Reland "Move ISpVoice instance to a worker thread"

This is a reland of b642822f

Original change's description:
> Move ISpVoice instance to a worker thread
>
> This CL is moving the speech synthesizer code to a worker thread.
>
> The TtsPlatformImpl is using an ISpVoice COM interface to perform
> text-to-speech actions. The instance was living on the UI Main
> Thread which was using STA Apartment.
>
> This CL is splitting the logic in two parts. The first part is the
> control logic which stay on the UI Thread. The second part is the
> synthesizer logic which is running on a worker thread.
>
> Bug: 869399, 1133813
> Change-Id: I8f8d939950cc39637f87d101ed362123a51e985e
> Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2462341
> Commit-Queue: Etienne Bergeron <etienneb@chromium.org>
> Reviewed-by: François Doray <fdoray@chromium.org>
> Reviewed-by: Dominic Mazzoni <dmazzoni@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#820340}

Bug: 869399
Bug: 1133813
Change-Id: I4f1e5bb143472b2d96cf4fe55d3f74d1e5c56877
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2502948Reviewed-by: François Doray <fdoray@chromium.org>
Reviewed-by: Dominic Mazzoni <dmazzoni@chromium.org>
Commit-Queue: Etienne Bergeron <etienneb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#821450}

content/browser/speech/tts_win.cc

@@ -7,30 +7,137 @@
 #include <sapi.h>
 #include <stdint.h>
 #include <wrl/client.h>
+#include <wrl/implements.h>
 
 #include <algorithm>
 
 #include "base/bind.h"
 #include "base/macros.h"
 #include "base/no_destructor.h"
+#include "base/sequenced_task_runner.h"
 #include "base/strings/string_number_conversions.h"
 #include "base/strings/string_piece.h"
 #include "base/strings/utf_string_conversions.h"
+#include "base/synchronization/lock.h"
+#include "base/task/task_traits.h"
+#include "base/task/thread_pool.h"
+#include "base/task_runner.h"
+#include "base/thread_annotations.h"
+#include "base/threading/sequence_bound.h"
 #include "base/values.h"
 #include "base/win/scoped_co_mem.h"
 #include "base/win/sphelper.h"
 #include "content/browser/speech/tts_platform_impl.h"
+#include "content/public/browser/browser_task_traits.h"
+#include "content/public/browser/browser_thread.h"
 #include "content/public/browser/tts_controller.h"
 
 namespace content {
 
 namespace {
 
+class TtsPlatformImplWin;
+class TtsPlatformImplBackgroundWorker;
+
+constexpr int kInvalidUtteranceId = -1;
+
 // ISpObjectToken key and value names.
 const wchar_t kAttributesKey[] = L"Attributes";
 const wchar_t kLanguageValue[] = L"Language";
 
-}  // anonymous namespace.
+// This COM interface is receiving the TTS events on the ISpVoice asynchronous
+// worker thread and is emitting a notification task
+// TtsPlatformImplBackgroundWorker::SendTtsEvent(...) on the worker sequence.
+class TtsEventSink
+    : public Microsoft::WRL::RuntimeClass<
+          Microsoft::WRL::RuntimeClassFlags<Microsoft::WRL::ClassicCom>,
+          ISpNotifySink> {
+ public:
+  TtsEventSink(TtsPlatformImplBackgroundWorker* worker,
+               scoped_refptr<base::TaskRunner> worker_task_runner)
+      : worker_(worker), worker_task_runner_(std::move(worker_task_runner)) {}
+
+  // ISpNotifySink:
+  IFACEMETHODIMP Notify(void) override;
+
+  int GetUtteranceId() {
+    base::AutoLock lock(lock_);
+    return utterance_id_;
+  }
+
+  void SetUtteranceId(int utterance_id) {
+    base::AutoLock lock(lock_);
+    utterance_id_ = utterance_id;
+  }
+
+ private:
+  // |worker_| is leaky and must never deleted because TtsEventSink posts
+  // asynchronous tasks to it.
+  TtsPlatformImplBackgroundWorker* worker_;
+  scoped_refptr<base::TaskRunner> worker_task_runner_;
+
+  base::Lock lock_;
+  int utterance_id_ GUARDED_BY(lock_);
+};
+
+class TtsPlatformImplBackgroundWorker {
+ public:
+  explicit TtsPlatformImplBackgroundWorker(
+      scoped_refptr<base::TaskRunner> task_runner)
+      : tts_event_sink_(
+            Microsoft::WRL::Make<TtsEventSink>(this, std::move(task_runner))) {}
+  TtsPlatformImplBackgroundWorker(const TtsPlatformImplBackgroundWorker&) =
+      delete;
+  TtsPlatