctly in the factory
+  // that did this instantiation.
+  max_window_dur_ = params_.onset_window();
+  if (params_.speech_on_window() > max_window_dur_)
+    max_window_dur_ = params_.speech_on_window();
+  if (params_.offset_window() > max_window_dur_)
+    max_window_dur_ = params_.offset_window();
+  Restart(true);
+
+  offset_confirm_dur_sec_ = params_.offset_window() -
+                            params_.offset_confirm_dur();
+  if (offset_confirm_dur_sec_ < 0.0)
+    offset_confirm_dur_sec_ = 0.0;
+
+  user_input_start_time_us_ = 0;
+
+  // Flag that indicates that  current input should be used for
+  // estimating the environment. The user has not yet started input
+  // by e.g. pressed the push-to-talk button. By default, this is
+  // false for backward compatibility.
+  estimating_environment_ = false;
+  // The initial value of the noise and speech levels is inconsequential.
+  // The level of the first frame will overwrite these values.
+  noise_level_ = params_.decision_threshold() / 2.0f;
+  fast_update_frames_ =
+      static_cast<int64_t>(params_.fast_update_dur() / params_.frame_period());
+
+  frame_counter_ = 0;  // Used for rapid initial update of levels.
+
+  sample_rate_ = params_.sample_rate();
+  start_lag_ = static_cast<int>(sample_rate_ /
+                                params_.max_fundamental_frequency());
+  end_lag_ = static_cast<int>(sample_rate_ /
+                              params_.min_fundamental_frequency());
+}
+
+void EnergyEndpointer::StartSession() {
+  Restart(true);
+}
+
+void EnergyEndpointer::EndSession() {
+  status_ = EP_POST_SPEECH;
+}
+
+void EnergyEndpointer::SetEnvironmentEstimationMode() {
+  Restart(true);
+  estimating_environment_ = true;
+}
+
+void EnergyEndpointer::SetUserInputMode() {
+  estimating_environment_ = false;
+  user_input_start_time_us_ = endpointer_time_us_;
+}
+
+void EnergyEndpointer::ProcessAudioFrame(int64_t time_us,
+                                         const int16_t* samples,
+                                         int num_samples,
+                                         float* rms_out) {
+  endpointer_time_us_ = time_us;
+  float rms = RMS(samples, num_samples);
+
+  // Check that this is user input audio vs. pre-input adaptation audio.
+  // Input audio starts when the user indicates start of input, by e.g.
+  // pressing push-to-talk. Audio recieved prior to that is used to update
+  // noise and speech level estimates.
+  if (!estimating_environment_) {
+    bool decision = false;
+    if ((endpointer_time_us_ - user_input_start_time_us_) <
+        Secs2Usecs(params_.contamination_rejection_period())) {
+      decision = false;
+      //PR_LOG(GetSpeechRecognitionLog(), PR_LOG_DEBUG, ("decision: forced to false, time: %d", endpointer_time_us_));
+    } else {
+      decision = (rms > decision_threshold_);
+    }
+
+    history_->Insert(endpointer_time_us_, decision);
+
+    switch (status_) {
+      case EP_PRE_SPEECH:
+        if (history_->RingSum(params_.onset_window()) >
+            params_.onset_detect_dur()) {
+          status_ = EP_POSSIBLE_ONSET;
+        }
+        break;
+
+      case EP_POSSIBLE_ONSET: {
+        float tsum = history_->RingSum(params_.onset_window());
+        if (tsum > params_.onset_confirm_dur()) {
+          status_ = EP_SPEECH_PRESENT;
+        } else {  // If signal is not maintained, drop back to pre-speech.
+          if (tsum <= params_.onset_detect_dur())
+            status_ = EP_PRE_SPEECH;
+        }
+        break;
+      }
+
+      case EP_SPEECH_PRESENT: {
+        // To induce hysteresis in the state residency, we allow a
+        // smaller residency time in the on_ring, than was required to
+        // enter the SPEECH_PERSENT state.
+        float on_time = history_->RingSum(params_.speech_on_window());
+        if (on_time < params_.on_maintain_dur())
+          status_ = EP_POSSIBLE_OFFSET;
+        break;
+      }
+
+      case EP_POSSIBLE_OFFSET:
+        if (history_->RingSum(params_.offset_window()) <=
+            offset_confirm_dur_sec_) {
+          // Note that this offset time may be beyond the end
+          // of the input buffer in a real-time system.  It will be up
+          // to the RecognizerSession to decide what to do.
+          status_ = EP_PRE_SPEECH;  // Automatically reset for next utterance.
+        } else {  // If speech picks up again we allow return to SPEECH_PRESENT.
+          if (history_->RingSum(params_.speech_on_window()) >=
+              params_.on_maintain_dur())
+            status_ = EP_SPEECH_PRESENT;
+        }
+        break;
+
+      default:
+        break;
+    }
+
+    // If this is a quiet, non-speech region, slowly adapt the detection
+    // threshold to be about 6dB above the average RMS.
+    if ((!decision) && (status_ == EP_PRE_SPEECH)) {
+      decision_threshold_ = (0.98f * decision_threshold_) + (0.02f * 2 * rms);
+      rms_adapt_ = decision_threshold_;
+    } else {
+      // If this is in a speech region, adapt the decision threshold to
+      // be about 10dB below the average RMS. If the noise level is high,
+      // the threshold is pushed up.
+      // Adaptation up to a higher level is 5 times faster than decay to
+      // a lower level.
+      if ((status_ == EP_SPEECH_PRESENT) && decision) {
+        if (rms_adapt_ > rms) {
+          rms_adapt_ = (0.99f * rms_adapt_) + (0.01f * rms);
+        } else {
+          rms_adapt_ = (0.95f * rms_adapt_) + (0.05f * rms);
+        }
+        float target_threshold = 0.3f * rms_adapt_ +  noise_level_;
+        decision_threshold_ = (.90f * decision_threshold_) +
+                              (0.10f * target_threshold);
+      }
+    }
+
+    // Set a floor
+    if (decision_threshold_ < params_.min_decision_threshold())
+      decision_threshold_ = params_.min_decision_threshold();
+  }
+
+  // Update speech and noise levels.
+  UpdateLevels(rms);
+  ++frame_counter_;
+
+  if (rms_out)
+    *rms_out = GetDecibel(rms);
+}
+
+float EnergyEndpointer::GetNoiseLevelDb() const {
+  return GetDecibel(noise_level_);
+}
+
+void EnergyEndpointer::UpdateLevels(float rms) {
+  // Update quickly initially. We assume this is noise and that
+  // speech is 6dB above the noise.
+  if (frame_counter_ < fast_update_frames_) {
+    // Alpha increases from 0 to (k-1)/k where k is the number of time
+    // steps in the initial adaptation period.
+    float alpha = static_cast<float>(frame_counter_) /
+        static_cast<float>(fast_update_frames_);
+    noise_level_ = (alpha * noise_level_) + ((1 - alpha) * rms);
+    //PR_LOG(GetSpeechRecognitionLog(), PR_LOG_DEBUG, ("FAST UPDATE, frame_counter_ %d, fast_update_frames_ %d", frame_counter_, fast_update_frames_));
+  } else {
+    // Update Noise level. The noise level adapts quickly downward, but
+    // slowly upward. The noise_level_ parameter is not currently used
+    // for threshold adaptation. It is used for UI feedback.
+    if (noise_level_ < rms)
+      noise_level_ = (0.999f * noise_level_) + (0.001f * rms);
+    else
+      noise_level_ = (0.95f * noise_level_) + (0.05f * rms);
+  }
+  if (estimating_environment_ || (frame_counter_ < fast_update_frames_)) {
+    decision_threshold_ = noise_level_ * 2; // 6dB above noise level.
+    // Set a floor
+    if (decision_threshold_ < params_.min_decision_threshold())
+      decision_threshold_ = params_.min_decision_threshold();
+  }
+}
+
+EpStatus EnergyEndpointer::Status(int64_t* status_time)  const {
+  *status_time = history_->EndTime();
+  return status_;
+}
+
+}  // namespace mozilla

dom/media/webspeech/recognition/energy_endpointer.h

@@ -0,0 +1,180 @@
+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//    * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//    * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//    * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// The EnergyEndpointer class finds likely speech onset and offset points.
+//
+// The implementation described here is about the simplest possible.
+// It is based on timings of threshold crossings for overall signal
+// RMS. It is suitable for light weight applications.
+//
+// As written, the basic idea is that one specifies intervals that
+// must be occupied by super- and sub-threshold energy levels, and
+// defers decisions re onset and offset times until these
+// specifications have been met.  Three basic intervals are tested: an
+// onset window, a speech-on window, and an offset window.  We require
+// super-threshold to exceed some mimimum total durations in the onset
+// and speech-on windows before declaring the speech onset time, and
+// we specify a required sub-threshold residency in the offset window
+// before declaring speech offset. As the various residency requirements are
+// met, the EnergyEndpointer instance assumes various states, and can return the
+// ID of these states to the client (see EpStatus below).
+//
+// The levels of the speech and background noise are continuously updated. It is
+// important that the background noise level be estimated initially for
+// robustness in noisy conditions. The first frames are assumed to be background
+// noise and a fast update rate is used for the noise level. The duration for
+// fast update is controlled by the fast_update_dur_ paramter.
+//
+// If used in noisy conditions, the endpointer should be started and run in the
+// EnvironmentEstimation mode, for at least 200ms, before switching to
+// UserInputMode.
+// Audio feedback contamination can appear in the input audio, if not cut
+// out or handled by echo cancellation. Audio feedback can trigger a false
+// accept. The false accepts can be ignored by setting
+// ep_contamination_rejection_period.
+
+#ifndef CONTENT_BROWSER_SPEECH_ENDPOINTER_ENERGY_ENDPOINTER_H_
+#define CONTENT_BROWSER_SPEECH_ENDPOINTER_ENERGY_ENDPOINTER_H_
+
+#include <vector>
+
+#include "nsAutoPtr.h"
+
+#include "energy_endpointer_params.h"
+
+namespace mozilla {
+
+// Endpointer status codes
+enum EpStatus {
+  EP_PRE_SPEECH = 10,
+  EP_POSSIBLE_ONSET,
+  EP_SPEECH_PRESENT,
+  EP_POSSIBLE_OFFSET,
+  EP_POST_SPEECH,
+};
+
+class EnergyEndpointer {
+ public:
+  // The default construction MUST be followed by Init(), before any
+  // other use can be made of the instance.
+  EnergyEndpointer();
+  virtual ~EnergyEndpointer();
+
+  void Init(const EnergyEndpointerParams& params);
+
+  // Start the endpointer. This should be called at the beginning of a session.
+  void StartSession();
+
+  // Stop the endpointer.
+  void EndSession();
+
+  // Start environment estimation. Audio will be used for environment estimation
+  // i.e. noise level estimation.
+  void SetEnvironmentEstimationMode();
+
+  // Start user input. This should be called when the user indicates start of
+  // input, e.g. by pressing a button.
+  void SetUserInputMode();
+
+  // Computes the next input frame and modifies EnergyEndpointer status as
+  // appropriate based on the computation.
+  void ProcessAudioFrame(int64_t time_us,
+                         const int16_t* samples, int num_samples,
+                         float* rms_out);
+
+  // Returns the current state of the EnergyEndpointer and the time
+  // corresponding to the most recently computed frame.
+  EpStatus Status(int64_t* status_time_us) const;
+
+  bool estimating_environment() const {
+    return estimating_environment_;
+  }
+
+  // Returns estimated noise level in dB.
+  float GetNoiseLevelDb() const;
+
+ private:
+  class HistoryRing;
+
+  // Resets the endpointer internal state.  If reset_threshold is true, the
+  // state will be reset completely, including adaptive thresholds and the
+  // removal of all history information.
+  void Restart(bool reset_threshold);
+
+  // Update internal speech and noise levels.
+  void UpdateLevels(float rms);
+
+  // Returns the number of frames (or frame number) corresponding to
+  // the 'time' (in seconds).
+  int TimeToFrame(float time) const;
+
+  EpStatus status_;  // The current state of this instance.
+  float offset_confirm_dur_sec_;  // max on time allowed to confirm POST_SPEECH
+  int64_t endpointer_time_us_;  // Time of the most recently received audio frame.
+  int64_t fast_update_frames_; // Number of frames for initial level adaptation.
+  int64_t frame_counter_;  // Number of frames seen. Used for initial adaptation.
+  float max_window_dur_;  // Largest search window size (seconds)
+  float sample_rate_;  // Sampling rate.
+
+  // Ring buffers to hold the speech activity history.
+  nsAutoPtr<HistoryRing> history_;
+
+  // Configuration parameters.
+  EnergyEndpointerParams params_;
+
+  // RMS which must be exceeded to conclude frame is speech.
+  float decision_threshold_;
+
+  // Flag to indicate that audio should be used to estimate environment, prior
+  // to receiving user input.
+  bool estimating_environment_;
+
+  // Estimate of the background noise level. Used externally for UI feedback.
+  float noise_level_;
+
+  // An adaptive threshold used to update decision_threshold_ when appropriate.
+  float rms_adapt_;
+
+  // Start lag corresponds to the highest fundamental frequency.
+  int start_lag_;
+
+  // End lag corresponds to the lowest fundamental frequency.
+  int end_lag_;
+
+  // Time when mode switched from environment estimation to user input. This
+  // is used to time forced rejection of audio feedback contamination.
+  int64_t user_input_start_time_us_;
+
+  // prevent copy constructor and assignment
+  EnergyEndpointer(const EnergyEndpointer&);
+  void operator=(const EnergyEndpointer&);
+};
+
+}  // namespace mozilla
+
+#endif  // CONTENT_BROWSER_SPEECH_ENDPOINTER_ENERGY_ENDPOINTER_H_

dom/media/webspeech/recognition/energy_endpointer_params.cc

@@ -0,0 +1,77 @@
+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//    * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//    * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//    * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include "energy_endpointer_params.h"
+
+namespace mozilla {
+
+EnergyEndpointerParams::EnergyEndpointerParams() {
+  SetDefaults();
+}
+
+void EnergyEndpointerParams::SetDefaults() {
+  frame_period_ = 0.01f;
+  frame_duration_ = 0.01f;
+  endpoint_margin_ = 0.2f;
+  onset_window_ = 0.15f;
+  speech_on_window_ = 0.4f;
+  offset_window_ = 0.15f;
+  onset_detect_dur_ = 0.09f;
+  onset_confirm_dur_ = 0.075f;
+  on_maintain_dur_ = 0.10f;
+  offset_confirm_dur_ = 0.12f;
+  decision_threshold_ = 150.0f;
+  min_decision_threshold_ = 50.0f;
+  fast_update_dur_ = 0.2f;
+  sample_rate_ = 8000.0f;
+  min_fundamental_frequency_ = 57.143f;
+  max_fundamental_frequency_ = 400.0f;
+  contamination_rejection_period_ = 0.25f;
+}
+
+void EnergyEndpointerParams::operator=(const EnergyEndpointerParams& source) {
+  frame_period_ = source.frame_period();
+  frame_duration_ = source.frame_duration();
+  endpoint_margin_ = source.endpoint_margin();
+  onset_window_ = source.onset_window();
+  speech_on_window_ = source.speech_on_window();
+  offset_window_ = source.offset_window();
+  onset_detect_dur_ = source.onset_detect_dur();
+  onset_confirm_dur_ = source.onset_confirm_dur();
+  on_maintain_dur_ = source.on_maintain_dur();
+  offset_confirm_dur_ = source.offset_confirm_dur();
+  decision_threshold_ = source.decision_threshold();
+  min_decision_threshold_ = source.min_decision_threshold();
+  fast_update_dur_ = source.fast_update_dur();
+  sample_rate_ = source.sample_rate();
+  min_fundamental_frequency_ = source.min_fundamental_frequency();
+  max_fundamental_frequency_ = source.max_fundamental_frequency();
+  contamination_rejection_period_ = source.contamination_rejection_period();
+}
+
+}  //  namespace mozilla

dom/media/webspeech/recognition/energy_endpointer_params.h

@@ -0,0 +1,159 @@
+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//    * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//    * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//    * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#ifndef CONTENT_BROWSER_SPEECH_ENDPOINTER_ENERGY_ENDPOINTER_PARAMS_H_
+#define CONTENT_BROWSER_SPEECH_ENDPOINTER_ENERGY_ENDPOINTER_PARAMS_H_
+
+namespace mozilla {
+
+// Input parameters for the EnergyEndpointer class.
+class EnergyEndpointerParams {
+ public:
+  EnergyEndpointerParams();
+
+  void SetDefaults();
+
+  void operator=(const EnergyEndpointerParams& source);
+
+  // Accessors and mutators
+  float frame_period() const { return frame_period_; }
+  void set_frame_period(float frame_period) {
+    frame_period_ = frame_period;
+  }
+
+  float frame_duration() const { return frame_duration_; }
+  void set_frame_duration(float frame_duration) {
+    frame_duration_ = frame_duration;
+  }
+
+  float endpoint_margin() const { return endpoint_margin_; }
+  void set_endpoint_margin(float endpoint_margin) {
+    endpoint_margin_ = endpoint_margin;
+  }
+
+  float onset_window() const { return onset_window_; }
+  void set_onset_window(float onset_window) { onset_window_ = onset_window; }
+
+  float speech_on_window() const { return speech_on_window_; }
+  void set_speech_on_window(float speech_on_window) {
+    speech_on_window_ = speech_on_window;
+  }
+
+  float offset_window() const { return offset_window_; }
+  void set_offset_window(float offset_window) {
+    offset_window_ = offset_window;
+  }
+
+  float onset_detect_dur() const { return onset_detect_dur_; }
+  void set_onset_detect_dur(float onset_detect_dur) {
+    onset_detect_dur_ = onset_detect_dur;
+  }
+
+  float onset_confirm_dur() const { return onset_confirm_dur_; }
+  void set_onset_confirm_dur(float onset_confirm_dur) {
+    onset_confirm_dur_ = onset_confirm_dur;
+  }
+
+  float on_maintain_dur() const { return on_maintain_dur_; }
+  void set_on_maintain_dur(float on_maintain_dur) {
+    on_maintain_dur_ = on_maintain_dur;
+  }
+
+  float offset_confirm_dur() const { return offset_confirm_dur_; }
+  void set_offset_confirm_dur(float offset_confirm_dur) {
+    offset_confirm_dur_ = offset_confirm_dur;
+  }
+
+  float decision_threshold() const { return decision_threshold_; }
+  void set_decision_threshold(float decision_threshold) {
+    decision_threshold_ = decision_threshold;
+  }
+
+  float min_decision_threshold() const { return min_decision_threshold_; }
+  void set_min_decision_threshold(float min_decision_threshold) {
+    min_decision_threshold_ = min_decision_threshold;
+  }
+
+  float fast_update_dur() const { return fast_update_dur_; }
+  void set_fast_update_dur(float fast_update_dur) {
+    fast_update_dur_ = fast_update_dur;
+  }
+
+  float sample_rate() const { return sample_rate_; }
+  void set_sample_rate(float sample_rate) { sample_rate_ = sample_rate; }
+
+  float min_fundamental_frequency() const { return min_fundamental_frequency_; }
+  void set_min_fundamental_frequency(float min_fundamental_frequency) {
+    min_fundamental_frequency_ = min_fundamental_frequency;
+  }
+
+  float max_fundamental_frequency() const { return max_fundamental_frequency_; }
+  void set_max_fundamental_frequency(float max_fundamental_frequency) {
+    max_fundamental_frequency_ = max_fundamental_frequency;
+  }
+
+  float contamination_rejection_period() const {
+    return contamination_rejection_period_;
+  }
+  void set_contamination_rejection_period(
+      float contamination_rejection_period) {
+    contamination_rejection_period_ = contamination_rejection_period;
+  }
+
+ private:
+  float frame_period_;  // Frame period
+  float frame_duration_;  // Window size
+  float onset_window_;  // Interval scanned for onset activity
+  float speech_on_window_;  // Inverval scanned for ongoing speech
+  float offset_window_;  // Interval scanned for offset evidence
+  float offset_confirm_dur_;  // Silence duration required to confirm offset
+  float decision_threshold_;  // Initial rms detection threshold
+  float min_decision_threshold_;  // Minimum rms detection threshold
+  float fast_update_dur_;  // Period for initial estimation of levels.
+  float sample_rate_;  // Expected sample rate.
+
+  // Time to add on either side of endpoint threshold crossings
+  float endpoint_margin_;
+  // Total dur within onset_window required to enter ONSET state
+  float onset_detect_dur_;
+  // Total on time within onset_window required to enter SPEECH_ON state
+  float onset_confirm_dur_;
+  // Minimum dur in SPEECH_ON state required to maintain ON state
+  float on_maintain_dur_;
+  // Minimum fundamental frequency for autocorrelation.
+  float min_fundamental_frequency_;
+  // Maximum fundamental frequency for autocorrelation.
+  float max_fundamental_frequency_;
+  // Period after start of user input that above threshold values are ignored.
+  // This is to reject audio feedback contamination.
+  float contamination_rejection_period_;
+};
+
+}  //  namespace mozilla
+
+#endif  // CONTENT_BROWSER_SPEECH_ENDPOINTER_ENERGY_ENDPOINTER_PARAMS_H_

dom/media/webspeech/recognition/models/en-US/README

@@ -0,0 +1,34 @@
+/* ====================================================================
+ * Copyright (c) 2015 Alpha Cephei Inc. All rights
+ * reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ALPHA CEPHEI INC. ``AS IS'' AND.
+ * ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,.
+ * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL ALPHA CEPHEI INC.
+ * NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT.
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,.
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY.
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT.
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE.
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ====================================================================
+ *
+ */
+
+This directory contains generic US english acoustic model trained with
+latest sphinxtrain.

dom/media/webspeech/recognition/models/en-US/feat.params

@@ -0,0 +1,12 @@
+-lowerf 130
+-upperf 3700
+-nfilt 20
+-transform dct
+-lifter 22
+-feat 1s_c_d_dd
+-svspec 0-12/13-25/26-38
+-agc none
+-cmn current
+-varnorm no
+-cmninit 30.09,0.49,5.51,7.43,-2.39,-1.64,-4.71,0.03,-5.11,0.20,4.84,6.14,0.35
+-model ptm

dom/media/webspeech/recognition/models/en-US/noisedict

@@ -0,0 +1,5 @@
+<s> SIL
+</s> SIL
+<sil> SIL
+[NOISE] +NSN+
+[SPEECH] +SPN+

dom/media/webspeech/recognition/moz.build

@@ -0,0 +1,88 @@
+# vim: set filetype=python:
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+MOCHITEST_MANIFESTS += ['test/mochitest.ini']
+
+XPIDL_MODULE = 'dom_webspeechrecognition'
+
+XPIDL_SOURCES = [
+  'nsISpeechRecognitionService.idl'
+]
+
+EXPORTS.mozilla.dom += [
+    'SpeechGrammar.h',
+    'SpeechGrammarList.h',
+    'SpeechRecognition.h',
+    'SpeechRecognitionAlternative.h',
+    'SpeechRecognitionResult.h',
+    'SpeechRecognitionResultList.h',
+    'SpeechStreamListener.h',
+]
+
+if CONFIG['MOZ_WEBSPEECH_TEST_BACKEND']:
+    EXPORTS.mozilla.dom += [
+        'test/FakeSpeechRecognitionService.h',
+    ]
+
+if CONFIG['MOZ_WEBSPEECH_POCKETSPHINX']:
+    EXPORTS.mozilla.dom += [
+        'PocketSphinxSpeechRecognitionService.h',
+    ]
+
+UNIFIED_SOURCES += [
+    'endpointer.cc',
+    'energy_endpointer.cc',
+    'energy_endpointer_params.cc',
+    'SpeechGrammar.cpp',
+    'SpeechGrammarList.cpp',
+    'SpeechRecognition.cpp',
+    'SpeechRecognitionAlternative.cpp',
+    'SpeechRecognitionResult.cpp',
+    'SpeechRecognitionResultList.cpp',
+    'SpeechStreamListener.cpp',
+]
+
+if CONFIG['MOZ_WEBSPEECH_TEST_BACKEND']:
+    UNIFIED_SOURCES += [
+        'test/FakeSpeechRecognitionService.cpp',
+    ]
+
+if CONFIG['MOZ_WEBSPEECH_POCKETSPHINX']:
+    UNIFIED_SOURCES += [
+        'PocketSphinxSpeechRecognitionService.cpp',
+    ]
+
+LOCAL_INCLUDES += [
+    '/dom/base',
+    '/media/sphinxbase',
+]
+
+if CONFIG['MOZ_WEBSPEECH_POCKETSPHINX']:
+    LOCAL_INCLUDES += [
+        '/media/pocketsphinx',
+    ]
+
+if CONFIG['MOZ_WEBSPEECH_MODELS']:
+    FINAL_TARGET_FILES.models.dict += [
+        'models/dict/en-US.dic',
+        'models/dict/en-US.dic.dmp',
+    ]
+    FINAL_TARGET_FILES.models['en-US'] += [
+		'models/en-US/feat.params',
+		'models/en-US/mdef',
+		'models/en-US/means',
+		'models/en-US/mixture_weights',
+		'models/en-US/noisedict',
+		'models/en-US/sendump',
+		'models/en-US/transition_matrices',
+		'models/en-US/variances',
+    ]
+
+include('/ipc/chromium/chromium-config.mozbuild')
+
+FINAL_LIBRARY = 'xul'
+
+if CONFIG['GNU_CXX']:
+    CXXFLAGS += ['-Wno-error=shadow']

dom/media/webspeech/recognition/nsISpeechRecognitionService.idl

@@ -0,0 +1,43 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsISupports.idl"
+
+%{C++
+#include "mozilla/WeakPtr.h"
+
+namespace mozilla {
+class AudioSegment;
+namespace dom {
+class SpeechRecognition;
+class SpeechRecognitionResultList;
+class SpeechGrammarList;
+class SpeechGrammar;
+}
+}
+%}
+
+native SpeechRecognitionWeakPtr(mozilla::WeakPtr<mozilla::dom::SpeechRecognition>);
+[ptr] native AudioSegmentPtr(mozilla::AudioSegment);
+[ptr] native SpeechGrammarPtr(mozilla::dom::SpeechGrammar);
+[ptr] native SpeechGrammarListPtr(mozilla::dom::SpeechGrammarList);
+
+[uuid(6fcb6ee8-a6db-49ba-9f06-355d7ee18ea7)]
+interface nsISpeechGrammarCompilationCallback : nsISupports {
+    void grammarCompilationEnd(in SpeechGrammarPtr grammarObject, in boolean success);
+};
+
+[uuid(8e97f287-f322-44e8-8888-8344fa408ef8)]
+interface nsISpeechRecognitionService : nsISupports {
+    void initialize(in SpeechRecognitionWeakPtr aSpeechRecognition);
+    void processAudioSegment(in AudioSegmentPtr aAudioSegment, in long aSampleRate);
+    void validateAndSetGrammarList(in SpeechGrammarPtr aSpeechGrammar, in nsISpeechGrammarCompilationCallback aCallback);
+    void soundEnd();
+    void abort();
+};
+
+%{C++
+#define NS_SPEECH_RECOGNITION_SERVICE_CONTRACTID_PREFIX "@mozilla.org/webspeech/service;1?name="
+%}

dom/media/webspeech/recognition/test/FakeSpeechRecognitionService.cpp

@@ -0,0 +1,119 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsThreadUtils.h"
+
+#include "FakeSpeechRecognitionService.h"
+#include "MediaPrefs.h"
+
+#include "SpeechRecognition.h"
+#include "SpeechRecognitionAlternative.h"
+#include "SpeechRecognitionResult.h"
+#include "SpeechRecognitionResultList.h"
+#include "nsIObserverService.h"
+#include "mozilla/Services.h"
+
+namespace mozilla {
+
+using namespace dom;
+
+NS_IMPL_ISUPPORTS(FakeSpeechRecognitionService, nsISpeechRecognitionService, nsIObserver)
+
+FakeSpeechRecognitionService::FakeSpeechRecognitionService()
+{
+}
+
+FakeSpeechRecognitionService::~FakeSpeechRecognitionService()
+{
+}
+
+NS_IMETHODIMP
+FakeSpeechRecognitionService::Initialize(WeakPtr<SpeechRecognition> aSpeechRecognition)
+{
+  mRecognition = aSpeechRecognition;
+  nsCOMPtr<nsIObserverService> obs = services::GetObserverService();
+  obs->AddObserver(this, SPEECH_RECOGNITION_TEST_EVENT_REQUEST_TOPIC, false);
+  obs->AddObserver(this, SPEECH_RECOGNITION_TEST_END_TOPIC, false);
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+FakeSpeechRecognitionService::ProcessAudioSegment(AudioSegment* aAudioSegment, int32_t aSampleRate)
+{
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+FakeSpeechRecognitionService::SoundEnd()
+{
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+FakeSpeechRecognitionService::ValidateAndSetGrammarList(mozilla::dom::SpeechGrammar*, nsISpeechGrammarCompilationCallback*)
+{
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+FakeSpeechRecognitionService::Abort()
+{
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+FakeSpeechRecognitionService::Observe(nsISupports* aSubject, const char* aTopic, const char16_t* aData)
+{
+  MOZ_ASSERT(MediaPrefs::WebSpeechFakeRecognitionService(),
+             "Got request to fake recognition service event, but "
+             TEST_PREFERENCE_FAKE_RECOGNITION_SERVICE " is not set");
+
+  if (!strcmp(aTopic, SPEECH_RECOGNITION_TEST_END_TOPIC)) {
+    nsCOMPtr<nsIObserverService> obs = services::GetObserverService();
+    obs->RemoveObserver(this, SPEECH_RECOGNITION_TEST_EVENT_REQUEST_TOPIC);
+    obs->RemoveObserver(this, SPEECH_RECOGNITION_TEST_END_TOPIC);
+
+    return NS_OK;
+  }
+
+  const nsDependentString eventName = nsDependentString(aData);
+
+  if (eventName.EqualsLiteral("EVENT_RECOGNITIONSERVICE_ERROR")) {
+    mRecognition->DispatchError(SpeechRecognition::EVENT_RECOGNITIONSERVICE_ERROR,
+                                SpeechRecognitionErrorCode::Network, // TODO different codes?
+                                NS_LITERAL_STRING("RECOGNITIONSERVICE_ERROR test event"));
+
+  } else if (eventName.EqualsLiteral("EVENT_RECOGNITIONSERVICE_FINAL_RESULT")) {
+    RefPtr<SpeechEvent> event =
+      new SpeechEvent(mRecognition,
+                      SpeechRecognition::EVENT_RECOGNITIONSERVICE_FINAL_RESULT);
+
+    event->mRecognitionResultList = BuildMockResultList();
+    NS_DispatchToMainThread(event);
+  }
+
+  return NS_OK;
+}
+
+SpeechRecognitionResultList*
+FakeSpeechRecognitionService::BuildMockResultList()
+{
+  SpeechRecognitionResultList* resultList = new SpeechRecognitionResultList(mRecognition);
+  SpeechRecognitionResult* result = new SpeechRecognitionResult(mRecognition);
+  if (0 < mRecognition->MaxAlternatives()) {
+    SpeechRecognitionAlternative* alternative = new SpeechRecognitionAlternative(mRecognition);
+
+    alternative->mTranscript = NS_LITERAL_STRING("Mock final result");
+    alternative->mConfidence = 0.0f;
+
+    result->mItems.AppendElement(alternative);
+  }
+  resultList->mItems.AppendElement(result);
+
+  return resultList;
+}
+
+} // namespace mozilla

dom/media/webspeech/recognition/test/FakeSpeechRecognitionService.h

@@ -0,0 +1,38 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_FakeSpeechRecognitionService_h
+#define mozilla_dom_FakeSpeechRecognitionService_h
+
+#include "nsCOMPtr.h"
+#include "nsIObserver.h"
+#include "nsISpeechRecognitionService.h"
+
+#define NS_FAKE_SPEECH_RECOGNITION_SERVICE_CID \
+  {0x48c345e7, 0x9929, 0x4f9a, {0xa5, 0x63, 0xf4, 0x78, 0x22, 0x2d, 0xab, 0xcd}};
+
+namespace mozilla {
+
+class FakeSpeechRecognitionService : public nsISpeechRecognitionService,
+                                     public nsIObserver
+{
+public:
+  NS_DECL_ISUPPORTS
+  NS_DECL_NSISPEECHRECOGNITIONSERVICE
+  NS_DECL_NSIOBSERVER
+
+  FakeSpeechRecognitionService();
+
+private:
+  virtual ~FakeSpeechRecognitionService();
+
+  WeakPtr<dom::SpeechRecognition> mRecognition;
+  dom::SpeechRecognitionResultList* BuildMockResultList();
+};
+
+} // namespace mozilla
+
+#endif

dom/media/webspeech/recognition/test/head.js

@@ -0,0 +1,181 @@
+"use strict";
+
+const DEFAULT_AUDIO_SAMPLE_FILE = "hello.ogg";
+const SPEECH_RECOGNITION_TEST_REQUEST_EVENT_TOPIC = "SpeechRecognitionTest:RequestEvent";
+const SPEECH_RECOGNITION_TEST_END_TOPIC = "SpeechRecognitionTest:End";
+
+var errorCodes = {
+  NO_SPEECH : "no-speech",
+  ABORTED : "aborted",
+  AUDIO_CAPTURE : "audio-capture",
+  NETWORK : "network",
+  NOT_ALLOWED : "not-allowed",
+  SERVICE_NOT_ALLOWED : "service-not-allowed",
+  BAD_GRAMMAR : "bad-grammar",
+  LANGUAGE_NOT_SUPPORTED : "language-not-supported"
+};
+
+var Services = SpecialPowers.Cu.import("resource://gre/modules/Services.jsm").Services;
+
+function EventManager(sr) {
+  var self = this;
+  var nEventsExpected = 0;
+  self.eventsReceived = [];
+
+  var allEvents = [
+    "audiostart",
+    "soundstart",
+    "speechstart",
+    "speechend",
+    "soundend",
+    "audioend",
+    "result",
+    "nomatch",
+    "error",
+    "start",
+    "end"
+  ];
+
+  var eventDependencies = {
+    "speechend": "speechstart",
+    "soundend": "soundstart",
+    "audioend": "audiostart"
+  };
+
+  var isDone = false;
+
+  // set up grammar
+  var sgl = new SpeechGrammarList();
+  sgl.addFromString("#JSGF V1.0; grammar test; public <simple> = hello ;", 1);
+  sr.grammars = sgl;
+
+  // AUDIO_DATA events are asynchronous,
+  // so we queue events requested while they are being
+  // issued to make them seem synchronous
+  var isSendingAudioData = false;
+  var queuedEventRequests = [];
+
+  // register default handlers
+  for (var i = 0; i < allEvents.length; i++) {
+    (function (eventName) {
+      sr["on" + eventName] = function (evt) {
+        var message = "unexpected event: " + eventName;
+        if (eventName == "error") {
+          message += " -- " + evt.message;
+        }
+
+        ok(false, message);
+        if (self.doneFunc && !isDone) {
+          isDone = true;
+          self.doneFunc();
+        }
+      };
+    })(allEvents[i]);
+  }
+
+  self.expect = function EventManager_expect(eventName, cb) {
+    nEventsExpected++;
+
+    sr["on" + eventName] = function(evt) {
+      self.eventsReceived.push(eventName);
+      ok(true, "received event " + eventName);
+
+      var dep = eventDependencies[eventName];
+      if (dep) {
+        ok(self.eventsReceived.indexOf(dep) >= 0,
+           eventName + " must come after " + dep);
+      }
+
+      cb && cb(evt, sr);
+      if (self.doneFunc && !isDone &&
+          nEventsExpected === self.eventsReceived.length) {
+        isDone = true;
+        self.doneFunc();
+      }
+    }
+  }
+
+  self.start = function EventManager_start() {
+    isSendingAudioData = true;
+    var audioTag = document.createElement("audio");
+    audioTag.src = self.audioSampleFile;
+
+    var stream = audioTag.mozCaptureStreamUntilEnded();
+    audioTag.addEventListener("ended", function() {
+      info("Sample stream ended, requesting queued events");
+      isSendingAudioData = false;
+      while (queuedEventRequests.length) {
+        self.requestFSMEvent(queuedEventRequests.shift());
+      }
+    });
+
+    audioTag.play();
+    sr.start(stream);
+  }
+
+  self.requestFSMEvent = function EventManager_requestFSMEvent(eventName) {
+    if (isSendingAudioData) {
+      info("Queuing event " + eventName + " until we're done sending audio data");
+      queuedEventRequests.push(eventName);
+      return;
+    }
+
+    info("requesting " + eventName);
+    Services.obs.notifyObservers(null,
+                                 SPEECH_RECOGNITION_TEST_REQUEST_EVENT_TOPIC,
+                                 eventName);
+  }
+
+  self.requestTestEnd = function EventManager_requestTestEnd() {
+    Services.obs.notifyObservers(null, SPEECH_RECOGNITION_TEST_END_TOPIC, null);
+  }
+}
+
+function buildResultCallback(transcript) {
+  return (function(evt) {
+    is(evt.results[0][0].transcript, transcript, "expect correct transcript");
+  });
+}
+
+function buildErrorCallback(errcode) {
+  return (function(err) {
+    is(err.error, errcode, "expect correct error code");
+  });
+}
+
+function performTest(options) {
+  var prefs = options.prefs;
+
+  prefs.unshift(
+    ["media.webspeech.recognition.enable", true],
+    ["media.webspeech.test.enable", true]
+  );
+
+  SpecialPowers.pushPrefEnv({set: prefs}, function() {
+    var sr = new SpeechRecognition();
+    var em = new EventManager(sr);
+
+    for (var eventName in options.expectedEvents) {
+      var cb = options.expectedEvents[eventName];
+      em.expect(eventName, cb);
+    }
+
+    em.doneFunc = function() {
+      em.requestTestEnd();
+      if (options.doneFunc) {
+        options.doneFunc();
+      }
+    }
+
+    em.audioSampleFile = DEFAULT_AUDIO_SAMPLE_FILE;
+    if (options.audioSampleFile) {
+      em.audioSampleFile = options.audioSampleFile;
+    }
+
+    em.start();
+
+    for (var i = 0; i < options.eventsToRequest.length; i++) {
+      em.requestFSMEvent(options.eventsToRequest[i]);
+    }
+  });
+}

dom/media/webspeech/recognition/test/hello.ogg^headers^

@@ -0,0 +1 @@
+Cache-Control: no-store

dom/media/webspeech/recognition/test/mochitest.ini

@@ -0,0 +1,21 @@
+[DEFAULT]
+tags=msg
+subsuite = media
+support-files =
+  head.js
+  hello.ogg
+  hello.ogg^headers^
+  silence.ogg
+  silence.ogg^headers^
+[test_abort.html]
+skip-if = toolkit == 'android' # bug 1037287
+[test_audio_capture_error.html]
+[test_call_start_from_end_handler.html]
+tags=capturestream
+skip-if = (android_version == '18' && debug) # bug 967606
+[test_nested_eventloop.html]
+skip-if = toolkit == 'android'
+[test_preference_enable.html]
+[test_recognition_service_error.html]
+[test_success_without_recognition_service.html]
+[test_timeout.html]

dom/media/webspeech/recognition/test/silence.ogg^headers^

@@ -0,0 +1 @@
+Cache-Control: no-store

dom/media/webspeech/recognition/test/test_abort.html

@@ -0,0 +1,71 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295 -- Call abort from inside handlers</title>
+  <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script>
+  <link rel="stylesheet" type="text/css" href="/tests/SimpleTest/test.css"/>
+  <script type="application/javascript" src="head.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="text/javascript">
+  SimpleTest.waitForExplicitFinish();
+
+  // Abort inside event handlers, should't get a
+  // result after that
+
+  var nextEventIdx = 0;
+  var eventsToAbortOn = [
+    "start",
+    "audiostart",
+    "speechstart",
+    "speechend",
+    "audioend"
+  ];
+
+  function doNextTest() {
+    var nextEvent = eventsToAbortOn[nextEventIdx];
+    var expectedEvents = {
+      "start": null,
+      "audiostart": null,
+      "audioend": null,
+      "end": null
+    };
+
+    if (nextEventIdx >= eventsToAbortOn.indexOf("speechstart")) {
+        expectedEvents["speechstart"] = null;
+    }
+
+    if (nextEventIdx >= eventsToAbortOn.indexOf("speechend")) {
+        expectedEvents["speechend"] = null;
+    }
+
+    info("Aborting on " + nextEvent);
+    expectedEvents[nextEvent] = function(evt, sr) {
+      sr.abort();
+    };
+
+    nextEventIdx++;
+
+    performTest({
+      eventsToRequest: [],
+      expectedEvents: expectedEvents,
+      doneFunc: (nextEventIdx < eventsToAbortOn.length) ? doNextTest : SimpleTest.finish,
+      prefs: [["media.webspeech.test.fake_fsm_events", true], ["media.webspeech.test.fake_recognition_service", true]]
+    });
+  }
+
+  doNextTest();
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/recognition/test/test_audio_capture_error.html

@@ -0,0 +1,40 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295 -- Behavior on audio error</title>
+  <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script>
+  <link rel="stylesheet" type="text/css" href="/tests/SimpleTest/test.css"/>
+  <script type="application/javascript" src="head.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="text/javascript">
+  SimpleTest.waitForExplicitFinish();
+
+  performTest({
+    eventsToRequest: ['EVENT_AUDIO_ERROR'],
+    expectedEvents: {
+      'start': null,
+      'audiostart': null,
+      'speechstart': null,
+      'speechend': null,
+      'audioend': null,
+      'error': buildErrorCallback(errorCodes.AUDIO_CAPTURE),
+      'end': null
+    },
+    doneFunc: SimpleTest.finish,
+    prefs: [["media.webspeech.test.fake_fsm_events", true], ["media.webspeech.test.fake_recognition_service", true]]
+  });
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/recognition/test/test_call_start_from_end_handler.html

@@ -0,0 +1,100 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295 -- Restart recognition from end handler</title>
+  <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script>
+  <link rel="stylesheet" type="text/css" href="/tests/SimpleTest/test.css"/>
+  <script type="application/javascript" src="head.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="text/javascript">
+  SimpleTest.waitForExplicitFinish();
+
+  function createAudioStream() {
+    var audioTag = document.createElement("audio");
+    audioTag.src = DEFAULT_AUDIO_SAMPLE_FILE;
+
+    var stream = audioTag.mozCaptureStreamUntilEnded();
+    audioTag.play();
+
+    return stream;
+  }
+
+  var done = false;
+  function endHandler(evt, sr) {
+    if (done) {
+      SimpleTest.finish();
+      return;
+    }
+
+    try {
+      var stream = createAudioStream();
+      sr.start(stream); // shouldn't fail
+    } catch (err) {
+      ok(false, "Failed to start() from end() callback");
+    }
+
+    // calling start() may cause some callbacks to fire, but we're
+    // no longer interested in them, except for onend, which is where
+    // we'll conclude the test.
+    sr.onstart = null;
+    sr.onaudiostart = null;
+    sr.onspeechstart = null;
+    sr.onspeechend = null;
+    sr.onaudioend = null;
+    sr.onresult = null;
+
+    // FIXME(ggp) the state transition caused by start() is async,
+    // but abort() is sync (see bug 1055093). until we normalize
+    // state transitions, we need to setTimeout here to make sure
+    // abort() finds the speech recognition object in the correct
+    // state (namely, STATE_STARTING).
+    setTimeout(function() {
+      sr.abort();
+      done = true;
+    });
+
+    info("Successfully start() from end() callback");
+  }
+
+  function expectExceptionHandler(evt, sr) {
+    try {
+      sr.start(createAudioStream());
+    } catch (err) {
+      is(err.name, "InvalidStateError");
+      return;
+    }
+
+    ok(false, "Calling start() didn't raise InvalidStateError");
+  }
+
+  performTest({
+    eventsToRequest: [
+      'EVENT_RECOGNITIONSERVICE_FINAL_RESULT'
+    ],
+    expectedEvents: {
+      'start': expectExceptionHandler,
+      'audiostart': expectExceptionHandler,
+      'speechstart': expectExceptionHandler,
+      'speechend': expectExceptionHandler,
+      'audioend': expectExceptionHandler,
+      'result': buildResultCallback("Mock final result"),
+      'end': endHandler,
+    },
+    prefs: [["media.webspeech.test.fake_fsm_events", true], ["media.webspeech.test.fake_recognition_service", true]]
+  });
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/recognition/test/test_nested_eventloop.html

@@ -0,0 +1,81 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295 -- Spin the event loop from inside a callback</title>
+  <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script>
+  <link rel="stylesheet" type="text/css" href="/tests/SimpleTest/test.css"/>
+  <script type="application/javascript" src="head.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="text/javascript">
+  SimpleTest.waitForExplicitFinish();
+
+  /*
+   * window.showModalDialog() can be used to spin the event loop, causing
+   * queued SpeechEvents (such as those created by calls to start(), stop()
+   * or abort()) to be processed immediately.
+   * When this is done from inside DOM event handlers, it is possible to
+   * cause reentrancy in our C++ code, which we should be able to withstand.
+   */
+  function abortAndSpinEventLoop(evt, sr) {
+    sr.abort();
+    SpecialPowers.spinEventLoop(window);
+  }
+  function doneFunc() {
+    // Trigger gc now and wait some time to make sure this test gets the blame
+    // for any assertions caused by showModalDialog
+    //
+    // NB - The assertions should be gone, but this looks too scary to touch
+    // during batch cleanup.
+    var count = 0, GC_COUNT = 4;
+
+    function triggerGCOrFinish() {
+      SpecialPowers.gc();
+      count++;
+
+      if (count == GC_COUNT) {
+        SimpleTest.finish();
+      }
+    }
+
+    for (var i = 0; i < GC_COUNT; i++) {
+      setTimeout(triggerGCOrFinish, 0);
+    }
+  }
+
+  /*
+   * We start by performing a normal start, then abort from the audiostart
+   * callback and force the EVENT_ABORT to be processed while still inside
+   * the event handler. This causes the recording to stop, which raises
+   * the audioend and (later on) end events.
+   * Then, we abort (once again spinning the event loop) from the audioend
+   * handler, attempting to cause a re-entry into the abort code. This second
+   * call should be ignored, and we get the end callback and finish.
+   */
+
+  performTest({
+    eventsToRequest: [],
+    expectedEvents: {
+      "audiostart": abortAndSpinEventLoop,
+      "audioend": abortAndSpinEventLoop,
+      "end": null
+    },
+    doneFunc: doneFunc,
+    prefs: [["media.webspeech.test.fake_fsm_events", true],
+            ["media.webspeech.test.fake_recognition_service", true]]
+  });
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/recognition/test/test_preference_enable.html

@@ -0,0 +1,43 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295 -- No objects should be visible with preference disabled</title>
+  <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script>
+  <link rel="stylesheet" type="text/css" href="/tests/SimpleTest/test.css"/>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="text/javascript">
+  SimpleTest.waitForExplicitFinish();
+
+  SpecialPowers.pushPrefEnv({
+    set: [["media.webspeech.recognition.enable", false]]
+  }, function() {
+    var objects = [
+      "SpeechRecognition",
+      "SpeechGrammar",
+      "SpeechRecognitionResult",
+      "SpeechRecognitionResultList",
+      "SpeechRecognitionAlternative"
+    ];
+
+    for (var i = 0; i < objects.length; i++) {
+      is(window[objects[i]], undefined,
+         objects[i] + " should be undefined with pref off");
+    }
+
+    SimpleTest.finish();
+  });
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/recognition/test/test_recognition_service_error.html

@@ -0,0 +1,43 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295 -- Behavior on recognition service error</title>
+  <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script>
+  <link rel="stylesheet" type="text/css" href="/tests/SimpleTest/test.css"/>
+  <script type="application/javascript" src="head.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="text/javascript">
+  SimpleTest.waitForExplicitFinish();
+
+  performTest({
+    eventsToRequest: [
+      'EVENT_RECOGNITIONSERVICE_ERROR'
+    ],
+    expectedEvents: {
+      'start': null,
+      'audiostart': null,
+      'speechstart': null,
+      'speechend': null,
+      'audioend': null,
+      'error': buildErrorCallback(errorCodes.NETWORK),
+      'end': null
+    },
+    doneFunc: SimpleTest.finish,
+    prefs: [["media.webspeech.test.fake_fsm_events", true], ["media.webspeech.test.fake_recognition_service", true]]
+  });
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/recognition/test/test_success_without_recognition_service.html

@@ -0,0 +1,43 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295 -- Success with fake recognition service</title>
+  <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script>
+  <link rel="stylesheet" type="text/css" href="/tests/SimpleTest/test.css"/>
+  <script type="application/javascript" src="head.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="text/javascript">
+  SimpleTest.waitForExplicitFinish();
+
+  performTest({
+    eventsToRequest: [
+      'EVENT_RECOGNITIONSERVICE_FINAL_RESULT'
+    ],
+    expectedEvents: {
+      'start': null,
+      'audiostart': null,
+      'speechstart': null,
+      'speechend': null,
+      'audioend': null,
+      'result': buildResultCallback("Mock final result"),
+      'end': null
+    },
+    doneFunc:SimpleTest.finish,
+    prefs: [["media.webspeech.test.fake_fsm_events", true], ["media.webspeech.test.fake_recognition_service", true]]
+  });
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/recognition/test/test_timeout.html

@@ -0,0 +1,40 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295 -- Timeout for user speech</title>
+  <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script>
+  <link rel="stylesheet" type="text/css" href="/tests/SimpleTest/test.css"/>
+  <script type="application/javascript" src="head.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="text/javascript">
+  SimpleTest.waitForExplicitFinish();
+
+  performTest({
+    eventsToRequest: [],
+    expectedEvents: {
+      "start": null,
+      "audiostart": null,
+      "audioend": null,
+      "error": buildErrorCallback(errorCodes.NO_SPEECH),
+      "end": null
+    },
+    doneFunc: SimpleTest.finish,
+    audioSampleFile: "silence.ogg",
+    prefs: [["media.webspeech.test.fake_fsm_events", true], ["media.webspeech.test.fake_recognition_service", true]]
+  });
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/SpeechSynthesis.cpp

@@ -0,0 +1,333 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsISupportsPrimitives.h"
+#include "nsSpeechTask.h"
+#include "mozilla/Logging.h"
+
+#include "mozilla/dom/ContentChild.h"
+#include "mozilla/dom/Element.h"
+
+#include "mozilla/dom/SpeechSynthesisBinding.h"
+#include "SpeechSynthesis.h"
+#include "nsSynthVoiceRegistry.h"
+#include "nsIDocument.h"
+
+#undef LOG
+mozilla::LogModule*
+GetSpeechSynthLog()
+{
+  static mozilla::LazyLogModule sLog("SpeechSynthesis");
+
+  return sLog;
+}
+#define LOG(type, msg) MOZ_LOG(GetSpeechSynthLog(), type, msg)
+
+namespace mozilla {
+namespace dom {
+
+NS_IMPL_CYCLE_COLLECTION_CLASS(SpeechSynthesis)
+
+NS_IMPL_CYCLE_COLLECTION_UNLINK_BEGIN_INHERITED(SpeechSynthesis, DOMEventTargetHelper)
+  NS_IMPL_CYCLE_COLLECTION_UNLINK(mCurrentTask)
+  NS_IMPL_CYCLE_COLLECTION_UNLINK(mSpeechQueue)
+  tmp->mVoiceCache.Clear();
+NS_IMPL_CYCLE_COLLECTION_UNLINK_END
+
+NS_IMPL_CYCLE_COLLECTION_TRAVERSE_BEGIN_INHERITED(SpeechSynthesis, DOMEventTargetHelper)
+  NS_IMPL_CYCLE_COLLECTION_TRAVERSE(mCurrentTask)
+  NS_IMPL_CYCLE_COLLECTION_TRAVERSE(mSpeechQueue)
+  for (auto iter = tmp->mVoiceCache.Iter(); !iter.Done(); iter.Next()) {
+    SpeechSynthesisVoice* voice = iter.UserData();
+    cb.NoteXPCOMChild(voice);
+  }
+NS_IMPL_CYCLE_COLLECTION_TRAVERSE_END
+
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION_INHERITED(SpeechSynthesis)
+  NS_INTERFACE_MAP_ENTRY(nsIObserver)
+  NS_INTERFACE_MAP_ENTRY(nsISupportsWeakReference)
+NS_INTERFACE_MAP_END_INHERITING(DOMEventTargetHelper)
+
+NS_IMPL_ADDREF_INHERITED(SpeechSynthesis, DOMEventTargetHelper)
+NS_IMPL_RELEASE_INHERITED(SpeechSynthesis, DOMEventTargetHelper)
+
+SpeechSynthesis::SpeechSynthesis(nsPIDOMWindowInner* aParent)
+  : DOMEventTargetHelper(aParent)
+  , mHoldQueue(false)
+  , mInnerID(aParent->WindowID())
+{
+  MOZ_ASSERT(aParent->IsInnerWindow());
+  MOZ_ASSERT(NS_IsMainThread());
+
+  nsCOMPtr<nsIObserverService> obs = mozilla::services::GetObserverService();
+  if (obs) {
+    obs->AddObserver(this, "inner-window-destroyed", true);
+    obs->AddObserver(this, "synth-voices-changed", true);
+  }
+
+}
+
+SpeechSynthesis::~SpeechSynthesis()
+{
+}
+
+JSObject*
+SpeechSynthesis::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto)
+{
+  return SpeechSynthesisBinding::Wrap(aCx, this, aGivenProto);
+}
+
+bool
+SpeechSynthesis::Pending() const
+{
+  switch (mSpeechQueue.Length()) {
+  case 0:
+    return false;
+
+  case 1:
+    return mSpeechQueue.ElementAt(0)->GetState() == SpeechSynthesisUtterance::STATE_PENDING;
+
+  default:
+    return true;
+  }
+}
+
+bool
+SpeechSynthesis::Speaking() const
+{
+  if (!mSpeechQueue.IsEmpty() &&
+      mSpeechQueue.ElementAt(0)->GetState() == SpeechSynthesisUtterance::STATE_SPEAKING) {
+    return true;
+  }
+
+  // Returns global speaking state if global queue is enabled. Or false.
+  return nsSynthVoiceRegistry::GetInstance()->IsSpeaking();
+}
+
+bool
+SpeechSynthesis::Paused() const
+{
+  return mHoldQueue || (mCurrentTask && mCurrentTask->IsPrePaused()) ||
+         (!mSpeechQueue.IsEmpty() && mSpeechQueue.ElementAt(0)->IsPaused());
+}
+
+bool
+SpeechSynthesis::HasEmptyQueue() const
+{
+  return mSpeechQueue.Length() == 0;
+}
+
+bool SpeechSynthesis::HasVoices() const
+{
+  uint32_t voiceCount = mVoiceCache.Count();
+  if (voiceCount == 0) {
+    nsresult rv = nsSynthVoiceRegistry::GetInstance()->GetVoiceCount(&voiceCount);
+    if(NS_WARN_IF(NS_FAILED(rv))) {
+      return false;
+    }
+  }
+
+  return voiceCount != 0;
+}
+
+void
+SpeechSynthesis::Speak(SpeechSynthesisUtterance& aUtterance)
+{
+  if (aUtterance.mState != SpeechSynthesisUtterance::STATE_NONE) {
+    // XXX: Should probably raise an error
+    return;
+  }
+
+  mSpeechQueue.AppendElement(&aUtterance);
+  aUtterance.mState = SpeechSynthesisUtterance::STATE_PENDING;
+
+  // If we only have one item in the queue, we aren't pre-paused, and
+  // we have voices available, speak it.
+  if (mSpeechQueue.Length() == 1 && !mCurrentTask && !mHoldQueue && HasVoices()) {
+    AdvanceQueue();
+  }
+}
+
+void
+SpeechSynthesis::AdvanceQueue()
+{
+  LOG(LogLevel::Debug,
+      ("SpeechSynthesis::AdvanceQueue length=%d", mSpeechQueue.Length()));
+
+  if (mSpeechQueue.IsEmpty()) {
+    return;
+  }
+
+  RefPtr<SpeechSynthesisUtterance> utterance = mSpeechQueue.ElementAt(0);
+
+  nsAutoString docLang;
+  nsCOMPtr<nsPIDOMWindowInner> window = GetOwner();
+  nsIDocument* doc = window ? window->GetExtantDoc() : nullptr;
+
+  if (doc) {
+    Element* elm = doc->GetHtmlElement();
+
+    if (elm) {
+      elm->GetLang(docLang);
+    }
+  }
+
+  mCurrentTask =
+    nsSynthVoiceRegistry::GetInstance()->SpeakUtterance(*utterance, docLang);
+
+  if (mCurrentTask) {
+    mCurrentTask->SetSpeechSynthesis(this);
+  }
+
+  return;
+}
+
+void
+SpeechSynthesis::Cancel()
+{
+  if (!mSpeechQueue.IsEmpty() &&
+      mSpeechQueue.ElementAt(0)->GetState() == SpeechSynthesisUtterance::STATE_SPEAKING) {
+    // Remove all queued utterances except for current one, we will remove it
+    // in OnEnd
+    mSpeechQueue.RemoveElementsAt(1, mSpeechQueue.Length() - 1);
+  } else {
+    mSpeechQueue.Clear();
+  }
+
+  if (mCurrentTask) {
+    mCurrentTask->Cancel();
+  }
+}
+
+void
+SpeechSynthesis::Pause()
+{
+  if (Paused()) {
+    return;
+  }
+
+  if (mCurrentTask && !mSpeechQueue.IsEmpty() &&
+      mSpeechQueue.ElementAt(0)->GetState() != SpeechSynthesisUtterance::STATE_ENDED) {
+    mCurrentTask->Pause();
+  } else {
+    mHoldQueue = true;
+  }
+}
+
+void
+SpeechSynthesis::Resume()
+{
+  if (!Paused()) {
+    return;
+  }
+
+  if (mCurrentTask) {
+    mCurrentTask->Resume();
+  } else {
+    mHoldQueue = false;
+    AdvanceQueue();
+  }
+}
+
+void
+SpeechSynthesis::OnEnd(const nsSpeechTask* aTask)
+{
+  MOZ_ASSERT(mCurrentTask == aTask);
+
+  if (!mSpeechQueue.IsEmpty()) {
+    mSpeechQueue.RemoveElementAt(0);
+  }
+
+  mCurrentTask = nullptr;
+  AdvanceQueue();
+}
+
+void
+SpeechSynthesis::GetVoices(nsTArray< RefPtr<SpeechSynthesisVoice> >& aResult)
+{
+  aResult.Clear();
+  uint32_t voiceCount = 0;
+
+  nsresult rv = nsSynthVoiceRegistry::GetInstance()->GetVoiceCount(&voiceCount);
+  if(NS_WARN_IF(NS_FAILED(rv))) {
+    return;
+  }
+
+  nsISupports* voiceParent = NS_ISUPPORTS_CAST(nsIObserver*, this);
+
+  for (uint32_t i = 0; i < voiceCount; i++) {
+    nsAutoString uri;
+    rv = nsSynthVoiceRegistry::GetInstance()->GetVoice(i, uri);
+
+    if (NS_FAILED(rv)) {
+      NS_WARNING("Failed to retrieve voice from registry");
+      continue;
+    }
+
+    SpeechSynthesisVoice* voice = mVoiceCache.GetWeak(uri);
+
+    if (!voice) {
+      voice = new SpeechSynthesisVoice(voiceParent, uri);
+    }
+
+    aResult.AppendElement(voice);
+  }
+
+  mVoiceCache.Clear();
+
+  for (uint32_t i = 0; i < aResult.Length(); i++) {
+    SpeechSynthesisVoice* voice = aResult[i];
+    mVoiceCache.Put(voice->mUri, voice);
+  }
+}
+
+// For testing purposes, allows us to cancel the current task that is
+// misbehaving, and flush the queue.
+void
+SpeechSynthesis::ForceEnd()
+{
+  if (mCurrentTask) {
+    mCurrentTask->ForceEnd();
+  }
+}
+
+NS_IMETHODIMP
+SpeechSynthesis::Observe(nsISupports* aSubject, const char* aTopic,
+                         const char16_t* aData)
+{
+  MOZ_ASSERT(NS_IsMainThread());
+
+
+  if (strcmp(aTopic, "inner-window-destroyed") == 0) {
+    nsCOMPtr<nsISupportsPRUint64> wrapper = do_QueryInterface(aSubject);
+    NS_ENSURE_TRUE(wrapper, NS_ERROR_FAILURE);
+
+    uint64_t innerID;
+    nsresult rv = wrapper->GetData(&innerID);
+    NS_ENSURE_SUCCESS(rv, rv);
+
+    if (innerID == mInnerID) {
+      Cancel();
+
+      nsCOMPtr<nsIObserverService> obs = mozilla::services::GetObserverService();
+      if (obs) {
+        obs->RemoveObserver(this, "inner-window-destroyed");
+      }
+    }
+  } else if (strcmp(aTopic, "synth-voices-changed") == 0) {
+    LOG(LogLevel::Debug, ("SpeechSynthesis::onvoiceschanged"));
+    DispatchTrustedEvent(NS_LITERAL_STRING("voiceschanged"));
+    // If we have a pending item, and voices become available, speak it.
+    if (!mCurrentTask && !mHoldQueue && HasVoices()) {
+      AdvanceQueue();
+    }
+  }
+
+  return NS_OK;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/SpeechSynthesis.h

@@ -0,0 +1,85 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechSynthesis_h
+#define mozilla_dom_SpeechSynthesis_h
+
+#include "nsCOMPtr.h"
+#include "nsIObserver.h"
+#include "nsRefPtrHashtable.h"
+#include "nsString.h"
+#include "nsWeakReference.h"
+#include "nsWrapperCache.h"
+#include "js/TypeDecls.h"
+
+#include "SpeechSynthesisUtterance.h"
+#include "SpeechSynthesisVoice.h"
+
+class nsIDOMWindow;
+
+namespace mozilla {
+namespace dom {
+
+class nsSpeechTask;
+
+class SpeechSynthesis final : public DOMEventTargetHelper
+                            , public nsIObserver
+                            , public nsSupportsWeakReference
+{
+public:
+  explicit SpeechSynthesis(nsPIDOMWindowInner* aParent);
+
+  NS_DECL_ISUPPORTS_INHERITED
+  NS_DECL_CYCLE_COLLECTION_CLASS_INHERITED(SpeechSynthesis, DOMEventTargetHelper)
+  NS_DECL_NSIOBSERVER
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  bool Pending() const;
+
+  bool Speaking() const;
+
+  bool Paused() const;
+
+  bool HasEmptyQueue() const;
+
+  void Speak(SpeechSynthesisUtterance& aUtterance);
+
+  void Cancel();
+
+  void Pause();
+
+  void Resume();
+
+  void OnEnd(const nsSpeechTask* aTask);
+
+  void GetVoices(nsTArray< RefPtr<SpeechSynthesisVoice> >& aResult);
+
+  void ForceEnd();
+
+  IMPL_EVENT_HANDLER(voiceschanged)
+
+private:
+  virtual ~SpeechSynthesis();
+
+  void AdvanceQueue();
+
+  bool HasVoices() const;
+
+  nsTArray<RefPtr<SpeechSynthesisUtterance> > mSpeechQueue;
+
+  RefPtr<nsSpeechTask> mCurrentTask;
+
+  nsRefPtrHashtable<nsStringHashKey, SpeechSynthesisVoice> mVoiceCache;
+
+  bool mHoldQueue;
+
+  uint64_t mInnerID;
+};
+
+} // namespace dom
+} // namespace mozilla
+#endif

dom/media/webspeech/synth/SpeechSynthesisUtterance.cpp

@@ -0,0 +1,178 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsCOMPtr.h"
+#include "nsCycleCollectionParticipant.h"
+#include "nsGkAtoms.h"
+
+#include "mozilla/dom/SpeechSynthesisEvent.h"
+#include "mozilla/dom/SpeechSynthesisUtteranceBinding.h"
+#include "SpeechSynthesisUtterance.h"
+#include "SpeechSynthesisVoice.h"
+
+#include <stdlib.h>
+
+namespace mozilla {
+namespace dom {
+
+NS_IMPL_CYCLE_COLLECTION_INHERITED(SpeechSynthesisUtterance,
+                                   DOMEventTargetHelper, mVoice);
+
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION_INHERITED(SpeechSynthesisUtterance)
+NS_INTERFACE_MAP_END_INHERITING(DOMEventTargetHelper)
+
+NS_IMPL_ADDREF_INHERITED(SpeechSynthesisUtterance, DOMEventTargetHelper)
+NS_IMPL_RELEASE_INHERITED(SpeechSynthesisUtterance, DOMEventTargetHelper)
+
+SpeechSynthesisUtterance::SpeechSynthesisUtterance(nsPIDOMWindowInner* aOwnerWindow,
+                                                   const nsAString& text)
+  : DOMEventTargetHelper(aOwnerWindow)
+  , mText(text)
+  , mVolume(1)
+  , mRate(1)
+  , mPitch(1)
+  , mState(STATE_NONE)
+  , mPaused(false)
+{
+}
+
+SpeechSynthesisUtterance::~SpeechSynthesisUtterance() {}
+
+JSObject*
+SpeechSynthesisUtterance::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto)
+{
+  return SpeechSynthesisUtteranceBinding::Wrap(aCx, this, aGivenProto);
+}
+
+nsISupports*
+SpeechSynthesisUtterance::GetParentObject() const
+{
+  return GetOwner();
+}
+
+already_AddRefed<SpeechSynthesisUtterance>
+SpeechSynthesisUtterance::Constructor(GlobalObject& aGlobal,
+                                      ErrorResult& aRv)
+{
+  return Constructor(aGlobal, EmptyString(), aRv);
+}
+
+already_AddRefed<SpeechSynthesisUtterance>
+SpeechSynthesisUtterance::Constructor(GlobalObject& aGlobal,
+                                      const nsAString& aText,
+                                      ErrorResult& aRv)
+{
+  nsCOMPtr<nsPIDOMWindowInner> win = do_QueryInterface(aGlobal.GetAsSupports());
+
+  if (!win) {
+    aRv.Throw(NS_ERROR_FAILURE);
+  }
+
+  MOZ_ASSERT(win->IsInnerWindow());
+  RefPtr<SpeechSynthesisUtterance> object =
+    new SpeechSynthesisUtterance(win, aText);
+  return object.forget();
+}
+
+void
+SpeechSynthesisUtterance::GetText(nsString& aResult) const
+{
+  aResult = mText;
+}
+
+void
+SpeechSynthesisUtterance::SetText(const nsAString& aText)
+{
+  mText = aText;
+}
+
+void
+SpeechSynthesisUtterance::GetLang(nsString& aResult) const
+{
+  aResult = mLang;
+}
+
+void
+SpeechSynthesisUtterance::SetLang(const nsAString& aLang)
+{
+  mLang = aLang;
+}
+
+SpeechSynthesisVoice*
+SpeechSynthesisUtterance::GetVoice() const
+{
+  return mVoice;
+}
+
+void
+SpeechSynthesisUtterance::SetVoice(SpeechSynthesisVoice* aVoice)
+{
+  mVoice = aVoice;
+}
+
+float
+SpeechSynthesisUtterance::Volume() const
+{
+  return mVolume;
+}
+
+void
+SpeechSynthesisUtterance::SetVolume(float aVolume)
+{
+  mVolume = std::max<float>(std::min<float>(aVolume, 1), 0);
+}
+
+float
+SpeechSynthesisUtterance::Rate() const
+{
+  return mRate;
+}
+
+void
+SpeechSynthesisUtterance::SetRate(float aRate)
+{
+  mRate = std::max<float>(std::min<float>(aRate, 10), 0.1f);
+}
+
+float
+SpeechSynthesisUtterance::Pitch() const
+{
+  return mPitch;
+}
+
+void
+SpeechSynthesisUtterance::SetPitch(float aPitch)
+{
+  mPitch = std::max<float>(std::min<float>(aPitch, 2), 0);
+}
+
+void
+SpeechSynthesisUtterance::GetChosenVoiceURI(nsString& aResult) const
+{
+  aResult = mChosenVoiceURI;
+}
+
+void
+SpeechSynthesisUtterance::DispatchSpeechSynthesisEvent(const nsAString& aEventType,
+                                                       uint32_t aCharIndex,
+                                                       float aElapsedTime,
+                                                       const nsAString& aName)
+{
+  SpeechSynthesisEventInit init;
+  init.mBubbles = false;
+  init.mCancelable = false;
+  init.mUtterance = this;
+  init.mCharIndex = aCharIndex;
+  init.mElapsedTime = aElapsedTime;
+  init.mName = aName;
+
+  RefPtr<SpeechSynthesisEvent> event =
+    SpeechSynthesisEvent::Constructor(this, aEventType, init);
+  DispatchTrustedEvent(event);
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/SpeechSynthesisUtterance.h

@@ -0,0 +1,124 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechSynthesisUtterance_h
+#define mozilla_dom_SpeechSynthesisUtterance_h
+
+#include "mozilla/DOMEventTargetHelper.h"
+#include "nsCOMPtr.h"
+#include "nsString.h"
+#include "js/TypeDecls.h"
+
+#include "nsSpeechTask.h"
+
+namespace mozilla {
+namespace dom {
+
+class SpeechSynthesisVoice;
+class SpeechSynthesis;
+class nsSynthVoiceRegistry;
+
+class SpeechSynthesisUtterance final : public DOMEventTargetHelper
+{
+  friend class SpeechSynthesis;
+  friend class nsSpeechTask;
+  friend class nsSynthVoiceRegistry;
+
+public:
+  SpeechSynthesisUtterance(nsPIDOMWindowInner* aOwnerWindow, const nsAString& aText);
+
+  NS_DECL_ISUPPORTS_INHERITED
+  NS_DECL_CYCLE_COLLECTION_CLASS_INHERITED(SpeechSynthesisUtterance,
+                                           DOMEventTargetHelper)
+  NS_REALLY_FORWARD_NSIDOMEVENTTARGET(DOMEventTargetHelper)
+
+  nsISupports* GetParentObject() const;
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  static
+  already_AddRefed<SpeechSynthesisUtterance> Constructor(GlobalObject& aGlobal,
+                                                         ErrorResult& aRv);
+  static
+  already_AddRefed<SpeechSynthesisUtterance> Constructor(GlobalObject& aGlobal,
+                                                         const nsAString& aText,
+                                                         ErrorResult& aRv);
+
+  void GetText(nsString& aResult) const;
+
+  void SetText(const nsAString& aText);
+
+  void GetLang(nsString& aResult) const;
+
+  void SetLang(const nsAString& aLang);
+
+  SpeechSynthesisVoice* GetVoice() const;
+
+  void SetVoice(SpeechSynthesisVoice* aVoice);
+
+  float Volume() const;
+
+  void SetVolume(float aVolume);
+
+  float Rate() const;
+
+  void SetRate(float aRate);
+
+  float Pitch() const;
+
+  void SetPitch(float aPitch);
+
+  void GetChosenVoiceURI(nsString& aResult) const;
+
+  enum {
+    STATE_NONE,
+    STATE_PENDING,
+    STATE_SPEAKING,
+    STATE_ENDED
+  };
+
+  uint32_t GetState() { return mState; }
+
+  bool IsPaused() { return mPaused; }
+
+  IMPL_EVENT_HANDLER(start)
+  IMPL_EVENT_HANDLER(end)
+  IMPL_EVENT_HANDLER(error)
+  IMPL_EVENT_HANDLER(pause)
+  IMPL_EVENT_HANDLER(resume)
+  IMPL_EVENT_HANDLER(mark)
+  IMPL_EVENT_HANDLER(boundary)
+
+private:
+  virtual ~SpeechSynthesisUtterance();
+
+  void DispatchSpeechSynthesisEvent(const nsAString& aEventType,
+                                    uint32_t aCharIndex,
+                                    float aElapsedTime, const nsAString& aName);
+
+  nsString mText;
+
+  nsString mLang;
+
+  float mVolume;
+
+  float mRate;
+
+  float mPitch;
+
+  nsString mChosenVoiceURI;
+
+  uint32_t mState;
+
+  bool mPaused;
+
+  RefPtr<SpeechSynthesisVoice> mVoice;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/synth/SpeechSynthesisVoice.cpp

@@ -0,0 +1,94 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechSynthesis.h"
+#include "nsSynthVoiceRegistry.h"
+#include "mozilla/dom/SpeechSynthesisVoiceBinding.h"
+
+namespace mozilla {
+namespace dom {
+
+NS_IMPL_CYCLE_COLLECTION_WRAPPERCACHE(SpeechSynthesisVoice, mParent)
+NS_IMPL_CYCLE_COLLECTING_ADDREF(SpeechSynthesisVoice)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(SpeechSynthesisVoice)
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(SpeechSynthesisVoice)
+  NS_WRAPPERCACHE_INTERFACE_MAP_ENTRY
+  NS_INTERFACE_MAP_ENTRY(nsISupports)
+NS_INTERFACE_MAP_END
+
+SpeechSynthesisVoice::SpeechSynthesisVoice(nsISupports* aParent,
+                                           const nsAString& aUri)
+  : mParent(aParent)
+  , mUri(aUri)
+{
+}
+
+SpeechSynthesisVoice::~SpeechSynthesisVoice()
+{
+}
+
+JSObject*
+SpeechSynthesisVoice::WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto)
+{
+  return SpeechSynthesisVoiceBinding::Wrap(aCx, this, aGivenProto);
+}
+
+nsISupports*
+SpeechSynthesisVoice::GetParentObject() const
+{
+  return mParent;
+}
+
+void
+SpeechSynthesisVoice::GetVoiceURI(nsString& aRetval) const
+{
+  aRetval = mUri;
+}
+
+void
+SpeechSynthesisVoice::GetName(nsString& aRetval) const
+{
+  DebugOnly<nsresult> rv =
+    nsSynthVoiceRegistry::GetInstance()->GetVoiceName(mUri, aRetval);
+  NS_WARNING_ASSERTION(NS_SUCCEEDED(rv),
+                       "Failed to get SpeechSynthesisVoice.name");
+}
+
+void
+SpeechSynthesisVoice::GetLang(nsString& aRetval) const
+{
+  DebugOnly<nsresult> rv =
+    nsSynthVoiceRegistry::GetInstance()->GetVoiceLang(mUri, aRetval);
+  NS_WARNING_ASSERTION(NS_SUCCEEDED(rv),
+                       "Failed to get SpeechSynthesisVoice.lang");
+}
+
+bool
+SpeechSynthesisVoice::LocalService() const
+{
+  bool isLocal;
+  DebugOnly<nsresult> rv =
+    nsSynthVoiceRegistry::GetInstance()->IsLocalVoice(mUri, &isLocal);
+  NS_WARNING_ASSERTION(
+    NS_SUCCEEDED(rv), "Failed to get SpeechSynthesisVoice.localService");
+
+  return isLocal;
+}
+
+bool
+SpeechSynthesisVoice::Default() const
+{
+  bool isDefault;
+  DebugOnly<nsresult> rv =
+    nsSynthVoiceRegistry::GetInstance()->IsDefaultVoice(mUri, &isDefault);
+  NS_WARNING_ASSERTION(
+    NS_SUCCEEDED(rv), "Failed to get SpeechSynthesisVoice.default");
+
+  return isDefault;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/SpeechSynthesisVoice.h

@@ -0,0 +1,60 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechSynthesisVoice_h
+#define mozilla_dom_SpeechSynthesisVoice_h
+
+#include "nsCOMPtr.h"
+#include "nsString.h"
+#include "nsWrapperCache.h"
+#include "js/TypeDecls.h"
+
+#include "nsISpeechService.h"
+
+namespace mozilla {
+namespace dom {
+
+class nsSynthVoiceRegistry;
+class SpeechSynthesis;
+
+class SpeechSynthesisVoice final : public nsISupports,
+                                   public nsWrapperCache
+{
+  friend class nsSynthVoiceRegistry;
+  friend class SpeechSynthesis;
+
+public:
+  SpeechSynthesisVoice(nsISupports* aParent, const nsAString& aUri);
+
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_SCRIPT_HOLDER_CLASS(SpeechSynthesisVoice)
+
+  nsISupports* GetParentObject() const;
+
+  JSObject* WrapObject(JSContext* aCx, JS::Handle<JSObject*> aGivenProto) override;
+
+  void GetVoiceURI(nsString& aRetval) const;
+
+  void GetName(nsString& aRetval) const;
+
+  void GetLang(nsString& aRetval) const;
+
+  bool LocalService() const;
+
+  bool Default() const;
+
+private:
+  virtual ~SpeechSynthesisVoice();
+
+  nsCOMPtr<nsISupports> mParent;
+
+  nsString mUri;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/synth/cocoa/OSXSpeechSynthesizerModule.cpp

@@ -0,0 +1,57 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim: set ts=8 sts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "mozilla/ModuleUtils.h"
+#include "nsIClassInfoImpl.h"
+
+#include "OSXSpeechSynthesizerService.h"
+
+using namespace mozilla::dom;
+
+#define OSXSPEECHSYNTHESIZERSERVICE_CID \
+  {0x914e73b4, 0x6337, 0x4bef, {0x97, 0xf3, 0x4d, 0x06, 0x9e, 0x05, 0x3a, 0x12}}
+
+#define OSXSPEECHSYNTHESIZERSERVICE_CONTRACTID "@mozilla.org/synthsystem;1"
+
+// Defines OSXSpeechSynthesizerServiceConstructor
+NS_GENERIC_FACTORY_SINGLETON_CONSTRUCTOR(OSXSpeechSynthesizerService,
+                                         OSXSpeechSynthesizerService::GetInstanceForService)
+
+// Defines kOSXSERVICE_CID
+NS_DEFINE_NAMED_CID(OSXSPEECHSYNTHESIZERSERVICE_CID);
+
+static const mozilla::Module::CIDEntry kCIDs[] = {
+  { &kOSXSPEECHSYNTHESIZERSERVICE_CID, true, nullptr, OSXSpeechSynthesizerServiceConstructor },
+  { nullptr }
+};
+
+static const mozilla::Module::ContractIDEntry kContracts[] = {
+  { OSXSPEECHSYNTHESIZERSERVICE_CONTRACTID, &kOSXSPEECHSYNTHESIZERSERVICE_CID },
+  { nullptr }
+};
+
+static const mozilla::Module::CategoryEntry kCategories[] = {
+  { "speech-synth-started", "OSX Speech Synth", OSXSPEECHSYNTHESIZERSERVICE_CONTRACTID },
+  { nullptr }
+};
+
+static void
+UnloadOSXSpeechSynthesizerModule()
+{
+  OSXSpeechSynthesizerService::Shutdown();
+}
+
+static const mozilla::Module kModule = {
+  mozilla::Module::kVersion,
+  kCIDs,
+  kContracts,
+  kCategories,
+  nullptr,
+  nullptr,
+  UnloadOSXSpeechSynthesizerModule
+};
+
+NSMODULE_DEFN(osxsynth) = &kModule;

dom/media/webspeech/synth/cocoa/OSXSpeechSynthesizerService.h

@@ -0,0 +1,44 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim: set ts=8 sts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_OsxSpeechSynthesizerService_h
+#define mozilla_dom_OsxSpeechSynthesizerService_h
+
+#include "nsISpeechService.h"
+#include "nsIObserver.h"
+#include "mozilla/StaticPtr.h"
+
+namespace mozilla {
+namespace dom {
+
+class OSXSpeechSynthesizerService final : public nsISpeechService
+                                        , public nsIObserver
+{
+public:
+  NS_DECL_THREADSAFE_ISUPPORTS
+  NS_DECL_NSISPEECHSERVICE
+  NS_DECL_NSIOBSERVER
+
+  bool Init();
+
+  static OSXSpeechSynthesizerService* GetInstance();
+  static already_AddRefed<OSXSpeechSynthesizerService> GetInstanceForService();
+  static void Shutdown();
+
+private:
+  OSXSpeechSynthesizerService();
+  virtual ~OSXSpeechSynthesizerService();
+
+  bool RegisterVoices();
+
+  bool mInitialized;
+  static mozilla::StaticRefPtr<OSXSpeechSynthesizerService> sSingleton;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/synth/cocoa/OSXSpeechSynthesizerService.mm

@@ -0,0 +1,499 @@
+/* -*- Mode: Objective-C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim: set ts=2 sw=2 et tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsISupports.h"
+#include "nsServiceManagerUtils.h"
+#include "nsObjCExceptions.h"
+#include "nsCocoaUtils.h"
+#include "nsThreadUtils.h"
+#include "mozilla/dom/nsSynthVoiceRegistry.h"
+#include "mozilla/dom/nsSpeechTask.h"
+#include "mozilla/Preferences.h"
+#include "mozilla/Assertions.h"
+#include "OSXSpeechSynthesizerService.h"
+
+#import <Cocoa/Cocoa.h>
+
+// We can escape the default delimiters ("[[" and "]]") by temporarily
+// changing the delimiters just before they appear, and changing them back
+// just after.
+#define DLIM_ESCAPE_START "[[dlim (( ))]]"
+#define DLIM_ESCAPE_END "((dlim [[ ]]))"
+
+using namespace mozilla;
+
+class SpeechTaskCallback final : public nsISpeechTaskCallback
+{
+public:
+  SpeechTaskCallback(nsISpeechTask* aTask,
+                     NSSpeechSynthesizer* aSynth,
+                     const nsTArray<size_t>& aOffsets)
+    : mTask(aTask)
+    , mSpeechSynthesizer(aSynth)
+    , mOffsets(aOffsets)
+  {
+    mStartingTime = TimeStamp::Now();
+  }
+
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_CLASS_AMBIGUOUS(SpeechTaskCallback, nsISpeechTaskCallback)
+
+  NS_DECL_NSISPEECHTASKCALLBACK
+
+  void OnWillSpeakWord(uint32_t aIndex);
+  void OnError(uint32_t aIndex);
+  void OnDidFinishSpeaking();
+
+private:
+  virtual ~SpeechTaskCallback()
+  {
+    [mSpeechSynthesizer release];
+  }
+
+  float GetTimeDurationFromStart();
+
+  nsCOMPtr<nsISpeechTask> mTask;
+  NSSpeechSynthesizer* mSpeechSynthesizer;
+  TimeStamp mStartingTime;
+  uint32_t mCurrentIndex;
+  nsTArray<size_t> mOffsets;
+};
+
+NS_IMPL_CYCLE_COLLECTION(SpeechTaskCallback, mTask);
+
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(SpeechTaskCallback)
+  NS_INTERFACE_MAP_ENTRY(nsISpeechTaskCallback)
+  NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsISpeechTaskCallback)
+NS_INTERFACE_MAP_END
+
+NS_IMPL_CYCLE_COLLECTING_ADDREF(SpeechTaskCallback)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(SpeechTaskCallback)
+
+NS_IMETHODIMP
+SpeechTaskCallback::OnCancel()
+{
+  NS_OBJC_BEGIN_TRY_ABORT_BLOCK_NSRESULT;
+
+  [mSpeechSynthesizer stopSpeaking];
+  return NS_OK;
+
+  NS_OBJC_END_TRY_ABORT_BLOCK_NSRESULT;
+}
+
+NS_IMETHODIMP
+SpeechTaskCallback::OnPause()
+{
+  NS_OBJC_BEGIN_TRY_ABORT_BLOCK_NSRESULT;
+
+  [mSpeechSynthesizer pauseSpeakingAtBoundary:NSSpeechImmediateBoundary];
+  if (!mTask) {
+    // When calling pause() on child porcess, it may not receive end event
+    // from chrome process yet.
+    return NS_ERROR_FAILURE;
+  }
+  mTask->DispatchPause(GetTimeDurationFromStart(), mCurrentIndex);
+  return NS_OK;
+
+  NS_OBJC_END_TRY_ABORT_BLOCK_NSRESULT;
+}
+
+NS_IMETHODIMP
+SpeechTaskCallback::OnResume()
+{
+  NS_OBJC_BEGIN_TRY_ABORT_BLOCK_NSRESULT;
+
+  [mSpeechSynthesizer continueSpeaking];
+  if (!mTask) {
+    // When calling resume() on child porcess, it may not receive end event
+    // from chrome process yet.
+    return NS_ERROR_FAILURE;
+  }
+  mTask->DispatchResume(GetTimeDurationFromStart(), mCurrentIndex);
+  return NS_OK;
+
+  NS_OBJC_END_TRY_ABORT_BLOCK_NSRESULT;
+}
+
+NS_IMETHODIMP
+SpeechTaskCallback::OnVolumeChanged(float aVolume)
+{
+  NS_OBJC_BEGIN_TRY_ABORT_BLOCK_NSRESULT;
+
+  [mSpeechSynthesizer setObject:[NSNumber numberWithFloat:aVolume]
+                    forProperty:NSSpeechVolumeProperty error:nil];
+  return NS_OK;
+
+  NS_OBJC_END_TRY_ABORT_BLOCK_NSRESULT;
+}
+
+float
+SpeechTaskCallback::GetTimeDurationFromStart()
+{
+  TimeDuration duration = TimeStamp::Now() - mStartingTime;
+  return duration.ToMilliseconds();
+}
+
+void
+SpeechTaskCallback::OnWillSpeakWord(uint32_t aIndex)
+{
+  mCurrentIndex = aIndex < mOffsets.Length() ? mOffsets[aIndex] : mCurrentIndex;
+  if (!mTask) {
+    return;
+  }
+  mTask->DispatchBoundary(NS_LITERAL_STRING("word"),
+                          GetTimeDurationFromStart(), mCurrentIndex);
+}
+
+void
+SpeechTaskCallback::OnError(uint32_t aIndex)
+{
+  if (!mTask) {
+    return;
+  }
+  mTask->DispatchError(GetTimeDurationFromStart(), aIndex);
+}
+
+void
+SpeechTaskCallback::OnDidFinishSpeaking()
+{
+  mTask->DispatchEnd(GetTimeDurationFromStart(), mCurrentIndex);
+  // no longer needed
+  [mSpeechSynthesizer setDelegate:nil];
+  mTask = nullptr;
+}
+
+@interface SpeechDelegate : NSObject<NSSpeechSynthesizerDelegate>
+{
+@private
+  SpeechTaskCallback* mCallback;
+}
+
+  - (id)initWithCallback:(SpeechTaskCallback*)aCallback;
+@end
+
+@implementation SpeechDelegate
+- (id)initWithCallback:(SpeechTaskCallback*)aCallback
+{
+  [super init];
+  mCallback = aCallback;
+  return self;
+}
+
+- (void)speechSynthesizer:(NSSpeechSynthesizer *)aSender
+            willSpeakWord:(NSRange)aRange ofString:(NSString*)aString
+{
+  mCallback->OnWillSpeakWord(aRange.location);
+}
+
+- (void)speechSynthesizer:(NSSpeechSynthesizer *)aSender
+        didFinishSpeaking:(BOOL)aFinishedSpeaking
+{
+  mCallback->OnDidFinishSpeaking();
+}
+
+- (void)speechSynthesizer:(NSSpeechSynthesizer*)aSender
+ didEncounterErrorAtIndex:(NSUInteger)aCharacterIndex
+                 ofString:(NSString*)aString
+                  message:(NSString*)aMessage
+{
+  mCallback->OnError(aCharacterIndex);
+}
+@end
+
+namespace mozilla {
+namespace dom {
+
+struct OSXVoice
+{
+  OSXVoice() : mIsDefault(false)
+  {
+  }
+
+  nsString mUri;
+  nsString mName;
+  nsString mLocale;
+  bool mIsDefault;
+};
+
+class RegisterVoicesRunnable final : public Runnable
+{
+public:
+  RegisterVoicesRunnable(OSXSpeechSynthesizerService* aSpeechService,
+                         nsTArray<OSXVoice>& aList)
+    : mSpeechService(aSpeechService)
+    , mVoices(aList)
+  {
+  }
+
+  NS_IMETHOD Run() override;
+
+private:
+  ~RegisterVoicesRunnable()
+  {
+  }
+
+  // This runnable always use sync mode.  It is unnecesarry to reference object
+  OSXSpeechSynthesizerService* mSpeechService;
+  nsTArray<OSXVoice>& mVoices;
+};
+
+NS_IMETHODIMP
+RegisterVoicesRunnable::Run()
+{
+  nsresult rv;
+  nsCOMPtr<nsISynthVoiceRegistry> registry =
+    do_GetService(NS_SYNTHVOICEREGISTRY_CONTRACTID, &rv);
+  if (!registry) {
+    return rv;
+  }
+
+  for (OSXVoice voice : mVoices) {
+    rv = registry->AddVoice(mSpeechService, voice.mUri, voice.mName, voice.mLocale, true, false);
+    if (NS_WARN_IF(NS_FAILED(rv))) {
+      continue;
+    }
+
+    if (voice.mIsDefault) {
+      registry->SetDefaultVoice(voice.mUri, true);
+    }
+  }
+
+  registry->NotifyVoicesChanged();
+
+  return NS_OK;
+}
+
+class EnumVoicesRunnable final : public Runnable
+{
+public:
+  explicit EnumVoicesRunnable(OSXSpeechSynthesizerService* aSpeechService)
+    : mSpeechService(aSpeechService)
+  {
+  }
+
+  NS_IMETHOD Run() override;
+
+private:
+  ~EnumVoicesRunnable()
+  {
+  }
+
+  RefPtr<OSXSpeechSynthesizerService> mSpeechService;
+};
+
+NS_IMETHODIMP
+EnumVoicesRunnable::Run()
+{
+  NS_OBJC_BEGIN_TRY_ABORT_BLOCK_NSRESULT;
+
+  AutoTArray<OSXVoice, 64> list;
+
+  NSArray* voices = [NSSpeechSynthesizer availableVoices];
+  NSString* defaultVoice = [NSSpeechSynthesizer defaultVoice];
+
+  for (NSString* voice in voices) {
+    OSXVoice item;
+
+    NSDictionary* attr = [NSSpeechSynthesizer attributesForVoice:voice];
+
+    nsAutoString identifier;
+    nsCocoaUtils::GetStringForNSString([attr objectForKey:NSVoiceIdentifier],
+                                       identifier);
+
+    nsCocoaUtils::GetStringForNSString([attr objectForKey:NSVoiceName], item.mName);
+
+    nsCocoaUtils::GetStringForNSString(
+      [attr objectForKey:NSVoiceLocaleIdentifier], item.mLocale);
+    item.mLocale.ReplaceChar('_', '-');
+
+    item.mUri.AssignLiteral("urn:moz-tts:osx:");
+    item.mUri.Append(identifier);
+
+    if ([voice isEqualToString:defaultVoice]) {
+      item.mIsDefault = true;
+    }
+
+    list.AppendElement(item);
+  }
+
+  RefPtr<RegisterVoicesRunnable> runnable = new RegisterVoicesRunnable(mSpeechService, list);
+  NS_DispatchToMainThread(runnable, NS_DISPATCH_SYNC);
+
+  return NS_OK;
+
+  NS_OBJC_END_TRY_ABORT_BLOCK_NSRESULT;
+}
+
+StaticRefPtr<OSXSpeechSynthesizerService> OSXSpeechSynthesizerService::sSingleton;
+
+NS_INTERFACE_MAP_BEGIN(OSXSpeechSynthesizerService)
+  NS_INTERFACE_MAP_ENTRY(nsISpeechService)
+  NS_INTERFACE_MAP_ENTRY(nsIObserver)
+  NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsISpeechService)
+NS_INTERFACE_MAP_END
+
+NS_IMPL_ADDREF(OSXSpeechSynthesizerService)
+NS_IMPL_RELEASE(OSXSpeechSynthesizerService)
+
+OSXSpeechSynthesizerService::OSXSpeechSynthesizerService()
+  : mInitialized(false)
+{
+}
+
+OSXSpeechSynthesizerService::~OSXSpeechSynthesizerService()
+{
+}
+
+bool
+OSXSpeechSynthesizerService::Init()
+{
+  if (Preferences::GetBool("media.webspeech.synth.test") ||
+      !Preferences::GetBool("media.webspeech.synth.enabled")) {
+    // When test is enabled, we shouldn't add OS backend (Bug 1160844)
+    return false;
+  }
+
+  nsCOMPtr<nsIThread> thread;
+  if (NS_FAILED(NS_NewNamedThread("SpeechWorker", getter_AddRefs(thread)))) {
+  	return false;
+  }
+
+  // Get all the voices and register in the SynthVoiceRegistry
+  nsCOMPtr<nsIRunnable> runnable = new EnumVoicesRunnable(this);
+  thread->Dispatch(runnable, NS_DISPATCH_NORMAL);
+
+  mInitialized = true;
+  return true;
+}
+
+NS_IMETHODIMP
+OSXSpeechSynthesizerService::Speak(const nsAString& aText,
+                                   const nsAString& aUri,
+                                   float aVolume,
+                                   float aRate,
+                                   float aPitch,
+                                   nsISpeechTask* aTask)
+{
+  NS_OBJC_BEGIN_TRY_ABORT_BLOCK_NSRESULT;
+
+  MOZ_ASSERT(StringBeginsWith(aUri, NS_LITERAL_STRING("urn:moz-tts:osx:")),
+             "OSXSpeechSynthesizerService doesn't allow this voice URI");
+
+  NSSpeechSynthesizer* synth = [[NSSpeechSynthesizer alloc] init];
+  // strlen("urn:moz-tts:osx:") == 16
+  NSString* identifier = nsCocoaUtils::ToNSString(Substring(aUri, 16));
+  [synth setVoice:identifier];
+
+  // default rate is 180-220
+  [synth setObject:[NSNumber numberWithInt:aRate * 200]
+         forProperty:NSSpeechRateProperty error:nil];
+  // volume allows 0.0-1.0
+  [synth setObject:[NSNumber numberWithFloat:aVolume]
+         forProperty:NSSpeechVolumeProperty error:nil];
+  // Use default pitch value to calculate this
+  NSNumber* defaultPitch =
+    [synth objectForProperty:NSSpeechPitchBaseProperty error:nil];
+  if (defaultPitch) {
+    int newPitch = [defaultPitch intValue] * (aPitch / 2 + 0.5);
+    [synth setObject:[NSNumber numberWithInt:newPitch]
+           forProperty:NSSpeechPitchBaseProperty error:nil];
+  }
+
+  nsAutoString escapedText;
+  // We need to map the the offsets from the given text to the escaped text.
+  // The index of the offsets array is the position in the escaped text,
+  // the element value is the position in the user-supplied text.
+  nsTArray<size_t> offsets;
+  offsets.SetCapacity(aText.Length());
+
+  // This loop looks for occurances of "[[" or "]]", escapes them, and
+  // populates the offsets array to supply a map to the original offsets.
+  for (size_t i = 0; i < aText.Length(); i++) {
+    if (aText.Length() > i + 1 &&
+        ((aText[i] == ']' && aText[i+1] == ']') ||
+         (aText[i] == '[' && aText[i+1] == '['))) {
+      escapedText.AppendLiteral(DLIM_ESCAPE_START);
+      offsets.AppendElements(strlen(DLIM_ESCAPE_START));
+      escapedText.Append(aText[i]);
+      offsets.AppendElement(i);
+      escapedText.Append(aText[++i]);
+      offsets.AppendElement(i);
+      escapedText.AppendLiteral(DLIM_ESCAPE_END);
+      offsets.AppendElements(strlen(DLIM_ESCAPE_END));
+    } else {
+      escapedText.Append(aText[i]);
+      offsets.AppendElement(i);
+    }
+  }
+
+  RefPtr<SpeechTaskCallback> callback = new SpeechTaskCallback(aTask, synth, offsets);
+  nsresult rv = aTask->Setup(callback, 0, 0, 0);
+  NS_ENSURE_SUCCESS(rv, rv);
+
+  SpeechDelegate* delegate = [[SpeechDelegate alloc] initWithCallback:callback];
+  [synth setDelegate:delegate];
+  [delegate release ];
+
+  NSString* text = nsCocoaUtils::ToNSString(escapedText);
+  BOOL success = [synth startSpeakingString:text];
+  NS_ENSURE_TRUE(success, NS_ERROR_FAILURE);
+
+  aTask->DispatchStart();
+  return NS_OK;
+
+  NS_OBJC_END_TRY_ABORT_BLOCK_NSRESULT;
+}
+
+NS_IMETHODIMP
+OSXSpeechSynthesizerService::GetServiceType(SpeechServiceType* aServiceType)
+{
+  *aServiceType = nsISpeechService::SERVICETYPE_INDIRECT_AUDIO;
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+OSXSpeechSynthesizerService::Observe(nsISupports* aSubject, const char* aTopic,
+                                     const char16_t* aData)
+{
+  return NS_OK;
+}
+
+OSXSpeechSynthesizerService*
+OSXSpeechSynthesizerService::GetInstance()
+{
+  MOZ_ASSERT(NS_IsMainThread());
+  if (XRE_GetProcessType() != GeckoProcessType_Default) {
+    return nullptr;
+  }
+
+  if (!sSingleton) {
+    RefPtr<OSXSpeechSynthesizerService> speechService =
+      new OSXSpeechSynthesizerService();
+    if (speechService->Init()) {
+      sSingleton = speechService;
+    }
+  }
+  return sSingleton;
+}
+
+already_AddRefed<OSXSpeechSynthesizerService>
+OSXSpeechSynthesizerService::GetInstanceForService()
+{
+  RefPtr<OSXSpeechSynthesizerService> speechService = GetInstance();
+  return speechService.forget();
+}
+
+void
+OSXSpeechSynthesizerService::Shutdown()
+{
+  if (!sSingleton) {
+    return;
+  }
+  sSingleton = nullptr;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/cocoa/moz.build

@@ -0,0 +1,12 @@
+# -*- Mode: python; indent-tabs-mode: nil; tab-width: 40 -*-
+# vim: set filetype=python:
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+SOURCES += [
+  'OSXSpeechSynthesizerModule.cpp',
+  'OSXSpeechSynthesizerService.mm'
+]
+
+FINAL_LIBRARY = 'xul'

dom/media/webspeech/synth/crashtests/1230428.html

@@ -0,0 +1,32 @@
+<!DOCTYPE html>
+<html class="reftest-wait">
+<head>
+<meta charset="utf-8">
+<script type="application/javascript">
+function f()
+{
+  if (speechSynthesis.getVoices().length == 0) {
+    // No synthesis backend to test this
+    document.documentElement.removeAttribute('class');
+    return;
+  }
+
+  var s = new SpeechSynthesisUtterance("hello world");
+  s.onerror = () => {
+    // No synthesis backend to test this
+    document.documentElement.removeAttribute('class');
+    return;
+  }
+  s.onend = () => {
+    document.documentElement.removeAttribute('class');
+  };
+  speechSynthesis.speak(s);
+  speechSynthesis.cancel();
+  speechSynthesis.pause();
+  speechSynthesis.resume();
+}
+  </script>
+</head>
+<body onload="f();">
+</body>
+</html>

dom/media/webspeech/synth/crashtests/crashtests.list

@@ -0,0 +1 @@
+skip-if(!cocoaWidget) pref(media.webspeech.synth.enabled,true) load 1230428.html # bug 1230428

dom/media/webspeech/synth/ipc/PSpeechSynthesis.ipdl

@@ -0,0 +1,49 @@
+/* -*- Mode: c++; c-basic-offset: 2; indent-tabs-mode: nil; tab-width: 40 -*- */
+/* vim: set ts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this file,
+ * You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+include protocol PContent;
+include protocol PSpeechSynthesisRequest;
+
+namespace mozilla {
+namespace dom {
+
+struct RemoteVoice {
+  nsString voiceURI;
+  nsString name;
+  nsString lang;
+  bool localService;
+  bool queued;
+};
+
+sync protocol PSpeechSynthesis
+{
+  manager PContent;
+  manages PSpeechSynthesisRequest;
+
+child:
+
+    async VoiceAdded(RemoteVoice aVoice);
+
+    async VoiceRemoved(nsString aUri);
+
+    async SetDefaultVoice(nsString aUri, bool aIsDefault);
+
+    async IsSpeakingChanged(bool aIsSpeaking);
+
+    async NotifyVoicesChanged();
+
+parent:
+    async __delete__();
+
+    async PSpeechSynthesisRequest(nsString aText, nsString aUri, nsString aLang,
+                                  float aVolume, float aRate, float aPitch);
+
+    sync ReadVoicesAndState() returns (RemoteVoice[] aVoices,
+                                       nsString[] aDefaults, bool aIsSpeaking);
+};
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/ipc/PSpeechSynthesisRequest.ipdl

@@ -0,0 +1,46 @@
+/* -*- Mode: c++; c-basic-offset: 2; indent-tabs-mode: nil; tab-width: 40 -*- */
+/* vim: set ts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this file,
+ * You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+include protocol PSpeechSynthesis;
+
+namespace mozilla {
+namespace dom {
+
+async protocol PSpeechSynthesisRequest
+{
+  manager PSpeechSynthesis;
+
+ parent:
+
+  async __delete__();
+
+  async Pause();
+
+  async Resume();
+
+  async Cancel();
+
+  async ForceEnd();
+
+  async SetAudioOutputVolume(float aVolume);
+
+ child:
+
+  async OnEnd(bool aIsError, float aElapsedTime, uint32_t aCharIndex);
+
+  async OnStart(nsString aUri);
+
+  async OnPause(float aElapsedTime, uint32_t aCharIndex);
+
+  async OnResume(float aElapsedTime, uint32_t aCharIndex);
+
+  async OnBoundary(nsString aName, float aElapsedTime, uint32_t aCharIndex);
+
+  async OnMark(nsString aName, float aElapsedTime, uint32_t aCharIndex);
+};
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/ipc/SpeechSynthesisChild.cpp

@@ -0,0 +1,213 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this file,
+ * You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechSynthesisChild.h"
+#include "nsSynthVoiceRegistry.h"
+
+namespace mozilla {
+namespace dom {
+
+SpeechSynthesisChild::SpeechSynthesisChild()
+{
+  MOZ_COUNT_CTOR(SpeechSynthesisChild);
+}
+
+SpeechSynthesisChild::~SpeechSynthesisChild()
+{
+  MOZ_COUNT_DTOR(SpeechSynthesisChild);
+}
+
+bool
+SpeechSynthesisChild::RecvVoiceAdded(const RemoteVoice& aVoice)
+{
+  nsSynthVoiceRegistry::RecvAddVoice(aVoice);
+  return true;
+}
+
+bool
+SpeechSynthesisChild::RecvVoiceRemoved(const nsString& aUri)
+{
+  nsSynthVoiceRegistry::RecvRemoveVoice(aUri);
+  return true;
+}
+
+bool
+SpeechSynthesisChild::RecvSetDefaultVoice(const nsString& aUri,
+                                          const bool& aIsDefault)
+{
+  nsSynthVoiceRegistry::RecvSetDefaultVoice(aUri, aIsDefault);
+  return true;
+}
+
+bool
+SpeechSynthesisChild::RecvIsSpeakingChanged(const bool& aIsSpeaking)
+{
+  nsSynthVoiceRegistry::RecvIsSpeakingChanged(aIsSpeaking);
+  return true;
+}
+
+bool
+SpeechSynthesisChild::RecvNotifyVoicesChanged()
+{
+  nsSynthVoiceRegistry::RecvNotifyVoicesChanged();
+  return true;
+}
+
+PSpeechSynthesisRequestChild*
+SpeechSynthesisChild::AllocPSpeechSynthesisRequestChild(const nsString& aText,
+                                                        const nsString& aLang,
+                                                        const nsString& aUri,
+                                                        const float& aVolume,
+                                                        const float& aRate,
+                                                        const float& aPitch)
+{
+  MOZ_CRASH("Caller is supposed to manually construct a request!");
+}
+
+bool
+SpeechSynthesisChild::DeallocPSpeechSynthesisRequestChild(PSpeechSynthesisRequestChild* aActor)
+{
+  delete aActor;
+  return true;
+}
+
+// SpeechSynthesisRequestChild
+
+SpeechSynthesisRequestChild::SpeechSynthesisRequestChild(SpeechTaskChild* aTask)
+  : mTask(aTask)
+{
+  mTask->mActor = this;
+  MOZ_COUNT_CTOR(SpeechSynthesisRequestChild);
+}
+
+SpeechSynthesisRequestChild::~SpeechSynthesisRequestChild()
+{
+  MOZ_COUNT_DTOR(SpeechSynthesisRequestChild);
+}
+
+bool
+SpeechSynthesisRequestChild::RecvOnStart(const nsString& aUri)
+{
+  mTask->DispatchStartImpl(aUri);
+  return true;
+}
+
+bool
+SpeechSynthesisRequestChild::RecvOnEnd(const bool& aIsError,
+                                       const float& aElapsedTime,
+                                       const uint32_t& aCharIndex)
+{
+  SpeechSynthesisRequestChild* actor = mTask->mActor;
+  mTask->mActor = nullptr;
+
+  if (aIsError) {
+    mTask->DispatchErrorImpl(aElapsedTime, aCharIndex);
+  } else {
+    mTask->DispatchEndImpl(aElapsedTime, aCharIndex);
+  }
+
+  actor->Send__delete__(actor);
+
+  return true;
+}
+
+bool
+SpeechSynthesisRequestChild::RecvOnPause(const float& aElapsedTime,
+                                         const uint32_t& aCharIndex)
+{
+  mTask->DispatchPauseImpl(aElapsedTime, aCharIndex);
+  return true;
+}
+
+bool
+SpeechSynthesisRequestChild::RecvOnResume(const float& aElapsedTime,
+                                          const uint32_t& aCharIndex)
+{
+  mTask->DispatchResumeImpl(aElapsedTime, aCharIndex);
+  return true;
+}
+
+bool
+SpeechSynthesisRequestChild::RecvOnBoundary(const nsString& aName,
+                                            const float& aElapsedTime,
+                                            const uint32_t& aCharIndex)
+{
+  mTask->DispatchBoundaryImpl(aName, aElapsedTime, aCharIndex);
+  return true;
+}
+
+bool
+SpeechSynthesisRequestChild::RecvOnMark(const nsString& aName,
+                                        const float& aElapsedTime,
+                                        const uint32_t& aCharIndex)
+{
+  mTask->DispatchMarkImpl(aName, aElapsedTime, aCharIndex);
+  return true;
+}
+
+// SpeechTaskChild
+
+SpeechTaskChild::SpeechTaskChild(SpeechSynthesisUtterance* aUtterance)
+  : nsSpeechTask(aUtterance)
+{
+}
+
+NS_IMETHODIMP
+SpeechTaskChild::Setup(nsISpeechTaskCallback* aCallback,
+                       uint32_t aChannels, uint32_t aRate, uint8_t argc)
+{
+  MOZ_CRASH("Should never be called from child");
+}
+
+NS_IMETHODIMP
+SpeechTaskChild::SendAudio(JS::Handle<JS::Value> aData, JS::Handle<JS::Value> aLandmarks,
+                           JSContext* aCx)
+{
+  MOZ_CRASH("Should never be called from child");
+}
+
+NS_IMETHODIMP
+SpeechTaskChild::SendAudioNative(int16_t* aData, uint32_t aDataLen)
+{
+  MOZ_CRASH("Should never be called from child");
+}
+
+void
+SpeechTaskChild::Pause()
+{
+  MOZ_ASSERT(mActor);
+  mActor->SendPause();
+}
+
+void
+SpeechTaskChild::Resume()
+{
+  MOZ_ASSERT(mActor);
+  mActor->SendResume();
+}
+
+void
+SpeechTaskChild::Cancel()
+{
+  MOZ_ASSERT(mActor);
+  mActor->SendCancel();
+}
+
+void
+SpeechTaskChild::ForceEnd()
+{
+  MOZ_ASSERT(mActor);
+  mActor->SendForceEnd();
+}
+
+void
+SpeechTaskChild::SetAudioOutputVolume(float aVolume)
+{
+  if (mActor) {
+    mActor->SendSetAudioOutputVolume(aVolume);
+  }
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/ipc/SpeechSynthesisChild.h

@@ -0,0 +1,106 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this file,
+ * You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechSynthesisChild_h
+#define mozilla_dom_SpeechSynthesisChild_h
+
+#include "mozilla/Attributes.h"
+#include "mozilla/dom/PSpeechSynthesisChild.h"
+#include "mozilla/dom/PSpeechSynthesisRequestChild.h"
+#include "nsSpeechTask.h"
+
+namespace mozilla {
+namespace dom {
+
+class nsSynthVoiceRegistry;
+class SpeechSynthesisRequestChild;
+class SpeechTaskChild;
+
+class SpeechSynthesisChild : public PSpeechSynthesisChild
+{
+  friend class nsSynthVoiceRegistry;
+
+public:
+  bool RecvVoiceAdded(const RemoteVoice& aVoice) override;
+
+  bool RecvVoiceRemoved(const nsString& aUri) override;
+
+  bool RecvSetDefaultVoice(const nsString& aUri, const bool& aIsDefault) override;
+
+  bool RecvIsSpeakingChanged(const bool& aIsSpeaking) override;
+
+  bool RecvNotifyVoicesChanged() override;
+
+protected:
+  SpeechSynthesisChild();
+  virtual ~SpeechSynthesisChild();
+
+  PSpeechSynthesisRequestChild* AllocPSpeechSynthesisRequestChild(const nsString& aLang,
+                                                                  const nsString& aUri,
+                                                                  const nsString& aText,
+                                                                  const float& aVolume,
+                                                                  const float& aPitch,
+                                                                  const float& aRate) override;
+  bool DeallocPSpeechSynthesisRequestChild(PSpeechSynthesisRequestChild* aActor) override;
+};
+
+class SpeechSynthesisRequestChild : public PSpeechSynthesisRequestChild
+{
+public:
+  explicit SpeechSynthesisRequestChild(SpeechTaskChild* aTask);
+  virtual ~SpeechSynthesisRequestChild();
+
+protected:
+  bool RecvOnStart(const nsString& aUri) override;
+
+  bool RecvOnEnd(const bool& aIsError,
+                 const float& aElapsedTime,
+                 const uint32_t& aCharIndex) override;
+
+  bool RecvOnPause(const float& aElapsedTime, const uint32_t& aCharIndex) override;
+
+  bool RecvOnResume(const float& aElapsedTime, const uint32_t& aCharIndex) override;
+
+  bool RecvOnBoundary(const nsString& aName, const float& aElapsedTime,
+                      const uint32_t& aCharIndex) override;
+
+  bool RecvOnMark(const nsString& aName, const float& aElapsedTime,
+                  const uint32_t& aCharIndex) override;
+
+  RefPtr<SpeechTaskChild> mTask;
+};
+
+class SpeechTaskChild : public nsSpeechTask
+{
+  friend class SpeechSynthesisRequestChild;
+public:
+
+  explicit SpeechTaskChild(SpeechSynthesisUtterance* aUtterance);
+
+  NS_IMETHOD Setup(nsISpeechTaskCallback* aCallback,
+                   uint32_t aChannels, uint32_t aRate, uint8_t argc) override;
+
+  NS_IMETHOD SendAudio(JS::Handle<JS::Value> aData, JS::Handle<JS::Value> aLandmarks,
+                       JSContext* aCx) override;
+
+  NS_IMETHOD SendAudioNative(int16_t* aData, uint32_t aDataLen) override;
+
+  void Pause() override;
+
+  void Resume() override;
+
+  void Cancel() override;
+
+  void ForceEnd() override;
+
+  void SetAudioOutputVolume(float aVolume) override;
+
+private:
+  SpeechSynthesisRequestChild* mActor;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/synth/ipc/SpeechSynthesisParent.cpp

@@ -0,0 +1,234 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this file,
+ * You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechSynthesisParent.h"
+#include "nsSynthVoiceRegistry.h"
+
+namespace mozilla {
+namespace dom {
+
+SpeechSynthesisParent::SpeechSynthesisParent()
+{
+  MOZ_COUNT_CTOR(SpeechSynthesisParent);
+}
+
+SpeechSynthesisParent::~SpeechSynthesisParent()
+{
+  MOZ_COUNT_DTOR(SpeechSynthesisParent);
+}
+
+void
+SpeechSynthesisParent::ActorDestroy(ActorDestroyReason aWhy)
+{
+  // Implement me! Bug 1005141
+}
+
+bool
+SpeechSynthesisParent::RecvReadVoicesAndState(InfallibleTArray<RemoteVoice>* aVoices,
+                                              InfallibleTArray<nsString>* aDefaults,
+                                              bool* aIsSpeaking)
+{
+  nsSynthVoiceRegistry::GetInstance()->SendVoicesAndState(aVoices, aDefaults,
+                                                          aIsSpeaking);
+  return true;
+}
+
+PSpeechSynthesisRequestParent*
+SpeechSynthesisParent::AllocPSpeechSynthesisRequestParent(const nsString& aText,
+                                                          const nsString& aLang,
+                                                          const nsString& aUri,
+                                                          const float& aVolume,
+                                                          const float& aRate,
+                                                          const float& aPitch)
+{
+  RefPtr<SpeechTaskParent> task = new SpeechTaskParent(aVolume, aText);
+  SpeechSynthesisRequestParent* actor = new SpeechSynthesisRequestParent(task);
+  return actor;
+}
+
+bool
+SpeechSynthesisParent::DeallocPSpeechSynthesisRequestParent(PSpeechSynthesisRequestParent* aActor)
+{
+  delete aActor;
+  return true;
+}
+
+bool
+SpeechSynthesisParent::RecvPSpeechSynthesisRequestConstructor(PSpeechSynthesisRequestParent* aActor,
+                                                              const nsString& aText,
+                                                              const nsString& aLang,
+                                                              const nsString& aUri,
+                                                              const float& aVolume,
+                                                              const float& aRate,
+                                                              const float& aPitch)
+{
+  MOZ_ASSERT(aActor);
+  SpeechSynthesisRequestParent* actor =
+    static_cast<SpeechSynthesisRequestParent*>(aActor);
+  nsSynthVoiceRegistry::GetInstance()->Speak(aText, aLang, aUri, aVolume, aRate,
+                                             aPitch, actor->mTask);
+  return true;
+}
+
+// SpeechSynthesisRequestParent
+
+SpeechSynthesisRequestParent::SpeechSynthesisRequestParent(SpeechTaskParent* aTask)
+  : mTask(aTask)
+{
+  mTask->mActor = this;
+  MOZ_COUNT_CTOR(SpeechSynthesisRequestParent);
+}
+
+SpeechSynthesisRequestParent::~SpeechSynthesisRequestParent()
+{
+  if (mTask) {
+    mTask->mActor = nullptr;
+    // If we still have a task, cancel it.
+    mTask->Cancel();
+  }
+  MOZ_COUNT_DTOR(SpeechSynthesisRequestParent);
+}
+
+void
+SpeechSynthesisRequestParent::ActorDestroy(ActorDestroyReason aWhy)
+{
+  // Implement me! Bug 1005141
+}
+
+bool
+SpeechSynthesisRequestParent::RecvPause()
+{
+  MOZ_ASSERT(mTask);
+  mTask->Pause();
+  return true;
+}
+
+bool
+SpeechSynthesisRequestParent::Recv__delete__()
+{
+  MOZ_ASSERT(mTask);
+  mTask->mActor = nullptr;
+  mTask = nullptr;
+  return true;
+}
+
+bool
+SpeechSynthesisRequestParent::RecvResume()
+{
+  MOZ_ASSERT(mTask);
+  mTask->Resume();
+  return true;
+}
+
+bool
+SpeechSynthesisRequestParent::RecvCancel()
+{
+  MOZ_ASSERT(mTask);
+  mTask->Cancel();
+  return true;
+}
+
+bool
+SpeechSynthesisRequestParent::RecvForceEnd()
+{
+  MOZ_ASSERT(mTask);
+  mTask->ForceEnd();
+  return true;
+}
+
+bool
+SpeechSynthesisRequestParent::RecvSetAudioOutputVolume(const float& aVolume)
+{
+  MOZ_ASSERT(mTask);
+  mTask->SetAudioOutputVolume(aVolume);
+  return true;
+}
+
+// SpeechTaskParent
+
+nsresult
+SpeechTaskParent::DispatchStartImpl(const nsAString& aUri)
+{
+  MOZ_ASSERT(mActor);
+  if(NS_WARN_IF(!(mActor->SendOnStart(nsString(aUri))))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+nsresult
+SpeechTaskParent::DispatchEndImpl(float aElapsedTime, uint32_t aCharIndex)
+{
+  if (!mActor) {
+    // Child is already gone.
+    return NS_OK;
+  }
+
+  if(NS_WARN_IF(!(mActor->SendOnEnd(false, aElapsedTime, aCharIndex)))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+nsresult
+SpeechTaskParent::DispatchPauseImpl(float aElapsedTime, uint32_t aCharIndex)
+{
+  MOZ_ASSERT(mActor);
+  if(NS_WARN_IF(!(mActor->SendOnPause(aElapsedTime, aCharIndex)))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+nsresult
+SpeechTaskParent::DispatchResumeImpl(float aElapsedTime, uint32_t aCharIndex)
+{
+  MOZ_ASSERT(mActor);
+  if(NS_WARN_IF(!(mActor->SendOnResume(aElapsedTime, aCharIndex)))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+nsresult
+SpeechTaskParent::DispatchErrorImpl(float aElapsedTime, uint32_t aCharIndex)
+{
+  MOZ_ASSERT(mActor);
+  if(NS_WARN_IF(!(mActor->SendOnEnd(true, aElapsedTime, aCharIndex)))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+nsresult
+SpeechTaskParent::DispatchBoundaryImpl(const nsAString& aName,
+                                       float aElapsedTime, uint32_t aCharIndex)
+{
+  MOZ_ASSERT(mActor);
+  if(NS_WARN_IF(!(mActor->SendOnBoundary(nsString(aName), aElapsedTime, aCharIndex)))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+nsresult
+SpeechTaskParent::DispatchMarkImpl(const nsAString& aName,
+                                   float aElapsedTime, uint32_t aCharIndex)
+{
+  MOZ_ASSERT(mActor);
+  if(NS_WARN_IF(!(mActor->SendOnMark(nsString(aName), aElapsedTime, aCharIndex)))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/ipc/SpeechSynthesisParent.h

@@ -0,0 +1,108 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this file,
+ * You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechSynthesisParent_h
+#define mozilla_dom_SpeechSynthesisParent_h
+
+#include "mozilla/dom/PSpeechSynthesisParent.h"
+#include "mozilla/dom/PSpeechSynthesisRequestParent.h"
+#include "nsSpeechTask.h"
+
+namespace mozilla {
+namespace dom {
+
+class ContentParent;
+class SpeechTaskParent;
+class SpeechSynthesisRequestParent;
+
+class SpeechSynthesisParent : public PSpeechSynthesisParent
+{
+  friend class ContentParent;
+  friend class SpeechSynthesisRequestParent;
+
+public:
+  void ActorDestroy(ActorDestroyReason aWhy) override;
+
+  bool RecvReadVoicesAndState(InfallibleTArray<RemoteVoice>* aVoices,
+                              InfallibleTArray<nsString>* aDefaults,
+                              bool* aIsSpeaking) override;
+
+protected:
+  SpeechSynthesisParent();
+  virtual ~SpeechSynthesisParent();
+  PSpeechSynthesisRequestParent* AllocPSpeechSynthesisRequestParent(const nsString& aText,
+                                                                    const nsString& aLang,
+                                                                    const nsString& aUri,
+                                                                    const float& aVolume,
+                                                                    const float& aRate,
+                                                                    const float& aPitch)
+                                                                    override;
+
+  bool DeallocPSpeechSynthesisRequestParent(PSpeechSynthesisRequestParent* aActor) override;
+
+  bool RecvPSpeechSynthesisRequestConstructor(PSpeechSynthesisRequestParent* aActor,
+                                              const nsString& aText,
+                                              const nsString& aLang,
+                                              const nsString& aUri,
+                                              const float& aVolume,
+                                              const float& aRate,
+                                              const float& aPitch) override;
+};
+
+class SpeechSynthesisRequestParent : public PSpeechSynthesisRequestParent
+{
+public:
+  explicit SpeechSynthesisRequestParent(SpeechTaskParent* aTask);
+  virtual ~SpeechSynthesisRequestParent();
+
+  RefPtr<SpeechTaskParent> mTask;
+
+protected:
+
+  void ActorDestroy(ActorDestroyReason aWhy) override;
+
+  bool RecvPause() override;
+
+  bool RecvResume() override;
+
+  bool RecvCancel() override;
+
+  bool RecvForceEnd() override;
+
+  bool RecvSetAudioOutputVolume(const float& aVolume) override;
+
+  bool Recv__delete__() override;
+};
+
+class SpeechTaskParent : public nsSpeechTask
+{
+  friend class SpeechSynthesisRequestParent;
+public:
+  SpeechTaskParent(float aVolume, const nsAString& aUtterance)
+    : nsSpeechTask(aVolume, aUtterance) {}
+
+  nsresult DispatchStartImpl(const nsAString& aUri);
+
+  nsresult DispatchEndImpl(float aElapsedTime, uint32_t aCharIndex);
+
+  nsresult DispatchPauseImpl(float aElapsedTime, uint32_t aCharIndex);
+
+  nsresult DispatchResumeImpl(float aElapsedTime, uint32_t aCharIndex);
+
+  nsresult DispatchErrorImpl(float aElapsedTime, uint32_t aCharIndex);
+
+  nsresult DispatchBoundaryImpl(const nsAString& aName,
+                                float aElapsedTime, uint32_t aCharIndex);
+
+  nsresult DispatchMarkImpl(const nsAString& aName,
+                            float aElapsedTime, uint32_t aCharIndex);
+
+private:
+  SpeechSynthesisRequestParent* mActor;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/synth/moz.build

@@ -0,0 +1,70 @@
+# vim: set filetype=python:
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+if CONFIG['MOZ_WEBSPEECH']:
+    MOCHITEST_MANIFESTS += [
+        'test/mochitest.ini',
+        'test/startup/mochitest.ini',
+    ]
+
+    XPIDL_MODULE = 'dom_webspeechsynth'
+
+    XPIDL_SOURCES += [
+        'nsISpeechService.idl',
+        'nsISynthVoiceRegistry.idl'
+    ]
+
+    EXPORTS.mozilla.dom += [
+        'ipc/SpeechSynthesisChild.h',
+        'ipc/SpeechSynthesisParent.h',
+        'nsSpeechTask.h',
+        'nsSynthVoiceRegistry.h',
+        'SpeechSynthesis.h',
+        'SpeechSynthesisUtterance.h',
+        'SpeechSynthesisVoice.h',
+    ]
+
+    UNIFIED_SOURCES += [
+        'ipc/SpeechSynthesisChild.cpp',
+        'ipc/SpeechSynthesisParent.cpp',
+        'nsSpeechTask.cpp',
+        'nsSynthVoiceRegistry.cpp',
+        'SpeechSynthesis.cpp',
+        'SpeechSynthesisUtterance.cpp',
+        'SpeechSynthesisVoice.cpp',
+    ]
+
+    if CONFIG['MOZ_WEBSPEECH_TEST_BACKEND']:
+        UNIFIED_SOURCES += [
+            'test/FakeSynthModule.cpp',
+            'test/nsFakeSynthServices.cpp'
+        ]
+
+    if CONFIG['MOZ_WIDGET_TOOLKIT'] == 'windows':
+        DIRS += ['windows']
+
+    if CONFIG['MOZ_WIDGET_TOOLKIT'] == 'cocoa':
+        DIRS += ['cocoa']
+
+    if CONFIG['MOZ_SYNTH_SPEECHD']:
+        DIRS += ['speechd']
+
+    if CONFIG['MOZ_SYNTH_PICO']:
+        DIRS += ['pico']
+
+IPDL_SOURCES += [
+    'ipc/PSpeechSynthesis.ipdl',
+    'ipc/PSpeechSynthesisRequest.ipdl',
+]
+
+include('/ipc/chromium/chromium-config.mozbuild')
+
+FINAL_LIBRARY = 'xul'
+LOCAL_INCLUDES += [
+    'ipc',
+]
+
+if CONFIG['GNU_CXX']:
+    CXXFLAGS += ['-Wno-error=shadow']

dom/media/webspeech/synth/nsISpeechService.idl

@@ -0,0 +1,173 @@
+/* -*- Mode: IDL; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this file,
+ * You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsISupports.idl"
+
+typedef unsigned short SpeechServiceType;
+
+/**
+ * A callback is implemented by the service. For direct audio services, it is
+ * required to implement these, although it could be helpful to use the
+ * cancel method for shutting down the speech resources.
+ */
+[scriptable, uuid(c576de0c-8a3d-4570-be7e-9876d3e5bed2)]
+interface nsISpeechTaskCallback : nsISupports
+{
+  /**
+   * The user or application has paused the speech.
+   */
+  void onPause();
+
+  /**
+   * The user or application has resumed the speech.
+   */
+  void onResume();
+
+  /**
+   * The user or application has canceled the speech.
+   */
+  void onCancel();
+
+  /**
+   * The user or application has changed the volume of this speech.
+   * This is only used on indirect audio service type.
+   */
+  void onVolumeChanged(in float aVolume);
+};
+
+
+/**
+ * A task is associated with a single utterance. It is provided by the browser
+ * to the service in the speak() method.
+ */
+[scriptable, builtinclass, uuid(ad59949c-2437-4b35-8eeb-d760caab75c5)]
+interface nsISpeechTask : nsISupports
+{
+  /**
+   * Prepare browser for speech.
+   *
+   * @param aCallback callback object for mid-speech operations.
+   * @param aChannels number of audio channels. Only required
+   *                    in direct audio services
+   * @param aRate     audio rate. Only required in direct audio services
+   */
+  [optional_argc] void setup(in nsISpeechTaskCallback aCallback,
+                               [optional] in uint32_t aChannels,
+                               [optional] in uint32_t aRate);
+
+  /**
+   * Send audio data to browser.
+   *
+   * @param aData     an Int16Array with PCM-16 audio data.
+   * @param aLandmarks an array of sample offset and landmark pairs.
+   *                     Used for emiting boundary and mark events.
+   */
+  [implicit_jscontext]
+  void sendAudio(in jsval aData, in jsval aLandmarks);
+
+  [noscript]
+  void sendAudioNative([array, size_is(aDataLen)] in short aData, in unsigned long aDataLen);
+
+  /**
+   * Dispatch start event.
+   */
+  void dispatchStart();
+
+  /**
+   * Dispatch end event.
+   *
+   * @param aElapsedTime time in seconds since speech has started.
+   * @param aCharIndex   offset of spoken characters.
+   */
+  void dispatchEnd(in float aElapsedTime, in unsigned long aCharIndex);
+
+  /**
+   * Dispatch pause event.
+   *
+   * @param aElapsedTime time in seconds since speech has started.
+   * @param aCharIndex   offset of spoken characters.
+   */
+  void dispatchPause(in float aElapsedTime, in unsigned long aCharIndex);
+
+  /**
+   * Dispatch resume event.
+   *
+   * @param aElapsedTime time in seconds since speech has started.
+   * @param aCharIndex   offset of spoken characters.
+   */
+  void dispatchResume(in float aElapsedTime, in unsigned long aCharIndex);
+
+  /**
+   * Dispatch error event.
+   *
+   * @param aElapsedTime time in seconds since speech has started.
+   * @param aCharIndex   offset of spoken characters.
+   */
+  void dispatchError(in float aElapsedTime, in unsigned long aCharIndex);
+
+  /**
+   * Dispatch boundary event.
+   *
+   * @param aName        name of boundary, 'word' or 'sentence'
+   * @param aElapsedTime time in seconds since speech has started.
+   * @param aCharIndex   offset of spoken characters.
+   */
+  void dispatchBoundary(in DOMString aName, in float aElapsedTime,
+                        in unsigned long aCharIndex);
+
+  /**
+   * Dispatch mark event.
+   *
+   * @param aName        mark identifier.
+   * @param aElapsedTime time in seconds since speech has started.
+   * @param aCharIndex   offset of spoken characters.
+   */
+  void dispatchMark(in DOMString aName, in float aElapsedTime, in unsigned long aCharIndex);
+};
+
+/**
+ * The main interface of a speech synthesis service.
+ *
+ * A service's speak method could be implemented in two ways:
+ *  1. Indirect audio - the service is responsible for outputting audio.
+ *    The service calls the nsISpeechTask.dispatch* methods directly. Starting
+ *    with dispatchStart() and ending with dispatchEnd or dispatchError().
+ *
+ *  2. Direct audio - the service provides us with PCM-16 data, and we output it.
+ *    The service does not call the dispatch task methods directly. Instead,
+ *    audio information is provided at setup(), and audio data is sent with
+ *    sendAudio(). The utterance is terminated with an empty sendAudio().
+ */
+[scriptable, uuid(9b7d59db-88ff-43d0-b6ee-9f63d042d08f)]
+interface nsISpeechService : nsISupports
+{
+  /**
+   * Speak the given text using the voice identified byu the given uri. See
+   * W3C Speech API spec for information about pitch and rate.
+   * https://dvcs.w3.org/hg/speech-api/raw-file/tip/speechapi.html#utterance-attributes
+   *
+   * @param aText   text to utter.
+   * @param aUri    unique voice identifier.
+   * @param aVolume volume to speak voice in. Only relevant for indirect audio.
+   * @param aRate   rate to speak voice in.
+   * @param aPitch  pitch to speak voice in.
+   * @param aTask  task instance for utterance, used for sending events or audio
+   *                 data back to browser.
+   */
+  void speak(in DOMString aText, in DOMString aUri,
+             in float aVolume, in float aRate, in float aPitch,
+             in nsISpeechTask aTask);
+
+  const SpeechServiceType SERVICETYPE_DIRECT_AUDIO = 1;
+  const SpeechServiceType SERVICETYPE_INDIRECT_AUDIO = 2;
+
+  readonly attribute SpeechServiceType serviceType;
+};
+
+%{C++
+// This is the service category speech services could use to start up as
+// a component.
+#define NS_SPEECH_SYNTH_STARTED "speech-synth-started"
+%}

dom/media/webspeech/synth/nsISynthVoiceRegistry.idl

@@ -0,0 +1,77 @@
+/* -*- Mode: IDL; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this file,
+ * You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsISupports.idl"
+
+interface nsISpeechService;
+
+[scriptable, builtinclass, uuid(5d7a0b38-77e5-4ee5-897c-ce5db9b85d44)]
+interface nsISynthVoiceRegistry : nsISupports
+{
+  /**
+   * Register a speech synthesis voice.
+   *
+   * @param aService          the service that provides this voice.
+   * @param aUri              a unique identifier for this voice.
+   * @param aName             human-readable name for this voice.
+   * @param aLang             a BCP 47 language tag.
+   * @param aLocalService     true if service does not require network.
+   * @param aQueuesUtterances true if voice only speaks one utterance at a time
+   */
+  void addVoice(in nsISpeechService aService, in DOMString aUri,
+                in DOMString aName, in DOMString aLang,
+                in boolean aLocalService, in boolean aQueuesUtterances);
+
+  /**
+   * Remove a speech synthesis voice.
+   *
+   * @param aService the service that was used to add the voice.
+   * @param aUri     a unique identifier of an existing voice.
+   */
+  void removeVoice(in nsISpeechService aService, in DOMString aUri);
+
+  /**
+   * Notify content of voice availability changes. This allows content
+   * to be notified of voice catalog changes in real time.
+   */
+  void notifyVoicesChanged();
+
+  /**
+   * Set a voice as default.
+   *
+   * @param aUri       a unique identifier of an existing voice.
+   * @param aIsDefault true if this voice should be toggled as default.
+   */
+  void setDefaultVoice(in DOMString aUri, in boolean aIsDefault);
+
+  readonly attribute uint32_t voiceCount;
+
+  AString getVoice(in uint32_t aIndex);
+
+  bool isDefaultVoice(in DOMString aUri);
+
+  bool isLocalVoice(in DOMString aUri);
+
+  AString getVoiceLang(in DOMString aUri);
+
+  AString getVoiceName(in DOMString aUri);
+};
+
+%{C++
+#define NS_SYNTHVOICEREGISTRY_CID                   \
+  { /* {7090524d-5574-4492-a77f-d8d558ced59d} */       \
+    0x7090524d,                                        \
+    0x5574,                                            \
+    0x4492,                                            \
+    { 0xa7, 0x7f, 0xd8, 0xd5, 0x58, 0xce, 0xd5, 0x9d } \
+  }
+
+#define NS_SYNTHVOICEREGISTRY_CONTRACTID \
+    "@mozilla.org/synth-voice-registry;1"
+
+#define NS_SYNTHVOICEREGISTRY_CLASSNAME \
+    "Speech Synthesis Voice Registry"
+
+%}

dom/media/webspeech/synth/nsSpeechTask.cpp

@@ -0,0 +1,783 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "AudioChannelAgent.h"
+#include "AudioChannelService.h"
+#include "AudioSegment.h"
+#include "MediaStreamListener.h"
+#include "nsSpeechTask.h"
+#include "nsSynthVoiceRegistry.h"
+#include "SharedBuffer.h"
+#include "SpeechSynthesis.h"
+
+// GetCurrentTime is defined in winbase.h as zero argument macro forwarding to
+// GetTickCount() and conflicts with nsSpeechTask::GetCurrentTime().
+#ifdef GetCurrentTime
+#undef GetCurrentTime
+#endif
+
+#undef LOG
+extern mozilla::LogModule* GetSpeechSynthLog();
+#define LOG(type, msg) MOZ_LOG(GetSpeechSynthLog(), type, msg)
+
+#define AUDIO_TRACK 1
+
+namespace mozilla {
+namespace dom {
+
+class SynthStreamListener : public MediaStreamListener
+{
+public:
+  explicit SynthStreamListener(nsSpeechTask* aSpeechTask,
+                               MediaStream* aStream) :
+    mSpeechTask(aSpeechTask),
+    mStream(aStream),
+    mStarted(false)
+  {
+  }
+
+  void DoNotifyStarted()
+  {
+    if (mSpeechTask) {
+      mSpeechTask->DispatchStartInner();
+    }
+  }
+
+  void DoNotifyFinished()
+  {
+    if (mSpeechTask) {
+      mSpeechTask->DispatchEndInner(mSpeechTask->GetCurrentTime(),
+                                    mSpeechTask->GetCurrentCharOffset());
+    }
+  }
+
+  void NotifyEvent(MediaStreamGraph* aGraph,
+                   MediaStreamGraphEvent event) override
+  {
+    switch (event) {
+      case MediaStreamGraphEvent::EVENT_FINISHED:
+        {
+          if (!mStarted) {
+            mStarted = true;
+            nsCOMPtr<nsIRunnable> startRunnable =
+              NewRunnableMethod(this, &SynthStreamListener::DoNotifyStarted);
+            aGraph->DispatchToMainThreadAfterStreamStateUpdate(startRunnable.forget());
+          }
+
+          nsCOMPtr<nsIRunnable> endRunnable =
+            NewRunnableMethod(this, &SynthStreamListener::DoNotifyFinished);
+          aGraph->DispatchToMainThreadAfterStreamStateUpdate(endRunnable.forget());
+        }
+        break;
+      case MediaStreamGraphEvent::EVENT_REMOVED:
+        mSpeechTask = nullptr;
+        // Dereference MediaStream to destroy safety
+        mStream = nullptr;
+        break;
+      default:
+        break;
+    }
+  }
+
+  void NotifyBlockingChanged(MediaStreamGraph* aGraph, Blocking aBlocked) override
+  {
+    if (aBlocked == MediaStreamListener::UNBLOCKED && !mStarted) {
+      mStarted = true;
+      nsCOMPtr<nsIRunnable> event =
+        NewRunnableMethod(this, &SynthStreamListener::DoNotifyStarted);
+      aGraph->DispatchToMainThreadAfterStreamStateUpdate(event.forget());
+    }
+  }
+
+private:
+  // Raw pointer; if we exist, the stream exists,
+  // and 'mSpeechTask' exclusively owns it and therefor exists as well.
+  nsSpeechTask* mSpeechTask;
+  // This is KungFuDeathGrip for MediaStream
+  RefPtr<MediaStream> mStream;
+
+  bool mStarted;
+};
+
+// nsSpeechTask
+
+NS_IMPL_CYCLE_COLLECTION(nsSpeechTask, mSpeechSynthesis, mUtterance, mCallback);
+
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(nsSpeechTask)
+  NS_INTERFACE_MAP_ENTRY(nsISpeechTask)
+  NS_INTERFACE_MAP_ENTRY(nsIAudioChannelAgentCallback)
+  NS_INTERFACE_MAP_ENTRY(nsISupportsWeakReference)
+  NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsISpeechTask)
+NS_INTERFACE_MAP_END
+
+NS_IMPL_CYCLE_COLLECTING_ADDREF(nsSpeechTask)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(nsSpeechTask)
+
+nsSpeechTask::nsSpeechTask(SpeechSynthesisUtterance* aUtterance)
+  : mUtterance(aUtterance)
+  , mInited(false)
+  , mPrePaused(false)
+  , mPreCanceled(false)
+  , mCallback(nullptr)
+  , mIndirectAudio(false)
+{
+  mText = aUtterance->mText;
+  mVolume = aUtterance->Volume();
+}
+
+nsSpeechTask::nsSpeechTask(float aVolume, const nsAString& aText)
+  : mUtterance(nullptr)
+  , mVolume(aVolume)
+  , mText(aText)
+  , mInited(false)
+  , mPrePaused(false)
+  , mPreCanceled(false)
+  , mCallback(nullptr)
+  , mIndirectAudio(false)
+{
+}
+
+nsSpeechTask::~nsSpeechTask()
+{
+  LOG(LogLevel::Debug, ("~nsSpeechTask"));
+  if (mStream) {
+    if (!mStream->IsDestroyed()) {
+      mStream->Destroy();
+    }
+
+    // This will finally destroyed by SynthStreamListener becasue
+    // MediaStream::Destroy() is async.
+    mStream = nullptr;
+  }
+
+  if (mPort) {
+    mPort->Destroy();
+    mPort = nullptr;
+  }
+}
+
+void
+nsSpeechTask::InitDirectAudio()
+{
+  mStream = MediaStreamGraph::GetInstance(MediaStreamGraph::AUDIO_THREAD_DRIVER,
+                                          AudioChannel::Normal)->
+    CreateSourceStream();
+  mIndirectAudio = false;
+  mInited = true;
+}
+
+void
+nsSpeechTask::InitIndirectAudio()
+{
+  mIndirectAudio = true;
+  mInited = true;
+}
+
+void
+nsSpeechTask::SetChosenVoiceURI(const nsAString& aUri)
+{
+  mChosenVoiceURI = aUri;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::Setup(nsISpeechTaskCallback* aCallback,
+                    uint32_t aChannels, uint32_t aRate, uint8_t argc)
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  LOG(LogLevel::Debug, ("nsSpeechTask::Setup"));
+
+  mCallback = aCallback;
+
+  if (mIndirectAudio) {
+    MOZ_ASSERT(!mStream);
+    if (argc > 0) {
+      NS_WARNING("Audio info arguments in Setup() are ignored for indirect audio services.");
+    }
+    return NS_OK;
+  }
+
+  // mStream is set up in Init() that should be called before this.
+  MOZ_ASSERT(mStream);
+
+  mStream->AddListener(new SynthStreamListener(this, mStream));
+
+  // XXX: Support more than one channel
+  if(NS_WARN_IF(!(aChannels == 1))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  mChannels = aChannels;
+
+  AudioSegment* segment = new AudioSegment();
+  mStream->AddAudioTrack(AUDIO_TRACK, aRate, 0, segment);
+  mStream->AddAudioOutput(this);
+  mStream->SetAudioOutputVolume(this, mVolume);
+
+  return NS_OK;
+}
+
+static RefPtr<mozilla::SharedBuffer>
+makeSamples(int16_t* aData, uint32_t aDataLen)
+{
+  RefPtr<mozilla::SharedBuffer> samples =
+    SharedBuffer::Create(aDataLen * sizeof(int16_t));
+  int16_t* frames = static_cast<int16_t*>(samples->Data());
+
+  for (uint32_t i = 0; i < aDataLen; i++) {
+    frames[i] = aData[i];
+  }
+
+  return samples;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::SendAudio(JS::Handle<JS::Value> aData, JS::Handle<JS::Value> aLandmarks,
+                        JSContext* aCx)
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  if(NS_WARN_IF(!(mStream))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+  if(NS_WARN_IF(mStream->IsDestroyed())) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+  if(NS_WARN_IF(!(mChannels))) {
+    return NS_ERROR_FAILURE;
+  }
+  if(NS_WARN_IF(!(aData.isObject()))) {
+    return NS_ERROR_INVALID_ARG;
+  }
+
+  if (mIndirectAudio) {
+    NS_WARNING("Can't call SendAudio from an indirect audio speech service.");
+    return NS_ERROR_FAILURE;
+  }
+
+  JS::Rooted<JSObject*> darray(aCx, &aData.toObject());
+  JSAutoCompartment ac(aCx, darray);
+
+  JS::Rooted<JSObject*> tsrc(aCx, nullptr);
+
+  // Allow either Int16Array or plain JS Array
+  if (JS_IsInt16Array(darray)) {
+    tsrc = darray;
+  } else {
+    bool isArray;
+    if (!JS_IsArrayObject(aCx, darray, &isArray)) {
+      return NS_ERROR_UNEXPECTED;
+    }
+    if (isArray) {
+      tsrc = JS_NewInt16ArrayFromArray(aCx, darray);
+    }
+  }
+
+  if (!tsrc) {
+    return NS_ERROR_DOM_TYPE_MISMATCH_ERR;
+  }
+
+  uint32_t dataLen = JS_GetTypedArrayLength(tsrc);
+  RefPtr<mozilla::SharedBuffer> samples;
+  {
+    JS::AutoCheckCannotGC nogc;
+    bool isShared;
+    int16_t* data = JS_GetInt16ArrayData(tsrc, &isShared, nogc);
+    if (isShared) {
+      // Must opt in to using shared data.
+      return NS_ERROR_DOM_TYPE_MISMATCH_ERR;
+    }
+    samples = makeSamples(data, dataLen);
+  }
+  SendAudioImpl(samples, dataLen);
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::SendAudioNative(int16_t* aData, uint32_t aDataLen)
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  if(NS_WARN_IF(!(mStream))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+  if(NS_WARN_IF(mStream->IsDestroyed())) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+  if(NS_WARN_IF(!(mChannels))) {
+    return NS_ERROR_FAILURE;
+  }
+
+  if (mIndirectAudio) {
+    NS_WARNING("Can't call SendAudio from an indirect audio speech service.");
+    return NS_ERROR_FAILURE;
+  }
+
+  RefPtr<mozilla::SharedBuffer> samples = makeSamples(aData, aDataLen);
+  SendAudioImpl(samples, aDataLen);
+
+  return NS_OK;
+}
+
+void
+nsSpeechTask::SendAudioImpl(RefPtr<mozilla::SharedBuffer>& aSamples, uint32_t aDataLen)
+{
+  if (aDataLen == 0) {
+    mStream->EndAllTrackAndFinish();
+    return;
+  }
+
+  AudioSegment segment;
+  AutoTArray<const int16_t*, 1> channelData;
+  channelData.AppendElement(static_cast<int16_t*>(aSamples->Data()));
+  segment.AppendFrames(aSamples.forget(), channelData, aDataLen,
+                       PRINCIPAL_HANDLE_NONE);
+  mStream->AppendToTrack(1, &segment);
+  mStream->AdvanceKnownTracksTime(STREAM_TIME_MAX);
+}
+
+NS_IMETHODIMP
+nsSpeechTask::DispatchStart()
+{
+  if (!mIndirectAudio) {
+    NS_WARNING("Can't call DispatchStart() from a direct audio speech service");
+    return NS_ERROR_FAILURE;
+  }
+
+  return DispatchStartInner();
+}
+
+nsresult
+nsSpeechTask::DispatchStartInner()
+{
+  nsSynthVoiceRegistry::GetInstance()->SetIsSpeaking(true);
+  return DispatchStartImpl();
+}
+
+nsresult
+nsSpeechTask::DispatchStartImpl()
+{
+  return DispatchStartImpl(mChosenVoiceURI);
+}
+
+nsresult
+nsSpeechTask::DispatchStartImpl(const nsAString& aUri)
+{
+  LOG(LogLevel::Debug, ("nsSpeechTask::DispatchStart"));
+
+  MOZ_ASSERT(mUtterance);
+  if(NS_WARN_IF(!(mUtterance->mState == SpeechSynthesisUtterance::STATE_PENDING))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  CreateAudioChannelAgent();
+
+  mUtterance->mState = SpeechSynthesisUtterance::STATE_SPEAKING;
+  mUtterance->mChosenVoiceURI = aUri;
+  mUtterance->DispatchSpeechSynthesisEvent(NS_LITERAL_STRING("start"), 0, 0,
+                                           EmptyString());
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::DispatchEnd(float aElapsedTime, uint32_t aCharIndex)
+{
+  if (!mIndirectAudio) {
+    NS_WARNING("Can't call DispatchEnd() from a direct audio speech service");
+    return NS_ERROR_FAILURE;
+  }
+
+  return DispatchEndInner(aElapsedTime, aCharIndex);
+}
+
+nsresult
+nsSpeechTask::DispatchEndInner(float aElapsedTime, uint32_t aCharIndex)
+{
+  if (!mPreCanceled) {
+    nsSynthVoiceRegistry::GetInstance()->SpeakNext();
+  }
+
+  return DispatchEndImpl(aElapsedTime, aCharIndex);
+}
+
+nsresult
+nsSpeechTask::DispatchEndImpl(float aElapsedTime, uint32_t aCharIndex)
+{
+  LOG(LogLevel::Debug, ("nsSpeechTask::DispatchEnd\n"));
+
+  DestroyAudioChannelAgent();
+
+  MOZ_ASSERT(mUtterance);
+  if(NS_WARN_IF(mUtterance->mState == SpeechSynthesisUtterance::STATE_ENDED)) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  // XXX: This should not be here, but it prevents a crash in MSG.
+  if (mStream) {
+    mStream->Destroy();
+  }
+
+  RefPtr<SpeechSynthesisUtterance> utterance = mUtterance;
+
+  if (mSpeechSynthesis) {
+    mSpeechSynthesis->OnEnd(this);
+  }
+
+  if (utterance->mState == SpeechSynthesisUtterance::STATE_PENDING) {
+    utterance->mState = SpeechSynthesisUtterance::STATE_NONE;
+  } else {
+    utterance->mState = SpeechSynthesisUtterance::STATE_ENDED;
+    utterance->DispatchSpeechSynthesisEvent(NS_LITERAL_STRING("end"),
+                                            aCharIndex, aElapsedTime,
+                                            EmptyString());
+  }
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::DispatchPause(float aElapsedTime, uint32_t aCharIndex)
+{
+  if (!mIndirectAudio) {
+    NS_WARNING("Can't call DispatchPause() from a direct audio speech service");
+    return NS_ERROR_FAILURE;
+  }
+
+  return DispatchPauseImpl(aElapsedTime, aCharIndex);
+}
+
+nsresult
+nsSpeechTask::DispatchPauseImpl(float aElapsedTime, uint32_t aCharIndex)
+{
+  LOG(LogLevel::Debug, ("nsSpeechTask::DispatchPause"));
+  MOZ_ASSERT(mUtterance);
+  if(NS_WARN_IF(mUtterance->mPaused)) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+  if(NS_WARN_IF(mUtterance->mState == SpeechSynthesisUtterance::STATE_ENDED)) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  mUtterance->mPaused = true;
+  if (mUtterance->mState == SpeechSynthesisUtterance::STATE_SPEAKING) {
+    mUtterance->DispatchSpeechSynthesisEvent(NS_LITERAL_STRING("pause"),
+                                             aCharIndex, aElapsedTime,
+                                             EmptyString());
+  }
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::DispatchResume(float aElapsedTime, uint32_t aCharIndex)
+{
+  if (!mIndirectAudio) {
+    NS_WARNING("Can't call DispatchResume() from a direct audio speech service");
+    return NS_ERROR_FAILURE;
+  }
+
+  return DispatchResumeImpl(aElapsedTime, aCharIndex);
+}
+
+nsresult
+nsSpeechTask::DispatchResumeImpl(float aElapsedTime, uint32_t aCharIndex)
+{
+  LOG(LogLevel::Debug, ("nsSpeechTask::DispatchResume"));
+  MOZ_ASSERT(mUtterance);
+  if(NS_WARN_IF(!(mUtterance->mPaused))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+  if(NS_WARN_IF(mUtterance->mState == SpeechSynthesisUtterance::STATE_ENDED)) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  mUtterance->mPaused = false;
+  if (mUtterance->mState == SpeechSynthesisUtterance::STATE_SPEAKING) {
+    mUtterance->DispatchSpeechSynthesisEvent(NS_LITERAL_STRING("resume"),
+                                             aCharIndex, aElapsedTime,
+                                             EmptyString());
+  }
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::DispatchError(float aElapsedTime, uint32_t aCharIndex)
+{
+  LOG(LogLevel::Debug, ("nsSpeechTask::DispatchError"));
+
+  if (!mIndirectAudio) {
+    NS_WARNING("Can't call DispatchError() from a direct audio speech service");
+    return NS_ERROR_FAILURE;
+  }
+
+  if (!mPreCanceled) {
+    nsSynthVoiceRegistry::GetInstance()->SpeakNext();
+  }
+
+  return DispatchErrorImpl(aElapsedTime, aCharIndex);
+}
+
+nsresult
+nsSpeechTask::DispatchErrorImpl(float aElapsedTime, uint32_t aCharIndex)
+{
+  MOZ_ASSERT(mUtterance);
+  if(NS_WARN_IF(mUtterance->mState == SpeechSynthesisUtterance::STATE_ENDED)) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  if (mSpeechSynthesis) {
+    mSpeechSynthesis->OnEnd(this);
+  }
+
+  mUtterance->mState = SpeechSynthesisUtterance::STATE_ENDED;
+  mUtterance->DispatchSpeechSynthesisEvent(NS_LITERAL_STRING("error"),
+                                           aCharIndex, aElapsedTime,
+                                           EmptyString());
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::DispatchBoundary(const nsAString& aName,
+                               float aElapsedTime, uint32_t aCharIndex)
+{
+  if (!mIndirectAudio) {
+    NS_WARNING("Can't call DispatchBoundary() from a direct audio speech service");
+    return NS_ERROR_FAILURE;
+  }
+
+  return DispatchBoundaryImpl(aName, aElapsedTime, aCharIndex);
+}
+
+nsresult
+nsSpeechTask::DispatchBoundaryImpl(const nsAString& aName,
+                                   float aElapsedTime, uint32_t aCharIndex)
+{
+  MOZ_ASSERT(mUtterance);
+  if(NS_WARN_IF(!(mUtterance->mState == SpeechSynthesisUtterance::STATE_SPEAKING))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  mUtterance->DispatchSpeechSynthesisEvent(NS_LITERAL_STRING("boundary"),
+                                           aCharIndex, aElapsedTime,
+                                           aName);
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::DispatchMark(const nsAString& aName,
+                           float aElapsedTime, uint32_t aCharIndex)
+{
+  if (!mIndirectAudio) {
+    NS_WARNING("Can't call DispatchMark() from a direct audio speech service");
+    return NS_ERROR_FAILURE;
+  }
+
+  return DispatchMarkImpl(aName, aElapsedTime, aCharIndex);
+}
+
+nsresult
+nsSpeechTask::DispatchMarkImpl(const nsAString& aName,
+                               float aElapsedTime, uint32_t aCharIndex)
+{
+  MOZ_ASSERT(mUtterance);
+  if(NS_WARN_IF(!(mUtterance->mState == SpeechSynthesisUtterance::STATE_SPEAKING))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  mUtterance->DispatchSpeechSynthesisEvent(NS_LITERAL_STRING("mark"),
+                                           aCharIndex, aElapsedTime,
+                                           aName);
+  return NS_OK;
+}
+
+void
+nsSpeechTask::Pause()
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  if (mCallback) {
+    DebugOnly<nsresult> rv = mCallback->OnPause();
+    NS_WARNING_ASSERTION(NS_SUCCEEDED(rv), "Unable to call onPause() callback");
+  }
+
+  if (mStream) {
+    mStream->Suspend();
+  }
+
+  if (!mInited) {
+    mPrePaused = true;
+  }
+
+  if (!mIndirectAudio) {
+    DispatchPauseImpl(GetCurrentTime(), GetCurrentCharOffset());
+  }
+}
+
+void
+nsSpeechTask::Resume()
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  if (mCallback) {
+    DebugOnly<nsresult> rv = mCallback->OnResume();
+    NS_WARNING_ASSERTION(NS_SUCCEEDED(rv),
+                         "Unable to call onResume() callback");
+  }
+
+  if (mStream) {
+    mStream->Resume();
+  }
+
+  if (mPrePaused) {
+    mPrePaused = false;
+    nsSynthVoiceRegistry::GetInstance()->ResumeQueue();
+  }
+
+  if (!mIndirectAudio) {
+    DispatchResumeImpl(GetCurrentTime(), GetCurrentCharOffset());
+  }
+}
+
+void
+nsSpeechTask::Cancel()
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  LOG(LogLevel::Debug, ("nsSpeechTask::Cancel"));
+
+  if (mCallback) {
+    DebugOnly<nsresult> rv = mCallback->OnCancel();
+    NS_WARNING_ASSERTION(NS_SUCCEEDED(rv),
+                         "Unable to call onCancel() callback");
+  }
+
+  if (mStream) {
+    mStream->Suspend();
+  }
+
+  if (!mInited) {
+    mPreCanceled = true;
+  }
+
+  if (!mIndirectAudio) {
+    DispatchEndInner(GetCurrentTime(), GetCurrentCharOffset());
+  }
+}
+
+void
+nsSpeechTask::ForceEnd()
+{
+  if (mStream) {
+    mStream->Suspend();
+  }
+
+  if (!mInited) {
+    mPreCanceled = true;
+  }
+
+  DispatchEndInner(GetCurrentTime(), GetCurrentCharOffset());
+}
+
+float
+nsSpeechTask::GetCurrentTime()
+{
+  return mStream ? (float)(mStream->GetCurrentTime() / 1000000.0) : 0;
+}
+
+uint32_t
+nsSpeechTask::GetCurrentCharOffset()
+{
+  return mStream && mStream->IsFinished() ? mText.Length() : 0;
+}
+
+void
+nsSpeechTask::SetSpeechSynthesis(SpeechSynthesis* aSpeechSynthesis)
+{
+  mSpeechSynthesis = aSpeechSynthesis;
+}
+
+void
+nsSpeechTask::CreateAudioChannelAgent()
+{
+  if (!mUtterance) {
+    return;
+  }
+
+  if (mAudioChannelAgent) {
+    mAudioChannelAgent->NotifyStoppedPlaying();
+  }
+
+  mAudioChannelAgent = new AudioChannelAgent();
+  mAudioChannelAgent->InitWithWeakCallback(mUtterance->GetOwner(),
+                                           static_cast<int32_t>(AudioChannelService::GetDefaultAudioChannel()),
+                                           this);
+
+  AudioPlaybackConfig config;
+  nsresult rv = mAudioChannelAgent->NotifyStartedPlaying(&config,
+                                                         AudioChannelService::AudibleState::eAudible);
+  if (NS_WARN_IF(NS_FAILED(rv))) {
+    return;
+  }
+
+  WindowVolumeChanged(config.mVolume, config.mMuted);
+  WindowSuspendChanged(config.mSuspend);
+}
+
+void
+nsSpeechTask::DestroyAudioChannelAgent()
+{
+  if (mAudioChannelAgent) {
+    mAudioChannelAgent->NotifyStoppedPlaying();
+    mAudioChannelAgent = nullptr;
+  }
+}
+
+NS_IMETHODIMP
+nsSpeechTask::WindowVolumeChanged(float aVolume, bool aMuted)
+{
+  SetAudioOutputVolume(aMuted ? 0.0 : mVolume * aVolume);
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::WindowSuspendChanged(nsSuspendedTypes aSuspend)
+{
+  if (!mUtterance) {
+    return NS_OK;
+  }
+
+  if (aSuspend == nsISuspendedTypes::NONE_SUSPENDED &&
+      mUtterance->mPaused) {
+    Resume();
+  } else if (aSuspend != nsISuspendedTypes::NONE_SUSPENDED &&
+             !mUtterance->mPaused) {
+    Pause();
+  }
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSpeechTask::WindowAudioCaptureChanged(bool aCapture)
+{
+  // This is not supported yet.
+  return NS_OK;
+}
+
+void
+nsSpeechTask::SetAudioOutputVolume(float aVolume)
+{
+  if (mStream && !mStream->IsDestroyed()) {
+    mStream->SetAudioOutputVolume(this, aVolume);
+  }
+  if (mIndirectAudio) {
+    mCallback->OnVolumeChanged(aVolume);
+  }
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/nsSpeechTask.h

@@ -0,0 +1,139 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_nsSpeechTask_h
+#define mozilla_dom_nsSpeechTask_h
+
+#include "MediaStreamGraph.h"
+#include "SpeechSynthesisUtterance.h"
+#include "nsIAudioChannelAgent.h"
+#include "nsISpeechService.h"
+
+namespace mozilla {
+
+class SharedBuffer;
+
+namespace dom {
+
+class SpeechSynthesisUtterance;
+class SpeechSynthesis;
+class SynthStreamListener;
+
+class nsSpeechTask : public nsISpeechTask
+                   , public nsIAudioChannelAgentCallback
+                   , public nsSupportsWeakReference
+{
+  friend class SynthStreamListener;
+
+public:
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_CLASS_AMBIGUOUS(nsSpeechTask, nsISpeechTask)
+
+  NS_DECL_NSISPEECHTASK
+  NS_DECL_NSIAUDIOCHANNELAGENTCALLBACK
+
+  explicit nsSpeechTask(SpeechSynthesisUtterance* aUtterance);
+  nsSpeechTask(float aVolume, const nsAString& aText);
+
+  virtual void Pause();
+
+  virtual void Resume();
+
+  virtual void Cancel();
+
+  virtual void ForceEnd();
+
+  float GetCurrentTime();
+
+  uint32_t GetCurrentCharOffset();
+
+  void SetSpeechSynthesis(SpeechSynthesis* aSpeechSynthesis);
+
+  void InitDirectAudio();
+  void InitIndirectAudio();
+
+  void SetChosenVoiceURI(const nsAString& aUri);
+
+  virtual void SetAudioOutputVolume(float aVolume);
+
+  bool IsPreCanceled()
+  {
+    return mPreCanceled;
+  };
+
+  bool IsPrePaused()
+  {
+    return mPrePaused;
+  }
+
+protected:
+  virtual ~nsSpeechTask();
+
+  nsresult DispatchStartImpl();
+
+  virtual nsresult DispatchStartImpl(const nsAString& aUri);
+
+  virtual nsresult DispatchEndImpl(float aElapsedTime, uint32_t aCharIndex);
+
+  virtual nsresult DispatchPauseImpl(float aElapsedTime, uint32_t aCharIndex);
+
+  virtual nsresult DispatchResumeImpl(float aElapsedTime, uint32_t aCharIndex);
+
+  virtual nsresult DispatchErrorImpl(float aElapsedTime, uint32_t aCharIndex);
+
+  virtual nsresult DispatchBoundaryImpl(const nsAString& aName,
+                                        float aElapsedTime,
+                                        uint32_t aCharIndex);
+
+  virtual nsresult DispatchMarkImpl(const nsAString& aName,
+                                    float aElapsedTime, uint32_t aCharIndex);
+
+  RefPtr<SpeechSynthesisUtterance> mUtterance;
+
+  float mVolume;
+
+  nsString mText;
+
+  bool mInited;
+
+  bool mPrePaused;
+
+  bool mPreCanceled;
+
+private:
+  void End();
+
+  void SendAudioImpl(RefPtr<mozilla::SharedBuffer>& aSamples, uint32_t aDataLen);
+
+  nsresult DispatchStartInner();
+
+  nsresult DispatchEndInner(float aElapsedTime, uint32_t aCharIndex);
+
+  void CreateAudioChannelAgent();
+
+  void DestroyAudioChannelAgent();
+
+  RefPtr<SourceMediaStream> mStream;
+
+  RefPtr<MediaInputPort> mPort;
+
+  nsCOMPtr<nsISpeechTaskCallback> mCallback;
+
+  nsCOMPtr<nsIAudioChannelAgent> mAudioChannelAgent;
+
+  uint32_t mChannels;
+
+  RefPtr<SpeechSynthesis> mSpeechSynthesis;
+
+  bool mIndirectAudio;
+
+  nsString mChosenVoiceURI;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/synth/nsSynthVoiceRegistry.cpp

@@ -0,0 +1,835 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsILocaleService.h"
+#include "nsISpeechService.h"
+#include "nsServiceManagerUtils.h"
+#include "nsCategoryManagerUtils.h"
+
+#include "MediaPrefs.h"
+#include "SpeechSynthesisUtterance.h"
+#include "SpeechSynthesisVoice.h"
+#include "nsSynthVoiceRegistry.h"
+#include "nsSpeechTask.h"
+#include "AudioChannelService.h"
+
+#include "nsString.h"
+#include "mozilla/StaticPtr.h"
+#include "mozilla/dom/ContentChild.h"
+#include "mozilla/dom/ContentParent.h"
+#include "mozilla/Unused.h"
+
+#include "SpeechSynthesisChild.h"
+#include "SpeechSynthesisParent.h"
+
+#undef LOG
+extern mozilla::LogModule* GetSpeechSynthLog();
+#define LOG(type, msg) MOZ_LOG(GetSpeechSynthLog(), type, msg)
+
+namespace {
+
+void
+GetAllSpeechSynthActors(InfallibleTArray<mozilla::dom::SpeechSynthesisParent*>& aActors)
+{
+  MOZ_ASSERT(NS_IsMainThread());
+  MOZ_ASSERT(aActors.IsEmpty());
+
+  AutoTArray<mozilla::dom::ContentParent*, 20> contentActors;
+  mozilla::dom::ContentParent::GetAll(contentActors);
+
+  for (uint32_t contentIndex = 0;
+       contentIndex < contentActors.Length();
+       ++contentIndex) {
+    MOZ_ASSERT(contentActors[contentIndex]);
+
+    AutoTArray<mozilla::dom::PSpeechSynthesisParent*, 5> speechsynthActors;
+    contentActors[contentIndex]->ManagedPSpeechSynthesisParent(speechsynthActors);
+
+    for (uint32_t speechsynthIndex = 0;
+         speechsynthIndex < speechsynthActors.Length();
+         ++speechsynthIndex) {
+      MOZ_ASSERT(speechsynthActors[speechsynthIndex]);
+
+      mozilla::dom::SpeechSynthesisParent* actor =
+        static_cast<mozilla::dom::SpeechSynthesisParent*>(speechsynthActors[speechsynthIndex]);
+      aActors.AppendElement(actor);
+    }
+  }
+}
+
+} // namespace
+
+namespace mozilla {
+namespace dom {
+
+// VoiceData
+
+class VoiceData final
+{
+private:
+  // Private destructor, to discourage deletion outside of Release():
+  ~VoiceData() {}
+
+public:
+  VoiceData(nsISpeechService* aService, const nsAString& aUri,
+            const nsAString& aName, const nsAString& aLang,
+            bool aIsLocal, bool aQueuesUtterances)
+    : mService(aService)
+    , mUri(aUri)
+    , mName(aName)
+    , mLang(aLang)
+    , mIsLocal(aIsLocal)
+    , mIsQueued(aQueuesUtterances) {}
+
+  NS_INLINE_DECL_REFCOUNTING(VoiceData)
+
+  nsCOMPtr<nsISpeechService> mService;
+
+  nsString mUri;
+
+  nsString mName;
+
+  nsString mLang;
+
+  bool mIsLocal;
+
+  bool mIsQueued;
+};
+
+// GlobalQueueItem
+
+class GlobalQueueItem final
+{
+private:
+  // Private destructor, to discourage deletion outside of Release():
+  ~GlobalQueueItem() {}
+
+public:
+  GlobalQueueItem(VoiceData* aVoice, nsSpeechTask* aTask, const nsAString& aText,
+                  const float& aVolume, const float& aRate, const float& aPitch)
+    : mVoice(aVoice)
+    , mTask(aTask)
+    , mText(aText)
+    , mVolume(aVolume)
+    , mRate(aRate)
+    , mPitch(aPitch) {}
+
+  NS_INLINE_DECL_REFCOUNTING(GlobalQueueItem)
+
+  RefPtr<VoiceData> mVoice;
+
+  RefPtr<nsSpeechTask> mTask;
+
+  nsString mText;
+
+  float mVolume;
+
+  float mRate;
+
+  float mPitch;
+
+  bool mIsLocal;
+};
+
+// nsSynthVoiceRegistry
+
+static StaticRefPtr<nsSynthVoiceRegistry> gSynthVoiceRegistry;
+
+NS_IMPL_ISUPPORTS(nsSynthVoiceRegistry, nsISynthVoiceRegistry)
+
+nsSynthVoiceRegistry::nsSynthVoiceRegistry()
+  : mSpeechSynthChild(nullptr)
+  , mUseGlobalQueue(false)
+  , mIsSpeaking(false)
+{
+  if (XRE_IsContentProcess()) {
+
+    mSpeechSynthChild = new SpeechSynthesisChild();
+    ContentChild::GetSingleton()->SendPSpeechSynthesisConstructor(mSpeechSynthChild);
+
+    InfallibleTArray<RemoteVoice> voices;
+    InfallibleTArray<nsString> defaults;
+    bool isSpeaking;
+
+    mSpeechSynthChild->SendReadVoicesAndState(&voices, &defaults, &isSpeaking);
+
+    for (uint32_t i = 0; i < voices.Length(); ++i) {
+      RemoteVoice voice = voices[i];
+      AddVoiceImpl(nullptr, voice.voiceURI(),
+                   voice.name(), voice.lang(),
+                   voice.localService(), voice.queued());
+    }
+
+    for (uint32_t i = 0; i < defaults.Length(); ++i) {
+      SetDefaultVoice(defaults[i], true);
+    }
+
+    mIsSpeaking = isSpeaking;
+  }
+}
+
+nsSynthVoiceRegistry::~nsSynthVoiceRegistry()
+{
+  LOG(LogLevel::Debug, ("~nsSynthVoiceRegistry"));
+
+  // mSpeechSynthChild's lifecycle is managed by the Content protocol.
+  mSpeechSynthChild = nullptr;
+
+  mUriVoiceMap.Clear();
+}
+
+nsSynthVoiceRegistry*
+nsSynthVoiceRegistry::GetInstance()
+{
+  MOZ_ASSERT(NS_IsMainThread());
+
+  if (!gSynthVoiceRegistry) {
+    gSynthVoiceRegistry = new nsSynthVoiceRegistry();
+    if (XRE_IsParentProcess()) {
+      // Start up all speech synth services.
+      NS_CreateServicesFromCategory(NS_SPEECH_SYNTH_STARTED, nullptr,
+        NS_SPEECH_SYNTH_STARTED);
+    }
+  }
+
+  return gSynthVoiceRegistry;
+}
+
+already_AddRefed<nsSynthVoiceRegistry>
+nsSynthVoiceRegistry::GetInstanceForService()
+{
+  RefPtr<nsSynthVoiceRegistry> registry = GetInstance();
+
+  return registry.forget();
+}
+
+void
+nsSynthVoiceRegistry::Shutdown()
+{
+  LOG(LogLevel::Debug, ("[%s] nsSynthVoiceRegistry::Shutdown()",
+                        (XRE_IsContentProcess()) ? "Content" : "Default"));
+  gSynthVoiceRegistry = nullptr;
+}
+
+void
+nsSynthVoiceRegistry::SendVoicesAndState(InfallibleTArray<RemoteVoice>* aVoices,
+                                         InfallibleTArray<nsString>* aDefaults,
+                                         bool* aIsSpeaking)
+{
+  for (uint32_t i=0; i < mVoices.Length(); ++i) {
+    RefPtr<VoiceData> voice = mVoices[i];
+
+    aVoices->AppendElement(RemoteVoice(voice->mUri, voice->mName, voice->mLang,
+                                       voice->mIsLocal, voice->mIsQueued));
+  }
+
+  for (uint32_t i=0; i < mDefaultVoices.Length(); ++i) {
+    aDefaults->AppendElement(mDefaultVoices[i]->mUri);
+  }
+
+  *aIsSpeaking = IsSpeaking();
+}
+
+void
+nsSynthVoiceRegistry::RecvRemoveVoice(const nsAString& aUri)
+{
+  // If we dont have a local instance of the registry yet, we will recieve current
+  // voices at contruction time.
+  if(!gSynthVoiceRegistry) {
+    return;
+  }
+
+  gSynthVoiceRegistry->RemoveVoice(nullptr, aUri);
+}
+
+void
+nsSynthVoiceRegistry::RecvAddVoice(const RemoteVoice& aVoice)
+{
+  // If we dont have a local instance of the registry yet, we will recieve current
+  // voices at contruction time.
+  if(!gSynthVoiceRegistry) {
+    return;
+  }
+
+  gSynthVoiceRegistry->AddVoiceImpl(nullptr, aVoice.voiceURI(),
+                                    aVoice.name(), aVoice.lang(),
+                                    aVoice.localService(), aVoice.queued());
+}
+
+void
+nsSynthVoiceRegistry::RecvSetDefaultVoice(const nsAString& aUri, bool aIsDefault)
+{
+  // If we dont have a local instance of the registry yet, we will recieve current
+  // voices at contruction time.
+  if(!gSynthVoiceRegistry) {
+    return;
+  }
+
+  gSynthVoiceRegistry->SetDefaultVoice(aUri, aIsDefault);
+}
+
+void
+nsSynthVoiceRegistry::RecvIsSpeakingChanged(bool aIsSpeaking)
+{
+  // If we dont have a local instance of the registry yet, we will get the
+  // speaking state on construction.
+  if(!gSynthVoiceRegistry) {
+    return;
+  }
+
+  gSynthVoiceRegistry->mIsSpeaking = aIsSpeaking;
+}
+
+void
+nsSynthVoiceRegistry::RecvNotifyVoicesChanged()
+{
+  // If we dont have a local instance of the registry yet, we don't care.
+  if(!gSynthVoiceRegistry) {
+    return;
+  }
+
+  gSynthVoiceRegistry->NotifyVoicesChanged();
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::AddVoice(nsISpeechService* aService,
+                               const nsAString& aUri,
+                               const nsAString& aName,
+                               const nsAString& aLang,
+                               bool aLocalService,
+                               bool aQueuesUtterances)
+{
+  LOG(LogLevel::Debug,
+      ("nsSynthVoiceRegistry::AddVoice uri='%s' name='%s' lang='%s' local=%s queued=%s",
+       NS_ConvertUTF16toUTF8(aUri).get(), NS_ConvertUTF16toUTF8(aName).get(),
+       NS_ConvertUTF16toUTF8(aLang).get(),
+       aLocalService ? "true" : "false",
+       aQueuesUtterances ? "true" : "false"));
+
+  if(NS_WARN_IF(XRE_IsContentProcess())) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  return AddVoiceImpl(aService, aUri, aName, aLang, aLocalService, aQueuesUtterances);
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::RemoveVoice(nsISpeechService* aService,
+                                  const nsAString& aUri)
+{
+  LOG(LogLevel::Debug,
+      ("nsSynthVoiceRegistry::RemoveVoice uri='%s' (%s)",
+       NS_ConvertUTF16toUTF8(aUri).get(),
+       (XRE_IsContentProcess()) ? "child" : "parent"));
+
+  bool found = false;
+  VoiceData* retval = mUriVoiceMap.GetWeak(aUri, &found);
+
+  if(NS_WARN_IF(!(found))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+  if(NS_WARN_IF(!(aService == retval->mService))) {
+    return NS_ERROR_INVALID_ARG;
+  }
+
+  mVoices.RemoveElement(retval);
+  mDefaultVoices.RemoveElement(retval);
+  mUriVoiceMap.Remove(aUri);
+
+  if (retval->mIsQueued && !MediaPrefs::WebSpeechForceGlobal()) {
+    // Check if this is the last queued voice, and disable the global queue if
+    // it is.
+    bool queued = false;
+    for (uint32_t i = 0; i < mVoices.Length(); i++) {
+      VoiceData* voice = mVoices[i];
+      if (voice->mIsQueued) {
+        queued = true;
+        break;
+      }
+    }
+    if (!queued) {
+      mUseGlobalQueue = false;
+    }
+  }
+
+  nsTArray<SpeechSynthesisParent*> ssplist;
+  GetAllSpeechSynthActors(ssplist);
+
+  for (uint32_t i = 0; i < ssplist.Length(); ++i)
+    Unused << ssplist[i]->SendVoiceRemoved(nsString(aUri));
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::NotifyVoicesChanged()
+{
+  if (XRE_IsParentProcess()) {
+    nsTArray<SpeechSynthesisParent*> ssplist;
+    GetAllSpeechSynthActors(ssplist);
+
+    for (uint32_t i = 0; i < ssplist.Length(); ++i)
+      Unused << ssplist[i]->SendNotifyVoicesChanged();
+  }
+
+  nsCOMPtr<nsIObserverService> obs = mozilla::services::GetObserverService();
+  if(NS_WARN_IF(!(obs))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  obs->NotifyObservers(nullptr, "synth-voices-changed", nullptr);
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::SetDefaultVoice(const nsAString& aUri,
+                                      bool aIsDefault)
+{
+  bool found = false;
+  VoiceData* retval = mUriVoiceMap.GetWeak(aUri, &found);
+  if(NS_WARN_IF(!(found))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  mDefaultVoices.RemoveElement(retval);
+
+  LOG(LogLevel::Debug, ("nsSynthVoiceRegistry::SetDefaultVoice %s %s",
+                     NS_ConvertUTF16toUTF8(aUri).get(),
+                     aIsDefault ? "true" : "false"));
+
+  if (aIsDefault) {
+    mDefaultVoices.AppendElement(retval);
+  }
+
+  if (XRE_IsParentProcess()) {
+    nsTArray<SpeechSynthesisParent*> ssplist;
+    GetAllSpeechSynthActors(ssplist);
+
+    for (uint32_t i = 0; i < ssplist.Length(); ++i) {
+      Unused << ssplist[i]->SendSetDefaultVoice(nsString(aUri), aIsDefault);
+    }
+  }
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::GetVoiceCount(uint32_t* aRetval)
+{
+  *aRetval = mVoices.Length();
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::GetVoice(uint32_t aIndex, nsAString& aRetval)
+{
+  if(NS_WARN_IF(!(aIndex < mVoices.Length()))) {
+    return NS_ERROR_INVALID_ARG;
+  }
+
+  aRetval = mVoices[aIndex]->mUri;
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::IsDefaultVoice(const nsAString& aUri, bool* aRetval)
+{
+  bool found;
+  VoiceData* voice = mUriVoiceMap.GetWeak(aUri, &found);
+  if(NS_WARN_IF(!(found))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  for (int32_t i = mDefaultVoices.Length(); i > 0; ) {
+    VoiceData* defaultVoice = mDefaultVoices[--i];
+
+    if (voice->mLang.Equals(defaultVoice->mLang)) {
+      *aRetval = voice == defaultVoice;
+      return NS_OK;
+    }
+  }
+
+  *aRetval = false;
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::IsLocalVoice(const nsAString& aUri, bool* aRetval)
+{
+  bool found;
+  VoiceData* voice = mUriVoiceMap.GetWeak(aUri, &found);
+  if(NS_WARN_IF(!(found))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  *aRetval = voice->mIsLocal;
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::GetVoiceLang(const nsAString& aUri, nsAString& aRetval)
+{
+  bool found;
+  VoiceData* voice = mUriVoiceMap.GetWeak(aUri, &found);
+  if(NS_WARN_IF(!(found))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  aRetval = voice->mLang;
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+nsSynthVoiceRegistry::GetVoiceName(const nsAString& aUri, nsAString& aRetval)
+{
+  bool found;
+  VoiceData* voice = mUriVoiceMap.GetWeak(aUri, &found);
+  if(NS_WARN_IF(!(found))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  aRetval = voice->mName;
+  return NS_OK;
+}
+
+nsresult
+nsSynthVoiceRegistry::AddVoiceImpl(nsISpeechService* aService,
+                                   const nsAString& aUri,
+                                   const nsAString& aName,
+                                   const nsAString& aLang,
+                                   bool aLocalService,
+                                   bool aQueuesUtterances)
+{
+  bool found = false;
+  mUriVoiceMap.GetWeak(aUri, &found);
+  if(NS_WARN_IF(found)) {
+    return NS_ERROR_INVALID_ARG;
+  }
+
+  RefPtr<VoiceData> voice = new VoiceData(aService, aUri, aName, aLang,
+                                            aLocalService, aQueuesUtterances);
+
+  mVoices.AppendElement(voice);
+  mUriVoiceMap.Put(aUri, voice);
+  mUseGlobalQueue |= aQueuesUtterances;
+
+  nsTArray<SpeechSynthesisParent*> ssplist;
+  GetAllSpeechSynthActors(ssplist);
+
+  if (!ssplist.IsEmpty()) {
+    mozilla::dom::RemoteVoice ssvoice(nsString(aUri),
+                                      nsString(aName),
+                                      nsString(aLang),
+                                      aLocalService,
+                                      aQueuesUtterances);
+
+    for (uint32_t i = 0; i < ssplist.Length(); ++i) {
+      Unused << ssplist[i]->SendVoiceAdded(ssvoice);
+    }
+  }
+
+  return NS_OK;
+}
+
+bool
+nsSynthVoiceRegistry::FindVoiceByLang(const nsAString& aLang,
+                                      VoiceData** aRetval)
+{
+  nsAString::const_iterator dashPos, start, end;
+  aLang.BeginReading(start);
+  aLang.EndReading(end);
+
+  while (true) {
+    nsAutoString langPrefix(Substring(start, end));
+
+    for (int32_t i = mDefaultVoices.Length(); i > 0; ) {
+      VoiceData* voice = mDefaultVoices[--i];
+
+      if (StringBeginsWith(voice->mLang, langPrefix)) {
+        *aRetval = voice;
+        return true;
+      }
+    }
+
+    for (int32_t i = mVoices.Length(); i > 0; ) {
+      VoiceData* voice = mVoices[--i];
+
+      if (StringBeginsWith(voice->mLang, langPrefix)) {
+        *aRetval = voice;
+        return true;
+      }
+    }
+
+    dashPos = end;
+    end = start;
+
+    if (!RFindInReadable(NS_LITERAL_STRING("-"), end, dashPos)) {
+      break;
+    }
+  }
+
+  return false;
+}
+
+VoiceData*
+nsSynthVoiceRegistry::FindBestMatch(const nsAString& aUri,
+                                    const nsAString& aLang)
+{
+  if (mVoices.IsEmpty()) {
+    return nullptr;
+  }
+
+  bool found = false;
+  VoiceData* retval = mUriVoiceMap.GetWeak(aUri, &found);
+
+  if (found) {
+    LOG(LogLevel::Debug, ("nsSynthVoiceRegistry::FindBestMatch - Matched URI"));
+    return retval;
+  }
+
+  // Try finding a match for given voice.
+  if (!aLang.IsVoid() && !aLang.IsEmpty()) {
+    if (FindVoiceByLang(aLang, &retval)) {
+      LOG(LogLevel::Debug,
+          ("nsSynthVoiceRegistry::FindBestMatch - Matched language (%s ~= %s)",
+           NS_ConvertUTF16toUTF8(aLang).get(),
+           NS_ConvertUTF16toUTF8(retval->mLang).get()));
+
+      return retval;
+    }
+  }
+
+  // Try UI language.
+  nsresult rv;
+  nsCOMPtr<nsILocaleService> localeService = do_GetService(NS_LOCALESERVICE_CONTRACTID, &rv);
+  if (NS_WARN_IF(NS_FAILED(rv))) {
+    return nullptr;
+  }
+
+  nsAutoString uiLang;
+  rv = localeService->GetLocaleComponentForUserAgent(uiLang);
+  if (NS_WARN_IF(NS_FAILED(rv))) {
+    return nullptr;
+  }
+
+  if (FindVoiceByLang(uiLang, &retval)) {
+    LOG(LogLevel::Debug,
+        ("nsSynthVoiceRegistry::FindBestMatch - Matched UI language (%s ~= %s)",
+         NS_ConvertUTF16toUTF8(uiLang).get(),
+         NS_ConvertUTF16toUTF8(retval->mLang).get()));
+
+    return retval;
+  }
+
+  // Try en-US, the language of locale "C"
+  if (FindVoiceByLang(NS_LITERAL_STRING("en-US"), &retval)) {
+    LOG(LogLevel::Debug,
+        ("nsSynthVoiceRegistry::FindBestMatch - Matched C locale language (en-US ~= %s)",
+         NS_ConvertUTF16toUTF8(retval->mLang).get()));
+
+    return retval;
+  }
+
+  // The top default voice is better than nothing...
+  if (!mDefaultVoices.IsEmpty()) {
+    return mDefaultVoices.LastElement();
+  }
+
+  return nullptr;
+}
+
+already_AddRefed<nsSpeechTask>
+nsSynthVoiceRegistry::SpeakUtterance(SpeechSynthesisUtterance& aUtterance,
+                                     const nsAString& aDocLang)
+{
+  nsString lang = nsString(aUtterance.mLang.IsEmpty() ? aDocLang : aUtterance.mLang);
+  nsAutoString uri;
+
+  if (aUtterance.mVoice) {
+    aUtterance.mVoice->GetVoiceURI(uri);
+  }
+
+  // Get current audio volume to apply speech call
+  float volume = aUtterance.Volume();
+  RefPtr<AudioChannelService> service = AudioChannelService::GetOrCreate();
+  if (service) {
+    if (nsCOMPtr<nsPIDOMWindowInner> topWindow = aUtterance.GetOwner()) {
+      // TODO : use audio channel agent, open new bug to fix it.
+      uint32_t channel = static_cast<uint32_t>(AudioChannelService::GetDefaultAudioChannel());
+      AudioPlaybackConfig config = service->GetMediaConfig(topWindow->GetOuterWindow(),
+                                                           channel);
+      volume = config.mMuted ? 0.0f : config.mVolume * volume;
+    }
+  }
+
+  RefPtr<nsSpeechTask> task;
+  if (XRE_IsContentProcess()) {
+    task = new SpeechTaskChild(&aUtterance);
+    SpeechSynthesisRequestChild* actor =
+      new SpeechSynthesisRequestChild(static_cast<SpeechTaskChild*>(task.get()));
+    mSpeechSynthChild->SendPSpeechSynthesisRequestConstructor(actor,
+                                                              aUtterance.mText,
+                                                              lang,
+                                                              uri,
+                                                              volume,
+                                                              aUtterance.Rate(),
+                                                              aUtterance.Pitch());
+  } else {
+    task = new nsSpeechTask(&aUtterance);
+    Speak(aUtterance.mText, lang, uri,
+          volume, aUtterance.Rate(), aUtterance.Pitch(), task);
+  }
+
+  return task.forget();
+}
+
+void
+nsSynthVoiceRegistry::Speak(const nsAString& aText,
+                            const nsAString& aLang,
+                            const nsAString& aUri,
+                            const float& aVolume,
+                            const float& aRate,
+                            const float& aPitch,
+                            nsSpeechTask* aTask)
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  VoiceData* voice = FindBestMatch(aUri, aLang);
+
+  if (!voice) {
+    NS_WARNING("No voices found.");
+    aTask->DispatchError(0, 0);
+    return;
+  }
+
+  aTask->SetChosenVoiceURI(voice->mUri);
+
+  if (mUseGlobalQueue || MediaPrefs::WebSpeechForceGlobal()) {
+    LOG(LogLevel::Debug,
+        ("nsSynthVoiceRegistry::Speak queueing text='%s' lang='%s' uri='%s' rate=%f pitch=%f",
+         NS_ConvertUTF16toUTF8(aText).get(), NS_ConvertUTF16toUTF8(aLang).get(),
+         NS_ConvertUTF16toUTF8(aUri).get(), aRate, aPitch));
+    RefPtr<GlobalQueueItem> item = new GlobalQueueItem(voice, aTask, aText,
+                                                         aVolume, aRate, aPitch);
+    mGlobalQueue.AppendElement(item);
+
+    if (mGlobalQueue.Length() == 1) {
+      SpeakImpl(item->mVoice, item->mTask, item->mText, item->mVolume, item->mRate,
+                item->mPitch);
+    }
+  } else {
+    SpeakImpl(voice, aTask, aText, aVolume, aRate, aPitch);
+  }
+}
+
+void
+nsSynthVoiceRegistry::SpeakNext()
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  LOG(LogLevel::Debug,
+      ("nsSynthVoiceRegistry::SpeakNext %d", mGlobalQueue.IsEmpty()));
+
+  SetIsSpeaking(false);
+
+  if (mGlobalQueue.IsEmpty()) {
+    return;
+  }
+
+  mGlobalQueue.RemoveElementAt(0);
+
+  while (!mGlobalQueue.IsEmpty()) {
+    RefPtr<GlobalQueueItem> item = mGlobalQueue.ElementAt(0);
+    if (item->mTask->IsPreCanceled()) {
+      mGlobalQueue.RemoveElementAt(0);
+      continue;
+    }
+    if (!item->mTask->IsPrePaused()) {
+      SpeakImpl(item->mVoice, item->mTask, item->mText, item->mVolume,
+                item->mRate, item->mPitch);
+    }
+    break;
+  }
+}
+
+void
+nsSynthVoiceRegistry::ResumeQueue()
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+  LOG(LogLevel::Debug,
+      ("nsSynthVoiceRegistry::ResumeQueue %d", mGlobalQueue.IsEmpty()));
+
+  if (mGlobalQueue.IsEmpty()) {
+    return;
+  }
+
+  RefPtr<GlobalQueueItem> item = mGlobalQueue.ElementAt(0);
+  if (!item->mTask->IsPrePaused()) {
+    SpeakImpl(item->mVoice, item->mTask, item->mText, item->mVolume,
+              item->mRate, item->mPitch);
+  }
+}
+
+bool
+nsSynthVoiceRegistry::IsSpeaking()
+{
+  return mIsSpeaking;
+}
+
+void
+nsSynthVoiceRegistry::SetIsSpeaking(bool aIsSpeaking)
+{
+  MOZ_ASSERT(XRE_IsParentProcess());
+
+  // Only set to 'true' if global queue is enabled.
+  mIsSpeaking =
+    aIsSpeaking && (mUseGlobalQueue || MediaPrefs::WebSpeechForceGlobal());
+
+  nsTArray<SpeechSynthesisParent*> ssplist;
+  GetAllSpeechSynthActors(ssplist);
+  for (uint32_t i = 0; i < ssplist.Length(); ++i) {
+    Unused << ssplist[i]->SendIsSpeakingChanged(aIsSpeaking);
+  }
+}
+
+void
+nsSynthVoiceRegistry::SpeakImpl(VoiceData* aVoice,
+                                nsSpeechTask* aTask,
+                                const nsAString& aText,
+                                const float& aVolume,
+                                const float& aRate,
+                                const float& aPitch)
+{
+  LOG(LogLevel::Debug,
+      ("nsSynthVoiceRegistry::SpeakImpl queueing text='%s' uri='%s' rate=%f pitch=%f",
+       NS_ConvertUTF16toUTF8(aText).get(), NS_ConvertUTF16toUTF8(aVoice->mUri).get(),
+       aRate, aPitch));
+
+  SpeechServiceType serviceType;
+
+  DebugOnly<nsresult> rv = aVoice->mService->GetServiceType(&serviceType);
+  NS_WARNING_ASSERTION(NS_SUCCEEDED(rv), "Failed to get speech service type");
+
+  if (serviceType == nsISpeechService::SERVICETYPE_INDIRECT_AUDIO) {
+    aTask->InitIndirectAudio();
+  } else {
+    aTask->InitDirectAudio();
+  }
+
+  if (NS_FAILED(aVoice->mService->Speak(aText, aVoice->mUri, aVolume, aRate,
+                                        aPitch, aTask))) {
+    if (serviceType == nsISpeechService::SERVICETYPE_INDIRECT_AUDIO) {
+      aTask->DispatchError(0, 0);
+    }
+    // XXX When using direct audio, no way to dispatch error
+  }
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/nsSynthVoiceRegistry.h

@@ -0,0 +1,109 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_nsSynthVoiceRegistry_h
+#define mozilla_dom_nsSynthVoiceRegistry_h
+
+#include "nsISynthVoiceRegistry.h"
+#include "nsRefPtrHashtable.h"
+#include "nsTArray.h"
+#include "MediaStreamGraph.h"
+
+class nsISpeechService;
+
+namespace mozilla {
+namespace dom {
+
+class RemoteVoice;
+class SpeechSynthesisUtterance;
+class SpeechSynthesisChild;
+class nsSpeechTask;
+class VoiceData;
+class GlobalQueueItem;
+
+class nsSynthVoiceRegistry final : public nsISynthVoiceRegistry
+{
+public:
+  NS_DECL_ISUPPORTS
+  NS_DECL_NSISYNTHVOICEREGISTRY
+
+  nsSynthVoiceRegistry();
+
+  already_AddRefed<nsSpeechTask> SpeakUtterance(SpeechSynthesisUtterance& aUtterance,
+                                                const nsAString& aDocLang);
+
+  void Speak(const nsAString& aText, const nsAString& aLang,
+             const nsAString& aUri, const float& aVolume,  const float& aRate,
+             const float& aPitch, nsSpeechTask* aTask);
+
+  void SendVoicesAndState(InfallibleTArray<RemoteVoice>* aVoices,
+                          InfallibleTArray<nsString>* aDefaults,
+                          bool* aIsSpeaking);
+
+  void SpeakNext();
+
+  void ResumeQueue();
+
+  bool IsSpeaking();
+
+  void SetIsSpeaking(bool aIsSpeaking);
+
+  static nsSynthVoiceRegistry* GetInstance();
+
+  static already_AddRefed<nsSynthVoiceRegistry> GetInstanceForService();
+
+  static void RecvRemoveVoice(const nsAString& aUri);
+
+  static void RecvAddVoice(const RemoteVoice& aVoice);
+
+  static void RecvSetDefaultVoice(const nsAString& aUri, bool aIsDefault);
+
+  static void RecvIsSpeakingChanged(bool aIsSpeaking);
+
+  static void RecvNotifyVoicesChanged();
+
+  static void Shutdown();
+
+private:
+  virtual ~nsSynthVoiceRegistry();
+
+  VoiceData* FindBestMatch(const nsAString& aUri, const nsAString& lang);
+
+  bool FindVoiceByLang(const nsAString& aLang, VoiceData** aRetval);
+
+  nsresult AddVoiceImpl(nsISpeechService* aService,
+                        const nsAString& aUri,
+                        const nsAString& aName,
+                        const nsAString& aLang,
+                        bool aLocalService,
+                        bool aQueuesUtterances);
+
+  void SpeakImpl(VoiceData* aVoice,
+                 nsSpeechTask* aTask,
+                 const nsAString& aText,
+                 const float& aVolume,
+                 const float& aRate,
+                 const float& aPitch);
+
+  nsTArray<RefPtr<VoiceData>> mVoices;
+
+  nsTArray<RefPtr<VoiceData>> mDefaultVoices;
+
+  nsRefPtrHashtable<nsStringHashKey, VoiceData> mUriVoiceMap;
+
+  SpeechSynthesisChild* mSpeechSynthChild;
+
+  bool mUseGlobalQueue;
+
+  nsTArray<RefPtr<GlobalQueueItem>> mGlobalQueue;
+
+  bool mIsSpeaking;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/synth/pico/PicoModule.cpp

@@ -0,0 +1,58 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "mozilla/ModuleUtils.h"
+#include "nsIClassInfoImpl.h"
+
+#ifdef MOZ_WEBRTC
+
+#include "nsPicoService.h"
+
+using namespace mozilla::dom;
+
+#define PICOSERVICE_CID \
+  {0x346c4fc8, 0x12fe, 0x459c, {0x81, 0x19, 0x9a, 0xa7, 0x73, 0x37, 0x7f, 0xf4}}
+
+#define PICOSERVICE_CONTRACTID "@mozilla.org/synthpico;1"
+
+// Defines nsPicoServiceConstructor
+NS_GENERIC_FACTORY_SINGLETON_CONSTRUCTOR(nsPicoService,
+                                         nsPicoService::GetInstanceForService)
+
+// Defines kPICOSERVICE_CID
+NS_DEFINE_NAMED_CID(PICOSERVICE_CID);
+
+static const mozilla::Module::CIDEntry kCIDs[] = {
+  { &kPICOSERVICE_CID, true, nullptr, nsPicoServiceConstructor },
+  { nullptr }
+};
+
+static const mozilla::Module::ContractIDEntry kContracts[] = {
+  { PICOSERVICE_CONTRACTID, &kPICOSERVICE_CID },
+  { nullptr }
+};
+
+static const mozilla::Module::CategoryEntry kCategories[] = {
+  { "profile-after-change", "Pico Speech Synth", PICOSERVICE_CONTRACTID },
+  { nullptr }
+};
+
+static void
+UnloadPicoModule()
+{
+  nsPicoService::Shutdown();
+}
+
+static const mozilla::Module kModule = {
+  mozilla::Module::kVersion,
+  kCIDs,
+  kContracts,
+  kCategories,
+  nullptr,
+  nullptr,
+  UnloadPicoModule
+};
+
+NSMODULE_DEFN(synthpico) = &kModule;
+#endif

dom/media/webspeech/synth/pico/moz.build

@@ -0,0 +1,13 @@
+# -*- Mode: python; indent-tabs-mode: nil; tab-width: 40 -*-
+# vim: set filetype=python:
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+UNIFIED_SOURCES += [
+    'nsPicoService.cpp',
+    'PicoModule.cpp'
+]
+include('/ipc/chromium/chromium-config.mozbuild')
+
+FINAL_LIBRARY = 'xul'

dom/media/webspeech/synth/pico/nsPicoService.cpp

@@ -0,0 +1,761 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsISupports.h"
+#include "nsPicoService.h"
+#include "nsPrintfCString.h"
+#include "nsIWeakReferenceUtils.h"
+#include "SharedBuffer.h"
+#include "nsISimpleEnumerator.h"
+
+#include "mozilla/dom/nsSynthVoiceRegistry.h"
+#include "mozilla/dom/nsSpeechTask.h"
+
+#include "nsIFile.h"
+#include "nsThreadUtils.h"
+#include "prenv.h"
+#include "mozilla/Preferences.h"
+#include "mozilla/DebugOnly.h"
+#include <dlfcn.h>
+
+// Pico API constants
+
+// Size of memory allocated for pico engine and voice resources.
+// We only have one voice and its resources loaded at once, so this
+// should always be enough.
+#define PICO_MEM_SIZE 2500000
+
+// Max length of returned strings. Pico will never return longer strings,
+// so this amount should be good enough for preallocating.
+#define PICO_RETSTRINGSIZE 200
+
+// Max amount we want from a single call of pico_getData
+#define PICO_MAX_CHUNK_SIZE 128
+
+// Arbitrary name for loaded voice, it doesn't mean anything outside of Pico
+#define PICO_VOICE_NAME "pico"
+
+// Return status from pico_getData meaning there is more data in the pipeline
+// to get from more calls to pico_getData
+#define PICO_STEP_BUSY 201
+
+// For performing a "soft" reset between utterances. This is used when one
+// utterance is interrupted by a new one.
+#define PICO_RESET_SOFT 0x10
+
+// Currently, Pico only provides mono output.
+#define PICO_CHANNELS_NUM 1
+
+// Pico's sample rate is always 16000
+#define PICO_SAMPLE_RATE 16000
+
+// The path to the language files in Gonk
+#define GONK_PICO_LANG_PATH "/system/tts/lang_pico"
+
+namespace mozilla {
+namespace dom {
+
+StaticRefPtr<nsPicoService> nsPicoService::sSingleton;
+
+class PicoApi
+{
+public:
+
+  PicoApi() : mInitialized(false) {}
+
+  bool Init()
+  {
+    if (mInitialized) {
+      return true;
+    }
+
+    void* handle = dlopen("libttspico.so", RTLD_LAZY);
+
+    if (!handle) {
+      NS_WARNING("Failed to open libttspico.so, pico cannot run");
+      return false;
+    }
+
+    pico_initialize =
+      (pico_Status (*)(void*, uint32_t, pico_System*))dlsym(
+        handle, "pico_initialize");
+
+    pico_terminate =
+      (pico_Status (*)(pico_System*))dlsym(handle, "pico_terminate");
+
+    pico_getSystemStatusMessage =
+      (pico_Status (*)(pico_System, pico_Status, pico_Retstring))dlsym(
+        handle, "pico_getSystemStatusMessage");;
+
+    pico_loadResource =
+      (pico_Status (*)(pico_System, const char*, pico_Resource*))dlsym(
+        handle, "pico_loadResource");
+
+    pico_unloadResource =
+      (pico_Status (*)(pico_System, pico_Resource*))dlsym(
+        handle, "pico_unloadResource");
+
+    pico_getResourceName =
+      (pico_Status (*)(pico_System, pico_Resource, pico_Retstring))dlsym(
+        handle, "pico_getResourceName");
+
+    pico_createVoiceDefinition =
+      (pico_Status (*)(pico_System, const char*))dlsym(
+        handle, "pico_createVoiceDefinition");
+
+    pico_addResourceToVoiceDefinition =
+      (pico_Status (*)(pico_System, const char*, const char*))dlsym(
+        handle, "pico_addResourceToVoiceDefinition");
+
+    pico_releaseVoiceDefinition =
+      (pico_Status (*)(pico_System, const char*))dlsym(
+        handle, "pico_releaseVoiceDefinition");
+
+    pico_newEngine =
+      (pico_Status (*)(pico_System, const char*, pico_Engine*))dlsym(
+        handle, "pico_newEngine");
+
+    pico_disposeEngine =
+      (pico_Status (*)(pico_System, pico_Engine*))dlsym(
+        handle, "pico_disposeEngine");
+
+    pico_resetEngine =
+      (pico_Status (*)(pico_Engine, int32_t))dlsym(handle, "pico_resetEngine");
+
+    pico_putTextUtf8 =
+      (pico_Status (*)(pico_Engine, const char*, const int16_t, int16_t*))dlsym(
+        handle, "pico_putTextUtf8");
+
+    pico_getData =
+      (pico_Status (*)(pico_Engine, void*, int16_t, int16_t*, int16_t*))dlsym(
+        handle, "pico_getData");
+
+    mInitialized = true;
+    return true;
+  }
+
+  typedef signed int pico_Status;
+  typedef char pico_Retstring[PICO_RETSTRINGSIZE];
+
+  pico_Status (* pico_initialize)(void*, uint32_t, pico_System*);
+  pico_Status (* pico_terminate)(pico_System*);
+  pico_Status (* pico_getSystemStatusMessage)(
+    pico_System, pico_Status, pico_Retstring);
+
+  pico_Status (* pico_loadResource)(pico_System, const char*, pico_Resource*);
+  pico_Status (* pico_unloadResource)(pico_System, pico_Resource*);
+  pico_Status (* pico_getResourceName)(
+    pico_System, pico_Resource, pico_Retstring);
+  pico_Status (* pico_createVoiceDefinition)(pico_System, const char*);
+  pico_Status (* pico_addResourceToVoiceDefinition)(
+    pico_System, const char*, const char*);
+  pico_Status (* pico_releaseVoiceDefinition)(pico_System, const char*);
+  pico_Status (* pico_newEngine)(pico_System, const char*, pico_Engine*);
+  pico_Status (* pico_disposeEngine)(pico_System, pico_Engine*);
+
+  pico_Status (* pico_resetEngine)(pico_Engine, int32_t);
+  pico_Status (* pico_putTextUtf8)(
+    pico_Engine, const char*, const int16_t, int16_t*);
+  pico_Status (* pico_getData)(
+    pico_Engine, void*, const int16_t, int16_t*, int16_t*);
+
+private:
+
+  bool mInitialized;
+
+} sPicoApi;
+
+#define PICO_ENSURE_SUCCESS_VOID(_funcName, _status)                      \
+  if (_status < 0) {                                                      \
+    PicoApi::pico_Retstring message;                                      \
+    sPicoApi.pico_getSystemStatusMessage(                                 \
+      nsPicoService::sSingleton->mPicoSystem, _status, message);          \
+    NS_WARNING(                                                           \
+      nsPrintfCString("Error running %s: %s", _funcName, message).get()); \
+    return;                                                               \
+  }
+
+#define PICO_ENSURE_SUCCESS(_funcName, _status, _rv)                      \
+  if (_status < 0) {                                                      \
+    PicoApi::pico_Retstring message;                                      \
+    sPicoApi.pico_getSystemStatusMessage(                                 \
+      nsPicoService::sSingleton->mPicoSystem, _status, message);          \
+    NS_WARNING(                                                           \
+      nsPrintfCString("Error running %s: %s", _funcName, message).get()); \
+    return _rv;                                                           \
+  }
+
+class PicoVoice
+{
+public:
+
+  PicoVoice(const nsAString& aLanguage)
+    : mLanguage(aLanguage) {}
+
+  NS_INLINE_DECL_THREADSAFE_REFCOUNTING(PicoVoice)
+
+  // Voice language, in BCB-47 syntax
+  nsString mLanguage;
+
+  // Language resource file
+  nsCString mTaFile;
+
+  // Speaker resource file
+  nsCString mSgFile;
+
+private:
+    ~PicoVoice() {}
+};
+
+class PicoCallbackRunnable : public Runnable,
+                             public nsISpeechTaskCallback
+{
+  friend class PicoSynthDataRunnable;
+
+public:
+  PicoCallbackRunnable(const nsAString& aText, PicoVoice* aVoice,
+                       float aRate, float aPitch, nsISpeechTask* aTask,
+                       nsPicoService* aService)
+    : mText(NS_ConvertUTF16toUTF8(aText))
+    , mRate(aRate)
+    , mPitch(aPitch)
+    , mFirstData(true)
+    , mTask(aTask)
+    , mVoice(aVoice)
+    , mService(aService) { }
+
+  NS_DECL_ISUPPORTS_INHERITED
+  NS_DECL_NSISPEECHTASKCALLBACK
+
+  NS_IMETHOD Run() override;
+
+  bool IsCurrentTask() { return mService->mCurrentTask == mTask; }
+
+private:
+  ~PicoCallbackRunnable() { }
+
+  void DispatchSynthDataRunnable(already_AddRefed<SharedBuffer>&& aBuffer,
+                                 size_t aBufferSize);
+
+  nsCString mText;
+
+  float mRate;
+
+  float mPitch;
+
+  bool mFirstData;
+
+  // We use this pointer to compare it with the current service task.
+  // If they differ, this runnable should stop.
+  nsISpeechTask* mTask;
+
+  // We hold a strong reference to the service, which in turn holds
+  // a strong reference to this voice.
+  PicoVoice* mVoice;
+
+  // By holding a strong reference to the service we guarantee that it won't be
+  // destroyed before this runnable.
+  RefPtr<nsPicoService> mService;
+};
+
+NS_IMPL_ISUPPORTS_INHERITED(PicoCallbackRunnable, Runnable, nsISpeechTaskCallback)
+
+// Runnable
+
+NS_IMETHODIMP
+PicoCallbackRunnable::Run()
+{
+  MOZ_ASSERT(!NS_IsMainThread());
+  PicoApi::pico_Status status = 0;
+
+  if (mService->CurrentVoice() != mVoice) {
+    mService->LoadEngine(mVoice);
+  } else {
+    status = sPicoApi.pico_resetEngine(mService->mPicoEngine, PICO_RESET_SOFT);
+    PICO_ENSURE_SUCCESS("pico_unloadResource", status, NS_ERROR_FAILURE);
+  }
+
+  // Add SSML markup for pitch and rate. Pico uses a minimal parser,
+  // so no namespace is needed.
+  nsPrintfCString markedUpText(
+    "<pitch level=\"%0.0f\"><speed level=\"%0.0f\">%s</speed></pitch>",
+    std::min(std::max(50.0f, mPitch * 100), 200.0f),
+    std::min(std::max(20.0f, mRate * 100), 500.0f),
+    mText.get());
+
+  const char* text = markedUpText.get();
+  size_t buffer_size = 512, buffer_offset = 0;
+  RefPtr<SharedBuffer> buffer = SharedBuffer::Create(buffer_size);
+  int16_t text_offset = 0, bytes_recv = 0, bytes_sent = 0, out_data_type = 0;
+  int16_t text_remaining = markedUpText.Length() + 1;
+
+  // Run this loop while this is the current task
+  while (IsCurrentTask()) {
+    if (text_remaining) {
+      status = sPicoApi.pico_putTextUtf8(mService->mPicoEngine,
+                                         text + text_offset, text_remaining,
+                                         &bytes_sent);
+      PICO_ENSURE_SUCCESS("pico_putTextUtf8", status, NS_ERROR_FAILURE);
+      // XXX: End speech task on error
+      text_remaining -= bytes_sent;
+      text_offset += bytes_sent;
+    } else {
+      // If we already fed all the text to the engine, send a zero length buffer
+      // and quit.
+      DispatchSynthDataRunnable(already_AddRefed<SharedBuffer>(), 0);
+      break;
+    }
+
+    do {
+      // Run this loop while the result of getData is STEP_BUSY, when it finishes
+      // synthesizing audio for the given text, it returns STEP_IDLE. We then
+      // break to the outer loop and feed more text, if there is any left.
+      if (!IsCurrentTask()) {
+        // If the task has changed, quit.
+        break;
+      }
+
+      if (buffer_size - buffer_offset < PICO_MAX_CHUNK_SIZE) {
+        // The next audio chunk retrieved may be bigger than our buffer,
+        // so send the data and flush the buffer.
+        DispatchSynthDataRunnable(buffer.forget(), buffer_offset);
+        buffer_offset = 0;
+        buffer = SharedBuffer::Create(buffer_size);
+      }
+
+      status = sPicoApi.pico_getData(mService->mPicoEngine,
+                                     (uint8_t*)buffer->Data() + buffer_offset,
+                                     PICO_MAX_CHUNK_SIZE,
+                                     &bytes_recv, &out_data_type);
+      PICO_ENSURE_SUCCESS("pico_getData", status, NS_ERROR_FAILURE);
+      buffer_offset += bytes_recv;
+    } while (status == PICO_STEP_BUSY);
+  }
+
+  return NS_OK;
+}
+
+void
+PicoCallbackRunnable::DispatchSynthDataRunnable(
+  already_AddRefed<SharedBuffer>&& aBuffer, size_t aBufferSize)
+{
+  class PicoSynthDataRunnable final : public Runnable
+  {
+  public:
+    PicoSynthDataRunnable(already_AddRefed<SharedBuffer>& aBuffer,
+                          size_t aBufferSize, bool aFirstData,
+                          PicoCallbackRunnable* aCallback)
+      : mBuffer(aBuffer)
+      , mBufferSize(aBufferSize)
+      , mFirstData(aFirstData)
+      , mCallback(aCallback) {
+    }
+
+    NS_IMETHOD Run() override
+    {
+      MOZ_ASSERT(NS_IsMainThread());
+
+      if (!mCallback->IsCurrentTask()) {
+        return NS_ERROR_NOT_AVAILABLE;
+      }
+
+      nsISpeechTask* task = mCallback->mTask;
+
+      if (mFirstData) {
+        task->Setup(mCallback, PICO_CHANNELS_NUM, PICO_SAMPLE_RATE, 2);
+      }
+
+      return task->SendAudioNative(
+        mBufferSize ? static_cast<short*>(mBuffer->Data()) : nullptr, mBufferSize / 2);
+    }
+
+  private:
+    RefPtr<SharedBuffer> mBuffer;
+
+    size_t mBufferSize;
+
+    bool mFirstData;
+
+    RefPtr<PicoCallbackRunnable> mCallback;
+  };
+
+  nsCOMPtr<nsIRunnable> sendEvent =
+    new PicoSynthDataRunnable(aBuffer, aBufferSize, mFirstData, this);
+  NS_DispatchToMainThread(sendEvent);
+  mFirstData = false;
+}
+
+// nsISpeechTaskCallback
+
+NS_IMETHODIMP
+PicoCallbackRunnable::OnPause()
+{
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+PicoCallbackRunnable::OnResume()
+{
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+PicoCallbackRunnable::OnCancel()
+{
+  mService->mCurrentTask = nullptr;
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+PicoCallbackRunnable::OnVolumeChanged(float aVolume)
+{
+  return NS_OK;
+}
+
+NS_INTERFACE_MAP_BEGIN(nsPicoService)
+  NS_INTERFACE_MAP_ENTRY(nsISpeechService)
+  NS_INTERFACE_MAP_ENTRY(nsIObserver)
+  NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsIObserver)
+NS_INTERFACE_MAP_END
+
+NS_IMPL_ADDREF(nsPicoService)
+NS_IMPL_RELEASE(nsPicoService)
+
+nsPicoService::nsPicoService()
+  : mInitialized(false)
+  , mVoicesMonitor("nsPicoService::mVoices")
+  , mCurrentTask(nullptr)
+  , mPicoSystem(nullptr)
+  , mPicoEngine(nullptr)
+  , mSgResource(nullptr)
+  , mTaResource(nullptr)
+  , mPicoMemArea(nullptr)
+{
+}
+
+nsPicoService::~nsPicoService()
+{
+  // We don't worry about removing the voices because this gets
+  // destructed at shutdown along with the voice registry.
+  MonitorAutoLock autoLock(mVoicesMonitor);
+  mVoices.Clear();
+
+  if (mThread) {
+    mThread->Shutdown();
+  }
+
+  UnloadEngine();
+}
+
+// nsIObserver
+
+NS_IMETHODIMP
+nsPicoService::Observe(nsISupports* aSubject, const char* aTopic,
+                       const char16_t* aData)
+{
+  MOZ_ASSERT(NS_IsMainThread());
+  if(NS_WARN_IF(!(!strcmp(aTopic, "profile-after-change")))) {
+    return NS_ERROR_UNEXPECTED;
+  }
+
+  if (!Preferences::GetBool("media.webspeech.synth.enabled") ||
+      Preferences::GetBool("media.webspeech.synth.test")) {
+    return NS_OK;
+  }
+
+  DebugOnly<nsresult> rv = NS_NewNamedThread("Pico Worker", getter_AddRefs(mThread));
+  MOZ_ASSERT(NS_SUCCEEDED(rv));
+  return mThread->Dispatch(
+    NewRunnableMethod(this, &nsPicoService::Init), NS_DISPATCH_NORMAL);
+}
+// nsISpeechService
+
+NS_IMETHODIMP
+nsPicoService::Speak(const nsAString& aText, const nsAString& aUri,
+                     float aVolume, float aRate, float aPitch,
+                     nsISpeechTask* aTask)
+{
+  if(NS_WARN_IF(!(mInitialized))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  MonitorAutoLock autoLock(mVoicesMonitor);
+  bool found = false;
+  PicoVoice* voice = mVoices.GetWeak(aUri, &found);
+  if(NS_WARN_IF(!(found))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  mCurrentTask = aTask;
+  RefPtr<PicoCallbackRunnable> cb = new PicoCallbackRunnable(aText, voice, aRate, aPitch, aTask, this);
+  return mThread->Dispatch(cb, NS_DISPATCH_NORMAL);
+}
+
+NS_IMETHODIMP
+nsPicoService::GetServiceType(SpeechServiceType* aServiceType)
+{
+  *aServiceType = nsISpeechService::SERVICETYPE_DIRECT_AUDIO;
+  return NS_OK;
+}
+
+// private methods
+
+void
+nsPicoService::Init()
+{
+  MOZ_ASSERT(!NS_IsMainThread());
+  MOZ_ASSERT(!mInitialized);
+
+  if (!sPicoApi.Init()) {
+    NS_WARNING("Failed to initialize pico library");
+    return;
+  }
+
+  // Use environment variable, or default android/b2g path
+  nsAutoCString langPath(PR_GetEnv("PICO_LANG_PATH"));
+
+  if (langPath.IsEmpty()) {
+    langPath.AssignLiteral(GONK_PICO_LANG_PATH);
+  }
+
+  nsCOMPtr<nsIFile> voicesDir;
+  NS_NewNativeLocalFile(langPath, true, getter_AddRefs(voicesDir));
+
+  nsCOMPtr<nsISimpleEnumerator> dirIterator;
+  nsresult rv = voicesDir->GetDirectoryEntries(getter_AddRefs(dirIterator));
+
+  if (NS_FAILED(rv)) {
+    NS_WARNING(nsPrintfCString("Failed to get contents of directory: %s", langPath.get()).get());
+    return;
+  }
+
+  bool hasMoreElements = false;
+  rv = dirIterator->HasMoreElements(&hasMoreElements);
+  MOZ_ASSERT(NS_SUCCEEDED(rv));
+
+  MonitorAutoLock autoLock(mVoicesMonitor);
+
+  while (hasMoreElements && NS_SUCCEEDED(rv)) {
+    nsCOMPtr<nsISupports> supports;
+    rv = dirIterator->GetNext(getter_AddRefs(supports));
+    MOZ_ASSERT(NS_SUCCEEDED(rv));
+
+    nsCOMPtr<nsIFile> voiceFile = do_QueryInterface(supports);
+    MOZ_ASSERT(voiceFile);
+
+    nsAutoCString leafName;
+    voiceFile->GetNativeLeafName(leafName);
+
+    nsAutoString lang;
+
+    if (GetVoiceFileLanguage(leafName, lang)) {
+      nsAutoString uri;
+      uri.AssignLiteral("urn:moz-tts:pico:");
+      uri.Append(lang);
+
+      bool found = false;
+      PicoVoice* voice = mVoices.GetWeak(uri, &found);
+
+      if (!found) {
+        voice = new PicoVoice(lang);
+        mVoices.Put(uri, voice);
+      }
+
+      // Each voice consists of two lingware files: A language resource file,
+      // suffixed by _ta.bin, and a speaker resource file, suffixed by _sb.bin.
+      // We currently assume that there is a pair of files for each language.
+      if (StringEndsWith(leafName, NS_LITERAL_CSTRING("_ta.bin"))) {
+        rv = voiceFile->GetPersistentDescriptor(voice->mTaFile);
+        MOZ_ASSERT(NS_SUCCEEDED(rv));
+      } else if (StringEndsWith(leafName, NS_LITERAL_CSTRING("_sg.bin"))) {
+        rv = voiceFile->GetPersistentDescriptor(voice->mSgFile);
+        MOZ_ASSERT(NS_SUCCEEDED(rv));
+      }
+    }
+
+    rv = dirIterator->HasMoreElements(&hasMoreElements);
+  }
+
+  NS_DispatchToMainThread(NewRunnableMethod(this, &nsPicoService::RegisterVoices));
+}
+
+void
+nsPicoService::RegisterVoices()
+{
+  nsSynthVoiceRegistry* registry = nsSynthVoiceRegistry::GetInstance();
+
+  for (auto iter = mVoices.Iter(); !iter.Done(); iter.Next()) {
+    const nsAString& uri = iter.Key();
+    RefPtr<PicoVoice>& voice = iter.Data();
+
+    // If we are missing either a language or a voice resource, it is invalid.
+    if (voice->mTaFile.IsEmpty() || voice->mSgFile.IsEmpty()) {
+      iter.Remove();
+      continue;
+    }
+
+    nsAutoString name;
+    name.AssignLiteral("Pico ");
+    name.Append(voice->mLanguage);
+
+    // This service is multi-threaded and can handle more than one utterance at a
+    // time before previous utterances end. So, aQueuesUtterances == false
+    DebugOnly<nsresult> rv =
+      registry->AddVoice(this, uri, name, voice->mLanguage, true, false);
+    NS_WARNING_ASSERTION(NS_SUCCEEDED(rv), "Failed to add voice");
+  }
+
+  mInitialized = true;
+}
+
+bool
+nsPicoService::GetVoiceFileLanguage(const nsACString& aFileName, nsAString& aLang)
+{
+  nsACString::const_iterator start, end;
+  aFileName.BeginReading(start);
+  aFileName.EndReading(end);
+
+  // The lingware filename syntax is language_(ta/sg).bin,
+  // we extract the language prefix here.
+  if (FindInReadable(NS_LITERAL_CSTRING("_"), start, end)) {
+    end = start;
+    aFileName.BeginReading(start);
+    aLang.Assign(NS_ConvertUTF8toUTF16(Substring(start, end)));
+    return true;
+  }
+
+  return false;
+}
+
+void
+nsPicoService::LoadEngine(PicoVoice* aVoice)
+{
+  PicoApi::pico_Status status = 0;
+
+  if (mPicoSystem) {
+    UnloadEngine();
+  }
+
+  if (!mPicoMemArea) {
+    mPicoMemArea = MakeUnique<uint8_t[]>(PICO_MEM_SIZE);
+  }
+
+  status = sPicoApi.pico_initialize(mPicoMemArea.get(),
+                                    PICO_MEM_SIZE, &mPicoSystem);
+  PICO_ENSURE_SUCCESS_VOID("pico_initialize", status);
+
+  status = sPicoApi.pico_loadResource(mPicoSystem, aVoice->mTaFile.get(), &mTaResource);
+  PICO_ENSURE_SUCCESS_VOID("pico_loadResource", status);
+
+  status = sPicoApi.pico_loadResource(mPicoSystem, aVoice->mSgFile.get(), &mSgResource);
+  PICO_ENSURE_SUCCESS_VOID("pico_loadResource", status);
+
+  status = sPicoApi.pico_createVoiceDefinition(mPicoSystem, PICO_VOICE_NAME);
+  PICO_ENSURE_SUCCESS_VOID("pico_createVoiceDefinition", status);
+
+  char taName[PICO_RETSTRINGSIZE];
+  status = sPicoApi.pico_getResourceName(mPicoSystem, mTaResource, taName);
+  PICO_ENSURE_SUCCESS_VOID("pico_getResourceName", status);
+
+  status = sPicoApi.pico_addResourceToVoiceDefinition(
+    mPicoSystem, PICO_VOICE_NAME, taName);
+  PICO_ENSURE_SUCCESS_VOID("pico_addResourceToVoiceDefinition", status);
+
+  char sgName[PICO_RETSTRINGSIZE];
+  status = sPicoApi.pico_getResourceName(mPicoSystem, mSgResource, sgName);
+  PICO_ENSURE_SUCCESS_VOID("pico_getResourceName", status);
+
+  status = sPicoApi.pico_addResourceToVoiceDefinition(
+    mPicoSystem, PICO_VOICE_NAME, sgName);
+  PICO_ENSURE_SUCCESS_VOID("pico_addResourceToVoiceDefinition", status);
+
+  status = sPicoApi.pico_newEngine(mPicoSystem, PICO_VOICE_NAME, &mPicoEngine);
+  PICO_ENSURE_SUCCESS_VOID("pico_newEngine", status);
+
+  if (sSingleton) {
+    sSingleton->mCurrentVoice = aVoice;
+  }
+}
+
+void
+nsPicoService::UnloadEngine()
+{
+  PicoApi::pico_Status status = 0;
+
+  if (mPicoEngine) {
+    status = sPicoApi.pico_disposeEngine(mPicoSystem, &mPicoEngine);
+    PICO_ENSURE_SUCCESS_VOID("pico_disposeEngine", status);
+    status = sPicoApi.pico_releaseVoiceDefinition(mPicoSystem, PICO_VOICE_NAME);
+    PICO_ENSURE_SUCCESS_VOID("pico_releaseVoiceDefinition", status);
+    mPicoEngine = nullptr;
+  }
+
+  if (mSgResource) {
+    status = sPicoApi.pico_unloadResource(mPicoSystem, &mSgResource);
+    PICO_ENSURE_SUCCESS_VOID("pico_unloadResource", status);
+    mSgResource = nullptr;
+  }
+
+  if (mTaResource) {
+    status = sPicoApi.pico_unloadResource(mPicoSystem, &mTaResource);
+    PICO_ENSURE_SUCCESS_VOID("pico_unloadResource", status);
+    mTaResource = nullptr;
+  }
+
+  if (mPicoSystem) {
+    status = sPicoApi.pico_terminate(&mPicoSystem);
+    PICO_ENSURE_SUCCESS_VOID("pico_terminate", status);
+    mPicoSystem = nullptr;
+  }
+}
+
+PicoVoice*
+nsPicoService::CurrentVoice()
+{
+  MOZ_ASSERT(!NS_IsMainThread());
+
+  return mCurrentVoice;
+}
+
+// static methods
+
+nsPicoService*
+nsPicoService::GetInstance()
+{
+  MOZ_ASSERT(NS_IsMainThread());
+  if (!XRE_IsParentProcess()) {
+    MOZ_ASSERT(false, "nsPicoService can only be started on main gecko process");
+    return nullptr;
+  }
+
+  if (!sSingleton) {
+    sSingleton = new nsPicoService();
+  }
+
+  return sSingleton;
+}
+
+already_AddRefed<nsPicoService>
+nsPicoService::GetInstanceForService()
+{
+  RefPtr<nsPicoService> picoService = GetInstance();
+  return picoService.forget();
+}
+
+void
+nsPicoService::Shutdown()
+{
+  if (!sSingleton) {
+    return;
+  }
+
+  sSingleton->mCurrentTask = nullptr;
+
+  sSingleton = nullptr;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/pico/nsPicoService.h

@@ -0,0 +1,93 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef nsPicoService_h
+#define nsPicoService_h
+
+#include "mozilla/Mutex.h"
+#include "nsTArray.h"
+#include "nsIObserver.h"
+#include "nsIThread.h"
+#include "nsISpeechService.h"
+#include "nsRefPtrHashtable.h"
+#include "mozilla/StaticPtr.h"
+#include "mozilla/Monitor.h"
+#include "mozilla/UniquePtr.h"
+
+namespace mozilla {
+namespace dom {
+
+class PicoVoice;
+class PicoCallbackRunnable;
+
+typedef void* pico_System;
+typedef void* pico_Resource;
+typedef void* pico_Engine;
+
+class nsPicoService : public nsIObserver,
+                      public nsISpeechService
+{
+  friend class PicoCallbackRunnable;
+  friend class PicoInitRunnable;
+
+public:
+  NS_DECL_THREADSAFE_ISUPPORTS
+  NS_DECL_NSISPEECHSERVICE
+  NS_DECL_NSIOBSERVER
+
+  nsPicoService();
+
+  static nsPicoService* GetInstance();
+
+  static already_AddRefed<nsPicoService> GetInstanceForService();
+
+  static void Shutdown();
+
+private:
+
+  virtual ~nsPicoService();
+
+  void Init();
+
+  void RegisterVoices();
+
+  bool GetVoiceFileLanguage(const nsACString& aFileName, nsAString& aLang);
+
+  void LoadEngine(PicoVoice* aVoice);
+
+  void UnloadEngine();
+
+  PicoVoice* CurrentVoice();
+
+  bool mInitialized;
+
+  nsCOMPtr<nsIThread> mThread;
+
+  nsRefPtrHashtable<nsStringHashKey, PicoVoice> mVoices;
+
+  Monitor mVoicesMonitor;
+
+  PicoVoice* mCurrentVoice;
+
+  Atomic<nsISpeechTask*> mCurrentTask;
+
+  pico_System mPicoSystem;
+
+  pico_Engine mPicoEngine;
+
+  pico_Resource mSgResource;
+
+  pico_Resource mTaResource;
+
+  mozilla::UniquePtr<uint8_t[]> mPicoMemArea;
+
+  static StaticRefPtr<nsPicoService> sSingleton;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif

dom/media/webspeech/synth/speechd/SpeechDispatcherModule.cpp

@@ -0,0 +1,56 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim: set ts=8 sts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "mozilla/ModuleUtils.h"
+#include "nsIClassInfoImpl.h"
+#include "SpeechDispatcherService.h"
+
+using namespace mozilla::dom;
+
+#define SPEECHDISPATCHERSERVICE_CID \
+  {0x8817b1cf, 0x5ada, 0x43bf, {0xbd, 0x73, 0x60, 0x76, 0x57, 0x70, 0x3d, 0x0d}}
+
+#define SPEECHDISPATCHERSERVICE_CONTRACTID "@mozilla.org/synthspeechdispatcher;1"
+
+// Defines SpeechDispatcherServiceConstructor
+NS_GENERIC_FACTORY_SINGLETON_CONSTRUCTOR(SpeechDispatcherService,
+                                         SpeechDispatcherService::GetInstanceForService)
+
+// Defines kSPEECHDISPATCHERSERVICE_CID
+NS_DEFINE_NAMED_CID(SPEECHDISPATCHERSERVICE_CID);
+
+static const mozilla::Module::CIDEntry kCIDs[] = {
+  { &kSPEECHDISPATCHERSERVICE_CID, true, nullptr, SpeechDispatcherServiceConstructor },
+  { nullptr }
+};
+
+static const mozilla::Module::ContractIDEntry kContracts[] = {
+  { SPEECHDISPATCHERSERVICE_CONTRACTID, &kSPEECHDISPATCHERSERVICE_CID },
+  { nullptr }
+};
+
+static const mozilla::Module::CategoryEntry kCategories[] = {
+  { "speech-synth-started", "SpeechDispatcher Speech Synth", SPEECHDISPATCHERSERVICE_CONTRACTID },
+  { nullptr }
+};
+
+static void
+UnloadSpeechDispatcherModule()
+{
+  SpeechDispatcherService::Shutdown();
+}
+
+static const mozilla::Module kModule = {
+  mozilla::Module::kVersion,
+  kCIDs,
+  kContracts,
+  kCategories,
+  nullptr,
+  nullptr,
+  UnloadSpeechDispatcherModule
+};
+
+NSMODULE_DEFN(synthspeechdispatcher) = &kModule;

dom/media/webspeech/synth/speechd/SpeechDispatcherService.cpp

@@ -0,0 +1,593 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim: set ts=8 sts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "SpeechDispatcherService.h"
+
+#include "mozilla/dom/nsSpeechTask.h"
+#include "mozilla/dom/nsSynthVoiceRegistry.h"
+#include "mozilla/Preferences.h"
+#include "nsEscape.h"
+#include "nsISupports.h"
+#include "nsPrintfCString.h"
+#include "nsReadableUtils.h"
+#include "nsServiceManagerUtils.h"
+#include "nsThreadUtils.h"
+#include "prlink.h"
+
+#include <math.h>
+#include <stdlib.h>
+
+#define URI_PREFIX "urn:moz-tts:speechd:"
+
+#define MAX_RATE static_cast<float>(2.5)
+#define MIN_RATE static_cast<float>(0.5)
+
+// Some structures for libspeechd
+typedef enum {
+  SPD_EVENT_BEGIN,
+  SPD_EVENT_END,
+  SPD_EVENT_INDEX_MARK,
+  SPD_EVENT_CANCEL,
+  SPD_EVENT_PAUSE,
+  SPD_EVENT_RESUME
+} SPDNotificationType;
+
+typedef enum {
+  SPD_BEGIN = 1,
+  SPD_END = 2,
+  SPD_INDEX_MARKS = 4,
+  SPD_CANCEL = 8,
+  SPD_PAUSE = 16,
+  SPD_RESUME = 32,
+
+  SPD_ALL = 0x3f
+} SPDNotification;
+
+typedef enum {
+  SPD_MODE_SINGLE = 0,
+  SPD_MODE_THREADED = 1
+} SPDConnectionMode;
+
+typedef void (*SPDCallback) (size_t msg_id, size_t client_id,
+                             SPDNotificationType state);
+
+typedef void (*SPDCallbackIM) (size_t msg_id, size_t client_id,
+                               SPDNotificationType state, char* index_mark);
+
+struct SPDConnection
+{
+  SPDCallback callback_begin;
+  SPDCallback callback_end;
+  SPDCallback callback_cancel;
+  SPDCallback callback_pause;
+  SPDCallback callback_resume;
+  SPDCallbackIM callback_im;
+
+  /* partial, more private fields in structure */
+};
+
+struct SPDVoice
+{
+  char* name;
+  char* language;
+  char* variant;
+};
+
+typedef enum {
+  SPD_IMPORTANT = 1,
+  SPD_MESSAGE = 2,
+  SPD_TEXT = 3,
+  SPD_NOTIFICATION = 4,
+  SPD_PROGRESS = 5
+} SPDPriority;
+
+#define SPEECHD_FUNCTIONS \
+  FUNC(spd_open, SPDConnection*, (const char*, const char*, const char*, SPDConnectionMode)) \
+  FUNC(spd_close, void, (SPDConnection*)) \
+  FUNC(spd_list_synthesis_voices, SPDVoice**, (SPDConnection*)) \
+  FUNC(spd_say, int, (SPDConnection*, SPDPriority, const char*)) \
+  FUNC(spd_cancel, int, (SPDConnection*)) \
+  FUNC(spd_set_volume, int, (SPDConnection*, int)) \
+  FUNC(spd_set_voice_rate, int, (SPDConnection*, int)) \
+  FUNC(spd_set_voice_pitch, int, (SPDConnection*, int)) \
+  FUNC(spd_set_synthesis_voice, int, (SPDConnection*, const char*)) \
+  FUNC(spd_set_notification_on, int, (SPDConnection*, SPDNotification))
+
+#define FUNC(name, type, params) \
+  typedef type (*_##name##_fn) params; \
+  static _##name##_fn _##name;
+
+SPEECHD_FUNCTIONS
+
+#undef FUNC
+
+#define spd_open _spd_open
+#define spd_close _spd_close
+#define spd_list_synthesis_voices _spd_list_synthesis_voices
+#define spd_say _spd_say
+#define spd_cancel _spd_cancel
+#define spd_set_volume _spd_set_volume
+#define spd_set_voice_rate _spd_set_voice_rate
+#define spd_set_voice_pitch _spd_set_voice_pitch
+#define spd_set_synthesis_voice _spd_set_synthesis_voice
+#define spd_set_notification_on _spd_set_notification_on
+
+static PRLibrary* speechdLib = nullptr;
+
+typedef void (*nsSpeechDispatcherFunc)();
+struct nsSpeechDispatcherDynamicFunction
+{
+  const char* functionName;
+  nsSpeechDispatcherFunc* function;
+};
+
+namespace mozilla {
+namespace dom {
+
+StaticRefPtr<SpeechDispatcherService> SpeechDispatcherService::sSingleton;
+
+class SpeechDispatcherVoice
+{
+public:
+
+  SpeechDispatcherVoice(const nsAString& aName, const nsAString& aLanguage)
+    : mName(aName), mLanguage(aLanguage) {}
+
+  NS_INLINE_DECL_THREADSAFE_REFCOUNTING(SpeechDispatcherVoice)
+
+  // Voice name
+  nsString mName;
+
+  // Voice language, in BCP-47 syntax
+  nsString mLanguage;
+
+private:
+  ~SpeechDispatcherVoice() {}
+};
+
+
+class SpeechDispatcherCallback final : public nsISpeechTaskCallback
+{
+public:
+  SpeechDispatcherCallback(nsISpeechTask* aTask, SpeechDispatcherService* aService)
+    : mTask(aTask)
+    , mService(aService) {}
+
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_CLASS_AMBIGUOUS(SpeechDispatcherCallback, nsISpeechTaskCallback)
+
+  NS_DECL_NSISPEECHTASKCALLBACK
+
+  bool OnSpeechEvent(SPDNotificationType state);
+
+private:
+  ~SpeechDispatcherCallback() { }
+
+  // This pointer is used to dispatch events
+  nsCOMPtr<nsISpeechTask> mTask;
+
+  // By holding a strong reference to the service we guarantee that it won't be
+  // destroyed before this runnable.
+  RefPtr<SpeechDispatcherService> mService;
+
+  TimeStamp mStartTime;
+};
+
+NS_IMPL_CYCLE_COLLECTION(SpeechDispatcherCallback, mTask);
+
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(SpeechDispatcherCallback)
+  NS_INTERFACE_MAP_ENTRY(nsISpeechTaskCallback)
+  NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsISpeechTaskCallback)
+NS_INTERFACE_MAP_END
+
+NS_IMPL_CYCLE_COLLECTING_ADDREF(SpeechDispatcherCallback)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(SpeechDispatcherCallback)
+
+NS_IMETHODIMP
+SpeechDispatcherCallback::OnPause()
+{
+  // XXX: Speech dispatcher does not pause immediately, but waits for the speech
+  // to reach an index mark so that it could resume from that offset.
+  // There is no support for word or sentence boundaries, so index marks would
+  // only occur in explicit SSML marks, and we don't support that yet.
+  // What in actuality happens, is that if you call spd_pause(), it will speak
+  // the utterance in its entirety, dispatch an end event, and then put speechd
+  // in a 'paused' state. Since it is after the utterance ended, we don't get
+  // that state change, and our speech api is in an unrecoverable state.
+  // So, since it is useless anyway, I am not implementing pause.
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+SpeechDispatcherCallback::OnResume()
+{
+  // XXX: Unsupported, see OnPause().
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+SpeechDispatcherCallback::OnCancel()
+{
+  if (spd_cancel(mService->mSpeechdClient) < 0) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+SpeechDispatcherCallback::OnVolumeChanged(float aVolume)
+{
+  // XXX: This currently does not change the volume mid-utterance, but it
+  // doesn't do anything bad either. So we could put this here with the hopes
+  // that speechd supports this in the future.
+  if (spd_set_volume(mService->mSpeechdClient, static_cast<int>(aVolume * 100)) < 0) {
+    return NS_ERROR_FAILURE;
+  }
+
+  return NS_OK;
+}
+
+bool
+SpeechDispatcherCallback::OnSpeechEvent(SPDNotificationType state)
+{
+  bool remove = false;
+
+  switch (state) {
+    case SPD_EVENT_BEGIN:
+      mStartTime = TimeStamp::Now();
+      mTask->DispatchStart();
+      break;
+
+    case SPD_EVENT_PAUSE:
+      mTask->DispatchPause((TimeStamp::Now() - mStartTime).ToSeconds(), 0);
+      break;
+
+    case SPD_EVENT_RESUME:
+      mTask->DispatchResume((TimeStamp::Now() - mStartTime).ToSeconds(), 0);
+      break;
+
+    case SPD_EVENT_CANCEL:
+    case SPD_EVENT_END:
+      mTask->DispatchEnd((TimeStamp::Now() - mStartTime).ToSeconds(), 0);
+      remove = true;
+      break;
+
+    case SPD_EVENT_INDEX_MARK:
+      // Not yet supported
+      break;
+
+    default:
+      break;
+  }
+
+  return remove;
+}
+
+static void
+speechd_cb(size_t msg_id, size_t client_id, SPDNotificationType state)
+{
+  SpeechDispatcherService* service = SpeechDispatcherService::GetInstance(false);
+
+  if (service) {
+    NS_DispatchToMainThread(
+      NewRunnableMethod<uint32_t, SPDNotificationType>(
+        service, &SpeechDispatcherService::EventNotify,
+        static_cast<uint32_t>(msg_id), state));
+  }
+}
+
+
+NS_INTERFACE_MAP_BEGIN(SpeechDispatcherService)
+  NS_INTERFACE_MAP_ENTRY(nsISpeechService)
+  NS_INTERFACE_MAP_ENTRY(nsIObserver)
+  NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsIObserver)
+NS_INTERFACE_MAP_END
+
+NS_IMPL_ADDREF(SpeechDispatcherService)
+NS_IMPL_RELEASE(SpeechDispatcherService)
+
+SpeechDispatcherService::SpeechDispatcherService()
+  : mInitialized(false)
+  , mSpeechdClient(nullptr)
+{
+}
+
+void
+SpeechDispatcherService::Init()
+{
+  if (!Preferences::GetBool("media.webspeech.synth.enabled") ||
+      Preferences::GetBool("media.webspeech.synth.test")) {
+    return;
+  }
+
+  // While speech dispatcher has a "threaded" mode, only spd_say() is async.
+  // Since synchronous socket i/o could impact startup time, we do
+  // initialization in a separate thread.
+  DebugOnly<nsresult> rv = NS_NewNamedThread("speechd init",
+                                             getter_AddRefs(mInitThread));
+  MOZ_ASSERT(NS_SUCCEEDED(rv));
+  rv = mInitThread->Dispatch(
+    NewRunnableMethod(this, &SpeechDispatcherService::Setup), NS_DISPATCH_NORMAL);
+  MOZ_ASSERT(NS_SUCCEEDED(rv));
+}
+
+SpeechDispatcherService::~SpeechDispatcherService()
+{
+  if (mInitThread) {
+    mInitThread->Shutdown();
+  }
+
+  if (mSpeechdClient) {
+    spd_close(mSpeechdClient);
+  }
+}
+
+void
+SpeechDispatcherService::Setup()
+{
+#define FUNC(name, type, params) { #name, (nsSpeechDispatcherFunc *)&_##name },
+  static const nsSpeechDispatcherDynamicFunction kSpeechDispatcherSymbols[] = {
+    SPEECHD_FUNCTIONS
+  };
+#undef FUNC
+
+  MOZ_ASSERT(!mInitialized);
+
+  speechdLib = PR_LoadLibrary("libspeechd.so.2");
+
+  if (!speechdLib) {
+    NS_WARNING("Failed to load speechd library");
+    return;
+  }
+
+  if (!PR_FindFunctionSymbol(speechdLib, "spd_get_volume")) {
+    // There is no version getter function, so we rely on a symbol that was
+    // introduced in release 0.8.2 in order to check for ABI compatibility.
+    NS_WARNING("Unsupported version of speechd detected");
+    return;
+  }
+
+  for (uint32_t i = 0; i < ArrayLength(kSpeechDispatcherSymbols); i++) {
+    *kSpeechDispatcherSymbols[i].function =
+      PR_FindFunctionSymbol(speechdLib, kSpeechDispatcherSymbols[i].functionName);
+
+    if (!*kSpeechDispatcherSymbols[i].function) {
+      NS_WARNING(nsPrintfCString("Failed to find speechd symbol for'%s'",
+                                 kSpeechDispatcherSymbols[i].functionName).get());
+      return;
+    }
+  }
+
+  mSpeechdClient = spd_open("firefox", "web speech api", "who", SPD_MODE_THREADED);
+  if (!mSpeechdClient) {
+    NS_WARNING("Failed to call spd_open");
+    return;
+  }
+
+  // Get all the voices from sapi and register in the SynthVoiceRegistry
+  SPDVoice** list = spd_list_synthesis_voices(mSpeechdClient);
+
+  mSpeechdClient->callback_begin = speechd_cb;
+  mSpeechdClient->callback_end = speechd_cb;
+  mSpeechdClient->callback_cancel = speechd_cb;
+  mSpeechdClient->callback_pause = speechd_cb;
+  mSpeechdClient->callback_resume = speechd_cb;
+
+  spd_set_notification_on(mSpeechdClient, SPD_BEGIN);
+  spd_set_notification_on(mSpeechdClient, SPD_END);
+  spd_set_notification_on(mSpeechdClient, SPD_CANCEL);
+
+  if (list != NULL) {
+    for (int i = 0; list[i]; i++) {
+      nsAutoString uri;
+
+      uri.AssignLiteral(URI_PREFIX);
+      nsAutoCString name;
+      NS_EscapeURL(list[i]->name, -1, esc_OnlyNonASCII | esc_AlwaysCopy, name);
+      uri.Append(NS_ConvertUTF8toUTF16(name));;
+      uri.AppendLiteral("?");
+
+      nsAutoCString lang(list[i]->language);
+
+      if (strcmp(list[i]->variant, "none") != 0) {
+        // In speech dispatcher, the variant will usually be the locale subtag
+        // with another, non-standard suptag after it. We keep the first one
+        // and convert it to uppercase.
+        const char* v = list[i]->variant;
+        const char* hyphen = strchr(v, '-');
+        nsDependentCSubstring variant(v, hyphen ? hyphen - v : strlen(v));
+        ToUpperCase(variant);
+
+        // eSpeak uses UK which is not a valid region subtag in BCP47.
+        if (variant.Equals("UK")) {
+          variant.AssignLiteral("GB");
+        }
+
+        lang.AppendLiteral("-");
+        lang.Append(variant);
+      }
+
+      uri.Append(NS_ConvertUTF8toUTF16(lang));
+
+      mVoices.Put(uri, new SpeechDispatcherVoice(
+                    NS_ConvertUTF8toUTF16(list[i]->name),
+                    NS_ConvertUTF8toUTF16(lang)));
+    }
+  }
+
+  NS_DispatchToMainThread(NewRunnableMethod(this, &SpeechDispatcherService::RegisterVoices));
+
+  //mInitialized = true;
+}
+
+// private methods
+
+void
+SpeechDispatcherService::RegisterVoices()
+{
+  RefPtr<nsSynthVoiceRegistry> registry = nsSynthVoiceRegistry::GetInstance();
+  for (auto iter = mVoices.Iter(); !iter.Done(); iter.Next()) {
+    RefPtr<SpeechDispatcherVoice>& voice = iter.Data();
+
+    // This service can only speak one utterance at a time, so we set
+    // aQueuesUtterances to true in order to track global state and schedule
+    // access to this service.
+    DebugOnly<nsresult> rv =
+      registry->AddVoice(this, iter.Key(), voice->mName, voice->mLanguage,
+                         voice->mName.EqualsLiteral("default"), true);
+
+    NS_WARNING_ASSERTION(NS_SUCCEEDED(rv), "Failed to add voice");
+  }
+
+  mInitThread->Shutdown();
+  mInitThread = nullptr;
+
+  mInitialized = true;
+
+  registry->NotifyVoicesChanged();
+}
+
+// nsIObserver
+
+NS_IMETHODIMP
+SpeechDispatcherService::Observe(nsISupports* aSubject, const char* aTopic,
+                                 const char16_t* aData)
+{
+  return NS_OK;
+}
+
+// nsISpeechService
+
+// TODO: Support SSML
+NS_IMETHODIMP
+SpeechDispatcherService::Speak(const nsAString& aText, const nsAString& aUri,
+                               float aVolume, float aRate, float aPitch,
+                               nsISpeechTask* aTask)
+{
+  if (NS_WARN_IF(!mInitialized)) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  RefPtr<SpeechDispatcherCallback> callback =
+    new SpeechDispatcherCallback(aTask, this);
+
+  bool found = false;
+  SpeechDispatcherVoice* voice = mVoices.GetWeak(aUri, &found);
+
+  if(NS_WARN_IF(!(found))) {
+    return NS_ERROR_NOT_AVAILABLE;
+  }
+
+  spd_set_synthesis_voice(mSpeechdClient,
+                          NS_ConvertUTF16toUTF8(voice->mName).get());
+
+  // We provide a volume of 0.0 to 1.0, speech-dispatcher expects 0 - 100.
+  spd_set_volume(mSpeechdClient, static_cast<int>(aVolume * 100));
+
+  // aRate is a value of 0.1 (0.1x) to 10 (10x) with 1 (1x) being normal rate.
+  // speechd expects -100 to 100 with 0 being normal rate.
+  float rate = 0;
+  if (aRate > 1) {
+    // Each step to 100 is logarithmically distributed up to 2.5x.
+    rate = log10(std::min(aRate, MAX_RATE)) / log10(MAX_RATE) * 100;
+  } else if (aRate < 1) {
+    // Each step to -100 is logarithmically distributed down to 0.5x.
+    rate = log10(std::max(aRate, MIN_RATE)) / log10(MIN_RATE) * -100;
+  }
+
+  spd_set_voice_rate(mSpeechdClient, static_cast<int>(rate));
+
+  // We provide a pitch of 0 to 2 with 1 being the default.
+  // speech-dispatcher expects -100 to 100 with 0 being default.
+  spd_set_voice_pitch(mSpeechdClient, static_cast<int>((aPitch - 1) * 100));
+
+  // The last three parameters don't matter for an indirect service
+  nsresult rv = aTask->Setup(callback, 0, 0, 0);
+
+  if (NS_FAILED(rv)) {
+    return rv;
+  }
+
+  if (aText.Length()) {
+    int msg_id = spd_say(
+      mSpeechdClient, SPD_MESSAGE, NS_ConvertUTF16toUTF8(aText).get());
+
+    if (msg_id < 0) {
+      return NS_ERROR_FAILURE;
+    }
+
+    mCallbacks.Put(msg_id, callback);
+  } else {
+    // Speech dispatcher does not work well with empty strings.
+    // In that case, don't send empty string to speechd,
+    // and just emulate a speechd start and end event.
+    NS_DispatchToMainThread(NewRunnableMethod<SPDNotificationType>(
+        callback, &SpeechDispatcherCallback::OnSpeechEvent, SPD_EVENT_BEGIN));
+
+    NS_DispatchToMainThread(NewRunnableMethod<SPDNotificationType>(
+        callback, &SpeechDispatcherCallback::OnSpeechEvent, SPD_EVENT_END));
+  }
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+SpeechDispatcherService::GetServiceType(SpeechServiceType* aServiceType)
+{
+  *aServiceType = nsISpeechService::SERVICETYPE_INDIRECT_AUDIO;
+  return NS_OK;
+}
+
+SpeechDispatcherService*
+SpeechDispatcherService::GetInstance(bool create)
+{
+  if (XRE_GetProcessType() != GeckoProcessType_Default) {
+    MOZ_ASSERT(false,
+               "SpeechDispatcherService can only be started on main gecko process");
+    return nullptr;
+  }
+
+  if (!sSingleton && create) {
+    sSingleton = new SpeechDispatcherService();
+    sSingleton->Init();
+  }
+
+  return sSingleton;
+}
+
+already_AddRefed<SpeechDispatcherService>
+SpeechDispatcherService::GetInstanceForService()
+{
+  MOZ_ASSERT(NS_IsMainThread());
+  RefPtr<SpeechDispatcherService> sapiService = GetInstance();
+  return sapiService.forget();
+}
+
+void
+SpeechDispatcherService::EventNotify(uint32_t aMsgId, uint32_t aState)
+{
+  SpeechDispatcherCallback* callback = mCallbacks.GetWeak(aMsgId);
+
+  if (callback) {
+    if (callback->OnSpeechEvent((SPDNotificationType)aState)) {
+      mCallbacks.Remove(aMsgId);
+    }
+  }
+}
+
+void
+SpeechDispatcherService::Shutdown()
+{
+  if (!sSingleton) {
+    return;
+  }
+
+  sSingleton = nullptr;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/speechd/SpeechDispatcherService.h

@@ -0,0 +1,67 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim: set ts=8 sts=2 et sw=2 tw=80: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef mozilla_dom_SpeechDispatcherService_h
+#define mozilla_dom_SpeechDispatcherService_h
+
+#include "mozilla/StaticPtr.h"
+#include "nsIObserver.h"
+#include "nsISpeechService.h"
+#include "nsIThread.h"
+#include "nsRefPtrHashtable.h"
+#include "nsTArray.h"
+
+struct SPDConnection;
+
+namespace mozilla {
+namespace dom {
+
+class SpeechDispatcherCallback;
+class SpeechDispatcherVoice;
+
+class SpeechDispatcherService final : public nsIObserver,
+                                      public nsISpeechService
+{
+  friend class SpeechDispatcherCallback;
+public:
+  NS_DECL_THREADSAFE_ISUPPORTS
+  NS_DECL_NSIOBSERVER
+  NS_DECL_NSISPEECHSERVICE
+
+  SpeechDispatcherService();
+
+  void Init();
+
+  void Setup();
+
+  void EventNotify(uint32_t aMsgId, uint32_t aState);
+
+  static SpeechDispatcherService* GetInstance(bool create = true);
+  static already_AddRefed<SpeechDispatcherService> GetInstanceForService();
+
+  static void Shutdown();
+
+  static StaticRefPtr<SpeechDispatcherService> sSingleton;
+
+private:
+  virtual ~SpeechDispatcherService();
+
+  void RegisterVoices();
+
+  bool mInitialized;
+
+  SPDConnection* mSpeechdClient;
+
+  nsRefPtrHashtable<nsUint32HashKey, SpeechDispatcherCallback> mCallbacks;
+
+  nsCOMPtr<nsIThread> mInitThread;
+
+  nsRefPtrHashtable<nsStringHashKey, SpeechDispatcherVoice> mVoices;
+};
+
+} // namespace dom
+} // namespace mozilla
+#endif

dom/media/webspeech/synth/speechd/moz.build

@@ -0,0 +1,13 @@
+# -*- Mode: python; indent-tabs-mode: nil; tab-width: 40 -*-
+# vim: set filetype=python:
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+UNIFIED_SOURCES += [
+  'SpeechDispatcherModule.cpp',
+  'SpeechDispatcherService.cpp'
+]
+include('/ipc/chromium/chromium-config.mozbuild')
+
+FINAL_LIBRARY = 'xul'

dom/media/webspeech/synth/test/FakeSynthModule.cpp

@@ -0,0 +1,55 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "mozilla/ModuleUtils.h"
+#include "nsIClassInfoImpl.h"
+
+#include "nsFakeSynthServices.h"
+
+using namespace mozilla::dom;
+
+#define FAKESYNTHSERVICE_CID \
+  {0xe7d52d9e, 0xc148, 0x47d8, {0xab, 0x2a, 0x95, 0xd7, 0xf4, 0x0e, 0xa5, 0x3d}}
+
+#define FAKESYNTHSERVICE_CONTRACTID "@mozilla.org/fakesynth;1"
+
+// Defines nsFakeSynthServicesConstructor
+NS_GENERIC_FACTORY_SINGLETON_CONSTRUCTOR(nsFakeSynthServices,
+                                         nsFakeSynthServices::GetInstanceForService)
+
+// Defines kFAKESYNTHSERVICE_CID
+NS_DEFINE_NAMED_CID(FAKESYNTHSERVICE_CID);
+
+static const mozilla::Module::CIDEntry kCIDs[] = {
+  { &kFAKESYNTHSERVICE_CID, true, nullptr, nsFakeSynthServicesConstructor },
+  { nullptr }
+};
+
+static const mozilla::Module::ContractIDEntry kContracts[] = {
+  { FAKESYNTHSERVICE_CONTRACTID, &kFAKESYNTHSERVICE_CID },
+  { nullptr }
+};
+
+static const mozilla::Module::CategoryEntry kCategories[] = {
+  { "speech-synth-started", "Fake Speech Synth", FAKESYNTHSERVICE_CONTRACTID },
+  { nullptr }
+};
+
+static void
+UnloadFakeSynthmodule()
+{
+  nsFakeSynthServices::Shutdown();
+}
+
+static const mozilla::Module kModule = {
+  mozilla::Module::kVersion,
+  kCIDs,
+  kContracts,
+  kCategories,
+  nullptr,
+  nullptr,
+  UnloadFakeSynthmodule
+};
+
+NSMODULE_DEFN(fakesynth) = &kModule;

dom/media/webspeech/synth/test/common.js

@@ -0,0 +1,91 @@
+function synthTestQueue(aTestArgs, aEndFunc) {
+  var utterances = [];
+  for (var i in aTestArgs) {
+    var uargs = aTestArgs[i][0];
+    var win = uargs.win || window;
+    var u = new win.SpeechSynthesisUtterance(uargs.text);
+
+    if (uargs.args) {
+      for (var attr in uargs.args)
+        u[attr] = uargs.args[attr];
+    }
+
+    function onend_handler(e) {
+      is(e.target, utterances.shift(), "Target matches utterances");
+      ok(!speechSynthesis.speaking, "speechSynthesis is not speaking.");
+
+      if (utterances.length) {
+        ok(speechSynthesis.pending, "other utterances queued");
+      } else {
+        ok(!speechSynthesis.pending, "queue is empty, nothing pending.");
+        if (aEndFunc)
+          aEndFunc();
+      }
+    }
+
+    u.addEventListener('start',
+      (function (expectedUri) {
+        return function (e) {
+          if (expectedUri) {
+            var chosenVoice = SpecialPowers.wrap(e).target.chosenVoiceURI;
+            is(chosenVoice, expectedUri, "Incorrect URI is used");
+          }
+        };
+      })(aTestArgs[i][1] ? aTestArgs[i][1].uri : null));
+
+    u.addEventListener('end', onend_handler);
+    u.addEventListener('error', onend_handler);
+
+    u.addEventListener('error',
+      (function (expectedError) {
+        return function onerror_handler(e) {
+          ok(expectedError, "Error in speech utterance '" + e.target.text + "'");
+        };
+      })(aTestArgs[i][1] ? aTestArgs[i][1].err : false));
+
+    utterances.push(u);
+    win.speechSynthesis.speak(u);
+  }
+
+  ok(!speechSynthesis.speaking, "speechSynthesis is not speaking yet.");
+  ok(speechSynthesis.pending, "speechSynthesis has an utterance queued.");
+}
+
+function loadFrame(frameId) {
+  return new Promise(function(resolve, reject) {
+    var frame = document.getElementById(frameId);
+    frame.addEventListener('load', function (e) {
+      frame.contentWindow.document.title = frameId;
+      resolve(frame);
+    });
+    frame.src = 'data:text/html,' + encodeURI('<html><head></head><body></body></html>');
+  });
+}
+
+function waitForVoices(win) {
+  return new Promise(resolve => {
+    function resolver() {
+      if (win.speechSynthesis.getVoices().length) {
+        win.speechSynthesis.removeEventListener('voiceschanged', resolver);
+        resolve();
+      }
+    }
+
+    win.speechSynthesis.addEventListener('voiceschanged', resolver);
+    resolver();
+  });
+}
+
+function loadSpeechTest(fileName, prefs, frameId="testFrame") {
+  loadFrame(frameId).then(frame => {
+    waitForVoices(frame.contentWindow).then(
+      () => document.getElementById("testFrame").src = fileName);
+  });
+}
+
+function testSynthState(win, expectedState) {
+  for (var attr in expectedState) {
+    is(win.speechSynthesis[attr], expectedState[attr],
+      win.document.title + ": '" + attr + '" does not match');
+  }
+}

dom/media/webspeech/synth/test/file_bfcache_frame.html

@@ -0,0 +1,28 @@
+<!DOCTYPE HTML>
+<html>
+<head>
+  <meta charset="utf-8">
+  <script type="application/javascript">
+    var frameUnloaded = function() {
+      var u = new SpeechSynthesisUtterance('hi');
+      u.addEventListener('end', function () {
+        parent.ok(true, 'Successfully spoke utterance from new frame.');
+        parent.onDone();
+      });
+      speechSynthesis.speak(u);
+    }
+    addEventListener('pageshow', function onshow(evt) {
+      var u = new SpeechSynthesisUtterance('hello');
+      u.lang = 'it-IT-noend';
+      u.addEventListener('start', function() {
+        location =
+          'data:text/html,<html><body onpageshow="' +
+          frameUnloaded.toSource() + '()"></body></html>';
+      });
+      speechSynthesis.speak(u);
+    });
+  </script>
+</head>
+<body>
+</body>
+</html>

dom/media/webspeech/synth/test/file_global_queue.html

@@ -0,0 +1,69 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=1188099
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 1188099: Global queue should correctly schedule utterances</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.info = parent.info;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+    window.todo = parent.todo;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=1188099">Mozilla Bug 1188099</a>
+<iframe id="frame1"></iframe>
+<iframe id="frame2"></iframe>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="application/javascript">
+  Promise.all([loadFrame('frame1'), loadFrame('frame2')]).then(function ([frame1, frame2]) {
+    var win1 = frame1.contentWindow;
+    var win2 = frame2.contentWindow;
+    var utterance1 = new win1.SpeechSynthesisUtterance("hello, losers");
+    var utterance2 = new win1.SpeechSynthesisUtterance("hello, losers three");
+    var utterance3 = new win2.SpeechSynthesisUtterance("hello, losers too");
+    var eventOrder = ['start1', 'end1', 'start3', 'end3', 'start2', 'end2'];
+    utterance1.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start1', 'start1');
+      testSynthState(win1, { speaking: true, pending: true });
+      testSynthState(win2, { speaking: true, pending: true });
+    });
+    utterance1.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end1', 'end1');
+    });
+    utterance3.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start3', 'start3');
+      testSynthState(win1, { speaking: true, pending: true });
+      testSynthState(win2, { speaking: true, pending: false });
+    });
+    utterance3.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end3', 'end3');
+    });
+    utterance2.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start2', 'start2');
+      testSynthState(win1, { speaking: true, pending: false });
+      testSynthState(win2, { speaking: true, pending: false });
+    });
+    utterance2.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end2', 'end2');
+      testSynthState(win1, { speaking: false, pending: false });
+      testSynthState(win2, { speaking: false, pending: false });
+      SimpleTest.finish();
+    });
+    win1.speechSynthesis.speak(utterance1);
+    win1.speechSynthesis.speak(utterance2);
+    win2.speechSynthesis.speak(utterance3);
+  });
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/file_global_queue_cancel.html

@@ -0,0 +1,88 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=1188099
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 1188099: Calling cancel() should work correctly with global queue</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.info = parent.info;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+    window.todo = parent.todo;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=1188099">Mozilla Bug 1188099</a>
+<iframe id="frame1"></iframe>
+<iframe id="frame2"></iframe>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="application/javascript">
+  Promise.all([loadFrame('frame1'), loadFrame('frame2')]).then(function ([frame1, frame2]) {
+    var win1 = frame1.contentWindow;
+    var win2 = frame2.contentWindow;
+
+    var utterance1 = new win1.SpeechSynthesisUtterance(
+      "u1: Donec ac nunc feugiat, posuere");
+    utterance1.lang = 'it-IT-noend';
+    var utterance2 = new win1.SpeechSynthesisUtterance("u2: hello, losers too");
+    utterance2.lang = 'it-IT-noend';
+    var utterance3 = new win1.SpeechSynthesisUtterance("u3: hello, losers three");
+
+    var utterance4 = new win2.SpeechSynthesisUtterance("u4: hello, losers same!");
+    utterance4.lang = 'it-IT-noend';
+    var utterance5 = new win2.SpeechSynthesisUtterance("u5: hello, losers too");
+    utterance5.lang = 'it-IT-noend';
+
+    var eventOrder = ['start1', 'end1', 'start2', 'end2'];
+    utterance1.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start1', 'start1');
+      testSynthState(win1, { speaking: true, pending: true });
+      testSynthState(win2, { speaking: true, pending: true });
+      win2.speechSynthesis.cancel();
+      SpecialPowers.wrap(win1.speechSynthesis).forceEnd();
+
+    });
+    utterance1.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end1', 'end1');
+      testSynthState(win1, { pending: true });
+      testSynthState(win2, { pending: false });
+    });
+    utterance2.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start2', 'start2');
+      testSynthState(win1, { speaking: true, pending: true });
+      testSynthState(win2, { speaking: true, pending: false });
+      win1.speechSynthesis.cancel();
+    });
+    utterance2.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end2', 'end2');
+      testSynthState(win1, { speaking: false, pending: false });
+      testSynthState(win2, { speaking: false, pending: false });
+      SimpleTest.finish();
+    });
+
+    function wrongUtterance(e) {
+      ok(false, 'This shall not be uttered: "' + e.target.text + '"');
+    }
+
+    utterance3.addEventListener('start', wrongUtterance);
+    utterance4.addEventListener('start', wrongUtterance);
+    utterance5.addEventListener('start', wrongUtterance);
+
+    win1.speechSynthesis.speak(utterance1);
+    win1.speechSynthesis.speak(utterance2);
+    win1.speechSynthesis.speak(utterance3);
+    win2.speechSynthesis.speak(utterance4);
+    win2.speechSynthesis.speak(utterance5);
+  });
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/file_global_queue_pause.html

@@ -0,0 +1,131 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=1188099
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 1188099: Calling pause() should work correctly with global queue</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.info = parent.info;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+    window.todo = parent.todo;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=1188099">Mozilla Bug 1188099</a>
+<iframe id="frame1"></iframe>
+<iframe id="frame2"></iframe>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="application/javascript">
+  Promise.all([loadFrame('frame1'), loadFrame('frame2')]).then(function ([frame1, frame2]) {
+    var win1 = frame1.contentWindow;
+    var win2 = frame2.contentWindow;
+
+    var utterance1 = new win1.SpeechSynthesisUtterance("Speak utterance 1.");
+    utterance1.lang = 'it-IT-noend';
+    var utterance2 = new win2.SpeechSynthesisUtterance("Speak utterance 2.");
+    var utterance3 = new win1.SpeechSynthesisUtterance("Speak utterance 3.");
+    var utterance4 = new win2.SpeechSynthesisUtterance("Speak utterance 4.");
+    var eventOrder = ['start1', 'pause1', 'resume1', 'end1', 'start2', 'end2',
+      'start4', 'end4', 'start3', 'end3'];
+
+    utterance1.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start1', 'start1');
+      win1.speechSynthesis.pause();
+    });
+    utterance1.addEventListener('pause', function(e) {
+      var expectedEvent = eventOrder.shift()
+      is(expectedEvent, 'pause1', 'pause1');
+      testSynthState(win1, { speaking: true, pending: false, paused: true});
+      testSynthState(win2, { speaking: true, pending: true, paused: false});
+
+      if (expectedEvent == 'pause1') {
+        win1.speechSynthesis.resume();
+      }
+    });
+    utterance1.addEventListener('resume', function(e) {
+      is(eventOrder.shift(), 'resume1', 'resume1');
+      testSynthState(win1, { speaking: true, pending: false, paused: false});
+      testSynthState(win2, { speaking: true, pending: true, paused: false});
+
+      win2.speechSynthesis.pause();
+
+      testSynthState(win1, { speaking: true, pending: false, paused: false});
+      // 1188099: currently, paused state is not gaurenteed to be immediate.
+      testSynthState(win2, { speaking: true, pending: true });
+
+      // We now make the utterance end.
+      SpecialPowers.wrap(win1.speechSynthesis).forceEnd();
+    });
+    utterance1.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end1', 'end1');
+      testSynthState(win1, { speaking: false, pending: false, paused: false});
+      testSynthState(win2, { speaking: false, pending: true, paused: true});
+
+      win2.speechSynthesis.resume();
+    });
+
+    utterance2.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start2', 'start2');
+      testSynthState(win1, { speaking: true, pending: false, paused: false});
+      testSynthState(win2, { speaking: true, pending: false, paused: false});
+    });
+    utterance2.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end2', 'end2');
+      testSynthState(win1, { speaking: false, pending: false, paused: false});
+      testSynthState(win2, { speaking: false, pending: false, paused: false});
+
+      win1.speechSynthesis.pause();
+
+      testSynthState(win1, { speaking: false, pending: false, paused: true});
+      testSynthState(win2, { speaking: false, pending: false, paused: false});
+
+      win1.speechSynthesis.speak(utterance3);
+      win2.speechSynthesis.speak(utterance4);
+
+      testSynthState(win1, { speaking: false, pending: true, paused: true});
+      testSynthState(win2, { speaking: false, pending: true, paused: false});
+    });
+
+    utterance4.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start4', 'start4');
+      testSynthState(win1, { speaking: true, pending: true, paused: true});
+      testSynthState(win2, { speaking: true, pending: false, paused: false});
+
+      win1.speechSynthesis.resume();
+    });
+    utterance4.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end4', 'end4');
+      testSynthState(win1, { speaking: false, pending: true, paused: false});
+      testSynthState(win2, { speaking: false, pending: false, paused: false});
+    });
+
+    utterance3.addEventListener('start', function(e) {
+      is(eventOrder.shift(), 'start3', 'start3');
+      testSynthState(win1, { speaking: true, pending: false, paused: false});
+      testSynthState(win2, { speaking: true, pending: false, paused: false});
+    });
+
+    utterance3.addEventListener('end', function(e) {
+      is(eventOrder.shift(), 'end3', 'end3');
+      testSynthState(win1, { speaking: false, pending: false, paused: false});
+      testSynthState(win2, { speaking: false, pending: false, paused: false});
+
+      SimpleTest.finish();
+    });
+
+    win1.speechSynthesis.speak(utterance1);
+    win2.speechSynthesis.speak(utterance2);
+  });
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/file_indirect_service_events.html

@@ -0,0 +1,102 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=1155034
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 1155034: Check that indirect audio services dispatch their own events</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.info = parent.info;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=1155034">Mozilla Bug 1155034</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="application/javascript">
+
+/** Test for Bug 1155034 **/
+
+function testFunc(done_cb) {
+  function test_with_events() {
+    info('test_with_events');
+    var utterance = new SpeechSynthesisUtterance("never end, callback events");
+    utterance.lang = 'it-IT-noend';
+
+    utterance.addEventListener('start', function(e) {
+      info('start test_with_events');
+      speechSynthesis.pause();
+    // Wait to see if we get some bad events we didn't expect.
+    });
+
+    utterance.addEventListener('pause', function(e) {
+      is(e.charIndex, 1, 'pause event charIndex matches service arguments');
+      is(e.elapsedTime, 1.5, 'pause event elapsedTime matches service arguments');
+      speechSynthesis.resume();
+    });
+
+    utterance.addEventListener('resume', function(e) {
+      is(e.charIndex, 1, 'resume event charIndex matches service arguments');
+      is(e.elapsedTime, 1.5, 'resume event elapsedTime matches service arguments');
+      speechSynthesis.cancel();
+    });
+
+    utterance.addEventListener('end', function(e) {
+      ok(e.charIndex, 1, 'resume event charIndex matches service arguments');
+      ok(e.elapsedTime, 1.5, 'end event elapsedTime matches service arguments');
+      test_no_events();
+    });
+
+    info('start speak');
+    speechSynthesis.speak(utterance);
+  }
+
+  function forbiddenEvent(e) {
+    ok(false, 'no "' + e.type + '" event was explicitly dispatched from the service')
+  }
+
+  function test_no_events() {
+    info('test_no_events');
+    var utterance = new SpeechSynthesisUtterance("never end");
+    utterance.lang = "it-IT-noevents-noend";
+    utterance.addEventListener('start', function(e) {
+      speechSynthesis.pause();
+      // Wait to see if we get some bad events we didn't expect.
+      setTimeout(function() {
+        ok(true, 'didn\'t get any unwanted events');
+        utterance.removeEventListener('end', forbiddenEvent);
+        SpecialPowers.wrap(speechSynthesis).forceEnd();
+        done_cb();
+      }, 1000);
+    });
+
+    utterance.addEventListener('pause', forbiddenEvent);
+    utterance.addEventListener('end', forbiddenEvent);
+
+    speechSynthesis.speak(utterance);
+  }
+
+  test_with_events();
+}
+
+// Run test with no global queue, and then run it with a global queue.
+testFunc(function() {
+  SpecialPowers.pushPrefEnv(
+    { set: [['media.webspeech.synth.force_global_queue', true]] }, function() {
+      testFunc(SimpleTest.finish)
+    });
+});
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/file_setup.html

@@ -0,0 +1,95 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=525444
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 525444: Web Speech API check all classes are present</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+  
+</div>
+<pre id="test">
+<script type="application/javascript">
+
+/** Test for Bug 525444 **/
+
+ok(SpeechSynthesis, "SpeechSynthesis exists in global scope");
+ok(SpeechSynthesisVoice, "SpeechSynthesisVoice exists in global scope");
+ok(SpeechSynthesisErrorEvent, "SpeechSynthesisErrorEvent exists in global scope");
+ok(SpeechSynthesisEvent, "SpeechSynthesisEvent exists in global scope");
+
+// SpeechSynthesisUtterance is the only type that has a constructor
+//  and writable properties
+ok(SpeechSynthesisUtterance, "SpeechSynthesisUtterance exists in global scope");
+var ssu = new SpeechSynthesisUtterance("hello world");
+is(typeof ssu, "object", "SpeechSynthesisUtterance instance is an object");
+is(ssu.text, "hello world", "SpeechSynthesisUtterance.text is correct");
+is(ssu.volume, 1, "SpeechSynthesisUtterance.volume default is correct");
+is(ssu.rate, 1, "SpeechSynthesisUtterance.rate default is correct");
+is(ssu.pitch, 1, "SpeechSynthesisUtterance.pitch default is correct");
+ssu.lang = "he-IL";
+ssu.volume = 0.5;
+ssu.rate = 2.0;
+ssu.pitch = 1.5;
+is(ssu.lang, "he-IL", "SpeechSynthesisUtterance.lang is correct");
+is(ssu.volume, 0.5,  "SpeechSynthesisUtterance.volume is correct");
+is(ssu.rate, 2.0,  "SpeechSynthesisUtterance.rate is correct");
+is(ssu.pitch, 1.5,  "SpeechSynthesisUtterance.pitch is correct");
+
+// Assign a rate that is out of bounds
+ssu.rate = 20;
+is(ssu.rate, 10, "SpeechSynthesisUtterance.rate enforces max of 10");
+ssu.rate = 0;
+is(ssu.rate.toPrecision(1), "0.1", "SpeechSynthesisUtterance.rate enforces min of 0.1");
+
+// Assign a volume which is out of bounds
+ssu.volume = 2;
+is(ssu.volume, 1, "SpeechSynthesisUtterance.volume enforces max of 1");
+ssu.volume = -1;
+is(ssu.volume, 0, "SpeechSynthesisUtterance.volume enforces min of 0");
+
+// Assign a pitch which is out of bounds
+ssu.pitch = 2.1;
+is(ssu.pitch, 2, "SpeechSynthesisUtterance.pitch enforces max of 2");
+ssu.pitch = -1;
+is(ssu.pitch, 0, "SpeechSynthesisUtterance.pitch enforces min of 0");
+
+// Test for singleton instance hanging off of window.
+ok(speechSynthesis, "speechSynthesis exists in global scope");
+is(typeof speechSynthesis, "object", "speechSynthesis instance is an object");
+is(typeof speechSynthesis.speak, "function", "speechSynthesis.speak is a function");
+is(typeof speechSynthesis.cancel, "function", "speechSynthesis.cancel is a function");
+is(typeof speechSynthesis.pause, "function", "speechSynthesis.pause is a function");
+is(typeof speechSynthesis.resume, "function", "speechSynthesis.resume is a function");
+is(typeof speechSynthesis.getVoices, "function", "speechSynthesis.getVoices is a function");
+
+is(typeof speechSynthesis.pending, "boolean", "speechSynthesis.pending is a boolean");
+is(typeof speechSynthesis.speaking, "boolean", "speechSynthesis.speaking is a boolean");
+is(typeof speechSynthesis.paused, "boolean", "speechSynthesis.paused is a boolean");
+
+var voices1 = speechSynthesis.getVoices();
+var voices2 = speechSynthesis.getVoices();
+
+ok(voices1.length == voices2.length, "Voice count matches");
+
+for (var i in voices1) {
+  ok(voices1[i] == voices2[i], "Voice instance matches");
+}
+
+SimpleTest.finish();
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/file_speech_cancel.html

@@ -0,0 +1,100 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=1150315
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 1150315: Check that successive cancel/speak calls work</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.info = parent.info;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=1150315">Mozilla Bug 1150315</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="application/javascript">
+
+/** Test for Bug 1150315 **/
+
+function testFunc(done_cb) {
+  var gotEndEvent = false;
+  // A long utterance that we will interrupt.
+  var utterance = new SpeechSynthesisUtterance("Donec ac nunc feugiat, posuere " +
+    "mauris id, pharetra velit. Donec fermentum orci nunc, sit amet maximus" +
+    "dui tincidunt ut. Sed ultricies ac nisi a laoreet. Proin interdum," +
+    "libero maximus hendrerit posuere, lorem risus egestas nisl, a" +
+    "ultricies massa justo eu nisi. Duis mattis nibh a ligula tincidunt" +
+    "tincidunt non eu erat. Sed bibendum varius vulputate. Cras leo magna," +
+    "ornare ac posuere vel, luctus id metus. Mauris nec quam ac augue" +
+    "consectetur bibendum. Integer a commodo tortor. Duis semper dolor eu" +
+    "facilisis facilisis. Etiam venenatis turpis est, quis tincidunt velit" +
+    "suscipit a. Cras semper orci in sapien rhoncus bibendum. Suspendisse" +
+    "eu ex lobortis, finibus enim in, condimentum quam. Maecenas eget dui" +
+    "ipsum. Aliquam tortor leo, interdum eget congue ut, tempor id elit.");
+  utterance.addEventListener('start', function(e) {
+    ok(true, 'start utterance 1');
+    speechSynthesis.cancel();
+    info('cancel!');
+    speechSynthesis.speak(utterance2);
+    info('speak??');
+  });
+
+  var utterance2 = new SpeechSynthesisUtterance("Proin ornare neque vitae " +
+    "risus mattis rutrum. Suspendisse a velit ut est convallis aliquet." +
+    "Nullam ante elit, malesuada vel luctus rutrum, ultricies nec libero." +
+    "Praesent eu iaculis orci. Sed nisl diam, sodales ac purus et," +
+    "volutpat interdum tortor. Nullam aliquam porta elit et maximus. Cras" +
+    "risus lectus, elementum vel sodales vel, ultricies eget lectus." +
+    "Curabitur velit lacus, mollis vel finibus et, molestie sit amet" +
+    "sapien. Proin vitae dolor ac augue posuere efficitur ac scelerisque" +
+    "diam. Nulla sed odio elit.");
+  utterance2.addEventListener('start', function() {
+    info('start');
+    speechSynthesis.cancel();
+    speechSynthesis.speak(utterance3);
+  });
+  utterance2.addEventListener('end', function(e) {
+    gotEndEvent = true;
+  });
+
+  var utterance3 = new SpeechSynthesisUtterance("Hello, world 3!");
+  utterance3.addEventListener('start', function() {
+    ok(gotEndEvent, "didn't get start event for this utterance");
+  });
+  utterance3.addEventListener('end', done_cb);
+
+  // Speak/cancel while paused (Bug 1187105)
+  speechSynthesis.pause();
+  speechSynthesis.speak(new SpeechSynthesisUtterance("hello."));
+  ok(speechSynthesis.pending, "paused speechSynthesis has an utterance queued.");
+  speechSynthesis.cancel();
+  ok(!speechSynthesis.pending, "paused speechSynthesis has no utterance queued.");
+  speechSynthesis.resume();
+
+  speechSynthesis.speak(utterance);
+  ok(!speechSynthesis.speaking, "speechSynthesis is not speaking yet.");
+  ok(speechSynthesis.pending, "speechSynthesis has an utterance queued.");
+}
+
+// Run test with no global queue, and then run it with a global queue.
+testFunc(function() {
+  SpecialPowers.pushPrefEnv(
+    { set: [['media.webspeech.synth.force_global_queue', true]] }, function() {
+      testFunc(SimpleTest.finish)
+    });
+});
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/file_speech_error.html

@@ -0,0 +1,46 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=1226015
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 1226015</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.info = parent.info;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=1226015">Mozilla Bug 1226015</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="application/javascript">
+
+/** Test for Bug 1226015 **/
+
+function testFunc(done_cb) {
+  var utterance = new SpeechSynthesisUtterance();
+  utterance.lang = 'it-IT-failatstart';
+
+  speechSynthesis.speak(utterance);
+  speechSynthesis.cancel();
+
+  ok(true, "we didn't crash, that is good.")
+  SimpleTest.finish();
+}
+
+// Run test with no global queue, and then run it with a global queue.
+testFunc();
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/file_speech_queue.html

@@ -0,0 +1,85 @@
+<!DOCTYPE HTML>
+<html lang="en-US">
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=525444
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 525444: Web Speech API, check speech synth queue</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=525444">Mozilla Bug 525444</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+
+</div>
+<pre id="test">
+<script type="application/javascript">
+
+/** Test for Bug 525444 **/
+
+// XXX: Rate and pitch are not tested.
+
+var langUriMap = {};
+
+for (var voice of speechSynthesis.getVoices()) {
+  langUriMap[voice.lang] = voice.voiceURI;
+  ok(true, voice.lang + ' ' + voice.voiceURI + ' ' + voice.default);
+  is(voice.default, voice.lang == 'en-JM', 'Only Jamaican voice should be default');
+}
+
+ok(langUriMap['en-JM'], 'No English-Jamaican voice');
+ok(langUriMap['en-GB'], 'No English-British voice');
+ok(langUriMap['en-CA'], 'No English-Canadian voice');
+ok(langUriMap['fr-CA'], 'No French-Canadian voice');
+ok(langUriMap['es-MX'], 'No Spanish-Mexican voice');
+ok(langUriMap['it-IT-fail'], 'No Failing Italian voice');
+
+function testFunc(done_cb) {
+  synthTestQueue(
+    [[{text: "Hello, world."},
+      { uri: langUriMap['en-JM'] }],
+     [{text: "Bonjour tout le monde .",
+       args: { lang: "fr", rate: 0.5, pitch: 0.75 }},
+      { uri: langUriMap['fr-CA'], rate: 0.5, pitch: 0.75}],
+     [{text: "How are you doing?", args: { lang: "en-GB" } },
+      { rate: 1, pitch: 1, uri: langUriMap['en-GB']}],
+     [{text: "Come stai?", args: { lang: "it-IT-fail" } },
+      { rate: 1, pitch: 1, uri: langUriMap['it-IT-fail'], err: true }],
+     [{text: "¡hasta mañana!", args: { lang: "es-MX" } },
+      { uri: langUriMap['es-MX'] }]],
+    function () {
+      var test_data = [];
+      var voices = speechSynthesis.getVoices();
+      for (var voice of voices) {
+        if (voice.voiceURI.indexOf('urn:moz-tts:fake-direct') < 0) {
+          continue;
+        }
+        test_data.push([{text: "Hello world", args: { voice: voice} },
+                        {uri: voice.voiceURI}]);
+      }
+
+      synthTestQueue(test_data, done_cb);
+    });
+}
+
+// Run test with no global queue, and then run it with a global queue.
+testFunc(function() {
+  SpecialPowers.pushPrefEnv(
+    { set: [['media.webspeech.synth.force_global_queue', true]] }, function() {
+      testFunc(SimpleTest.finish)
+    });
+});
+
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/file_speech_simple.html

@@ -0,0 +1,53 @@
+<!DOCTYPE HTML>
+<html>
+<!--
+https://bugzilla.mozilla.org/show_bug.cgi?id=650295
+-->
+<head>
+  <meta charset="utf-8">
+  <title>Test for Bug 650295: Web Speech API check all classes are present</title>
+  <script type="application/javascript">
+    window.SimpleTest = parent.SimpleTest;
+    window.info = parent.info;
+    window.is = parent.is;
+    window.isnot = parent.isnot;
+    window.ok = parent.ok;
+  </script>
+  <script type="application/javascript" src="common.js"></script>
+</head>
+<body>
+<a target="_blank" href="https://bugzilla.mozilla.org/show_bug.cgi?id=650295">Mozilla Bug 650295</a>
+<p id="display"></p>
+<div id="content" style="display: none">
+  
+</div>
+<pre id="test">
+<script type="application/javascript">
+
+/** Test for Bug 525444 **/
+
+var gotStartEvent = false;
+var gotBoundaryEvent = false;
+var utterance = new SpeechSynthesisUtterance("Hello, world!");
+utterance.addEventListener('start', function(e) {
+  ok(speechSynthesis.speaking, "speechSynthesis is speaking.");
+  ok(!speechSynthesis.pending, "speechSynthesis has no other utterances queued.");
+  gotStartEvent = true;
+});
+
+utterance.addEventListener('end', function(e) {
+  ok(!speechSynthesis.speaking, "speechSynthesis is not speaking.");
+  ok(!speechSynthesis.pending, "speechSynthesis has no other utterances queued.");
+  ok(gotStartEvent, "Got 'start' event.");
+  info('end ' + e.elapsedTime);
+  SimpleTest.finish();
+});
+
+speechSynthesis.speak(utterance);
+ok(!speechSynthesis.speaking, "speechSynthesis is not speaking yet.");
+ok(speechSynthesis.pending, "speechSynthesis has an utterance queued.");
+
+</script>
+</pre>
+</body>
+</html>

dom/media/webspeech/synth/test/mochitest.ini

@@ -0,0 +1,26 @@
+[DEFAULT]
+tags=msg
+subsuite = media
+support-files =
+  common.js
+  file_bfcache_frame.html
+  file_setup.html
+  file_speech_queue.html
+  file_speech_simple.html
+  file_speech_cancel.html
+  file_speech_error.html
+  file_indirect_service_events.html
+  file_global_queue.html
+  file_global_queue_cancel.html
+  file_global_queue_pause.html
+
+[test_setup.html]
+[test_speech_queue.html]
+[test_speech_simple.html]
+[test_speech_cancel.html]
+[test_speech_error.html]
+[test_indirect_service_events.html]
+[test_global_queue.html]
+[test_global_queue_cancel.html]
+[test_global_queue_pause.html]
+[test_bfcache.html]

dom/media/webspeech/synth/test/nsFakeSynthServices.cpp

@@ -0,0 +1,401 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "nsISupports.h"
+#include "nsFakeSynthServices.h"
+#include "nsPrintfCString.h"
+#include "nsIWeakReferenceUtils.h"
+#include "SharedBuffer.h"
+#include "nsISimpleEnumerator.h"
+
+#include "mozilla/dom/nsSynthVoiceRegistry.h"
+#include "mozilla/dom/nsSpeechTask.h"
+
+#include "nsThreadUtils.h"
+#include "prenv.h"
+#include "mozilla/Preferences.h"
+#include "mozilla/DebugOnly.h"
+
+#define CHANNELS 1
+#define SAMPLERATE 1600
+
+namespace mozilla {
+namespace dom {
+
+StaticRefPtr<nsFakeSynthServices> nsFakeSynthServices::sSingleton;
+
+enum VoiceFlags
+{
+  eSuppressEvents = 1,
+  eSuppressEnd = 2,
+  eFailAtStart = 4,
+  eFail = 8
+};
+
+struct VoiceDetails
+{
+  const char* uri;
+  const char* name;
+  const char* lang;
+  bool defaultVoice;
+  uint32_t flags;
+};
+
+static const VoiceDetails sDirectVoices[] = {
+  {"urn:moz-tts:fake-direct:bob", "Bob Marley", "en-JM", true, 0},
+  {"urn:moz-tts:fake-direct:amy", "Amy Winehouse", "en-GB", false, 0},
+  {"urn:moz-tts:fake-direct:lenny", "Leonard Cohen", "en-CA", false, 0},
+  {"urn:moz-tts:fake-direct:celine", "Celine Dion", "fr-CA", false, 0},
+  {"urn:moz-tts:fake-direct:julie", "Julieta Venegas", "es-MX", false, },
+};
+
+static const VoiceDetails sIndirectVoices[] = {
+  {"urn:moz-tts:fake-indirect:zanetta", "Zanetta Farussi", "it-IT", false, 0},
+  {"urn:moz-tts:fake-indirect:margherita", "Margherita Durastanti", "it-IT-noevents-noend", false, eSuppressEvents | eSuppressEnd},
+  {"urn:moz-tts:fake-indirect:teresa", "Teresa Cornelys", "it-IT-noend", false, eSuppressEnd},
+  {"urn:moz-tts:fake-indirect:cecilia", "Cecilia Bartoli", "it-IT-failatstart", false, eFailAtStart},
+  {"urn:moz-tts:fake-indirect:gottardo", "Gottardo Aldighieri", "it-IT-fail", false, eFail},
+};
+
+// FakeSynthCallback
+class FakeSynthCallback : public nsISpeechTaskCallback
+{
+public:
+  explicit FakeSynthCallback(nsISpeechTask* aTask) : mTask(aTask) { }
+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
+  NS_DECL_CYCLE_COLLECTION_CLASS_AMBIGUOUS(FakeSynthCallback, nsISpeechTaskCallback)
+
+  NS_IMETHOD OnPause() override
+  {
+    if (mTask) {
+      mTask->DispatchPause(1.5, 1);
+    }
+
+    return NS_OK;
+  }
+
+  NS_IMETHOD OnResume() override
+  {
+    if (mTask) {
+      mTask->DispatchResume(1.5, 1);
+    }
+
+    return NS_OK;
+  }
+
+  NS_IMETHOD OnCancel() override
+  {
+    if (mTask) {
+      mTask->DispatchEnd(1.5, 1);
+    }
+
+    return NS_OK;
+  }
+
+  NS_IMETHOD OnVolumeChanged(float aVolume) override
+  {
+    return NS_OK;
+  }
+
+private:
+  virtual ~FakeSynthCallback() { }
+
+  nsCOMPtr<nsISpeechTask> mTask;
+};
+
+NS_IMPL_CYCLE_COLLECTION(FakeSynthCallback, mTask);
+
+NS_INTERFACE_MAP_BEGIN_CYCLE_COLLECTION(FakeSynthCallback)
+  NS_INTERFACE_MAP_ENTRY(nsISpeechTaskCallback)
+  NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsISpeechTaskCallback)
+NS_INTERFACE_MAP_END
+
+NS_IMPL_CYCLE_COLLECTING_ADDREF(FakeSynthCallback)
+NS_IMPL_CYCLE_COLLECTING_RELEASE(FakeSynthCallback)
+
+// FakeDirectAudioSynth
+
+class FakeDirectAudioSynth : public nsISpeechService
+{
+
+public:
+  FakeDirectAudioSynth() { }
+
+  NS_DECL_ISUPPORTS
+  NS_DECL_NSISPEECHSERVICE
+
+private:
+  virtual ~FakeDirectAudioSynth() { }
+};
+
+NS_IMPL_ISUPPORTS(FakeDirectAudioSynth, nsISpeechService)
+
+NS_IMETHODIMP
+FakeDirectAudioSynth::Speak(const nsAString& aText, const nsAString& aUri,
+                            float aVolume, float aRate, float aPitch,
+                            nsISpeechTask* aTask)
+{
+  class Runnable final : public mozilla::Runnable
+  {
+  public:
+    Runnable(nsISpeechTask* aTask, const nsAString& aText) :
+      mTask(aTask), mText(aText)
+    {
+    }
+
+    NS_IMETHOD Run() override
+    {
+      RefPtr<FakeSynthCallback> cb = new FakeSynthCallback(nullptr);
+      mTask->Setup(cb, CHANNELS, SAMPLERATE, 2);
+
+      // Just an arbitrary multiplier. Pretend that each character is
+      // synthesized to 40 frames.
+      uint32_t frames_length = 40 * mText.Length();
+      auto frames = MakeUnique<int16_t[]>(frames_length);
+      mTask->SendAudioNative(frames.get(), frames_length);
+
+      mTask->SendAudioNative(nullptr, 0);
+
+      return NS_OK;
+    }
+
+  private:
+    nsCOMPtr<nsISpeechTask> mTask;
+    nsString mText;
+  };
+
+  nsCOMPtr<nsIRunnable> runnable = new Runnable(aTask, aText);
+  NS_DispatchToMainThread(runnable);
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+FakeDirectAudioSynth::GetServiceType(SpeechServiceType* aServiceType)
+{
+  *aServiceType = nsISpeechService::SERVICETYPE_DIRECT_AUDIO;
+  return NS_OK;
+}
+
+// FakeDirectAudioSynth
+
+class FakeIndirectAudioSynth : public nsISpeechService
+{
+
+public:
+  FakeIndirectAudioSynth() {}
+
+  NS_DECL_ISUPPORTS
+  NS_DECL_NSISPEECHSERVICE
+
+private:
+  virtual ~FakeIndirectAudioSynth() { }
+};
+
+NS_IMPL_ISUPPORTS(FakeIndirectAudioSynth, nsISpeechService)
+
+NS_IMETHODIMP
+FakeIndirectAudioSynth::Speak(const nsAString& aText, const nsAString& aUri,
+                              float aVolume, float aRate, float aPitch,
+                              nsISpeechTask* aTask)
+{
+  class DispatchStart final : public Runnable
+  {
+  public:
+    explicit DispatchStart(nsISpeechTask* aTask) :
+      mTask(aTask)
+    {
+    }
+
+    NS_IMETHOD Run() override
+    {
+      mTask->DispatchStart();
+
+      return NS_OK;
+    }
+
+  private:
+    nsCOMPtr<nsISpeechTask> mTask;
+  };
+
+  class DispatchEnd final : public Runnable
+  {
+  public:
+    DispatchEnd(nsISpeechTask* aTask, const nsAString& aText) :
+      mTask(aTask), mText(aText)
+    {
+    }
+
+    NS_IMETHOD Run() override
+    {
+      mTask->DispatchEnd(mText.Length()/2, mText.Length());
+
+      return NS_OK;
+    }
+
+  private:
+    nsCOMPtr<nsISpeechTask> mTask;
+    nsString mText;
+  };
+
+  class DispatchError final : public Runnable
+  {
+  public:
+    DispatchError(nsISpeechTask* aTask, const nsAString& aText) :
+      mTask(aTask), mText(aText)
+    {
+    }
+
+    NS_IMETHOD Run() override
+    {
+      mTask->DispatchError(mText.Length()/2, mText.Length());
+
+      return NS_OK;
+    }
+
+  private:
+    nsCOMPtr<nsISpeechTask> mTask;
+    nsString mText;
+  };
+
+  uint32_t flags = 0;
+  for (uint32_t i = 0; i < ArrayLength(sIndirectVoices); i++) {
+    if (aUri.EqualsASCII(sIndirectVoices[i].uri)) {
+      flags = sIndirectVoices[i].flags;
+    }
+  }
+
+  if (flags & eFailAtStart) {
+    return NS_ERROR_FAILURE;
+  }
+
+  RefPtr<FakeSynthCallback> cb = new FakeSynthCallback(
+    (flags & eSuppressEvents) ? nullptr : aTask);
+
+  aTask->Setup(cb, 0, 0, 0);
+
+  nsCOMPtr<nsIRunnable> runnable = new DispatchStart(aTask);
+  NS_DispatchToMainThread(runnable);
+
+  if (flags & eFail) {
+    runnable = new DispatchError(aTask, aText);
+    NS_DispatchToMainThread(runnable);
+  } else if ((flags & eSuppressEnd) == 0) {
+    runnable = new DispatchEnd(aTask, aText);
+    NS_DispatchToMainThread(runnable);
+  }
+
+  return NS_OK;
+}
+
+NS_IMETHODIMP
+FakeIndirectAudioSynth::GetServiceType(SpeechServiceType* aServiceType)
+{
+  *aServiceType = nsISpeechService::SERVICETYPE_INDIRECT_AUDIO;
+  return NS_OK;
+}
+
+// nsFakeSynthService
+
+NS_INTERFACE_MAP_BEGIN(nsFakeSynthServices)
+  NS_INTERFACE_MAP_ENTRY(nsIObserver)
+  NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsIObserver)
+NS_INTERFACE_MAP_END
+
+NS_IMPL_ADDREF(nsFakeSynthServices)
+NS_IMPL_RELEASE(nsFakeSynthServices)
+
+nsFakeSynthServices::nsFakeSynthServices()
+{
+}
+
+nsFakeSynthServices::~nsFakeSynthServices()
+{
+}
+
+static void
+AddVoices(nsISpeechService* aService, const VoiceDetails* aVoices, uint32_t aLength)
+{
+  RefPtr<nsSynthVoiceRegistry> registry = nsSynthVoiceRegistry::GetInstance();
+  for (uint32_t i = 0; i < aLength; i++) {
+    NS_ConvertUTF8toUTF16 name(aVoices[i].name);
+    NS_ConvertUTF8toUTF16 uri(aVoices[i].uri);
+    NS_ConvertUTF8toUTF16 lang(aVoices[i].lang);
+    // These services can handle more than one utterance at a time and have
+    // several speaking simultaniously. So, aQueuesUtterances == false
+    registry->AddVoice(aService, uri, name, lang, true, false);
+    if (aVoices[i].defaultVoice) {
+      registry->SetDefaultVoice(uri, true);
+    }
+  }
+
+  registry->NotifyVoicesChanged();
+}
+
+void
+nsFakeSynthServices::Init()
+{
+  mDirectService = new FakeDirectAudioSynth();
+  AddVoices(mDirectService, sDirectVoices, ArrayLength(sDirectVoices));
+
+  mIndirectService = new FakeIndirectAudioSynth();
+  AddVoices(mIndirectService, sIndirectVoices, ArrayLength(sIndirectVoices));
+}
+
+// nsIObserver
+
+NS_IMETHODIMP
+nsFakeSynthServices::Observe(nsISupports* aSubject, const char* aTopic,
+                             const char16_t* aData)
+{
+  MOZ_ASSERT(NS_IsMainThread());
+  if(NS_WARN_IF(!(!strcmp(aTopic, "speech-synth-started")))) {
+    return NS_ERROR_UNEXPECTED;
+  }
+
+  if (Preferences::GetBool("media.webspeech.synth.test")) {
+    NS_DispatchToMainThread(NewRunnableMethod(this, &nsFakeSynthServices::Init));
+  }
+
+  return NS_OK;
+}
+
+// static methods
+
+nsFakeSynthServices*
+nsFakeSynthServices::GetInstance()
+{
+  MOZ_ASSERT(NS_IsMainThread());
+  if (!XRE_IsParentProcess()) {
+    MOZ_ASSERT(false, "nsFakeSynthServices can only be started on main gecko process");
+    return nullptr;
+  }
+
+  if (!sSingleton) {
+    sSingleton = new nsFakeSynthServices();
+  }
+
+  return sSingleton;
+}
+
+already_AddRefed<nsFakeSynthServices>
+nsFakeSynthServices::GetInstanceForService()
+{
+  RefPtr<nsFakeSynthServices> picoService = GetInstance();
+  return picoService.forget();
+}
+
+void
+nsFakeSynthServices::Shutdown()
+{
+  if (!sSingleton) {
+    return;
+  }
+
+  sSingleton = nullptr;
+}
+
+} // namespace dom
+} // namespace mozilla

dom/media/webspeech/synth/test/nsFakeSynthServices.h

@@ -0,0 +1,52 @@
+/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* vim:set ts=2 sw=2 sts=2 et cindent: */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef nsFakeSynthServices_h
+#define nsFakeSynthServices_h
+
+#include "nsTArray.h"
+#include "nsIObserver.h"
+#include "nsIThread.h"
+#include "nsISpeechService.h"
+#include "nsRefPtrHashtable.h"
+#include "mozilla/StaticPtr.h"
+#include "mozilla/Monitor.h"
+
+namespace mozilla {
+namespace dom {
+
+class nsFakeSynthServices : public nsIObserver
+{
+
+public:
+  NS_DECL_ISUPPORTS
+  NS_DECL_NSIOBSERVER
+
+  nsFakeSynthServices();
+
+  static nsFakeSynthServices* GetInstance();
+
+  static already_AddRefed<nsFakeSynthServices> GetInstanceForService();
+
+  static void Shutdown();
+
+private:
+
+  virtual ~nsFakeSynthServices();
+
+  void Init();
+
+  nsCOMPtr<nsISpeechService> mDirectService;
+
+  nsCOMPtr<nsISpeechService> mIndirectService;
+
+  static StaticRefPtr<nsFakeSynthServices> sSingleton;
+};
+
+} // namespace dom
+} // namespace mozilla
+
+#endif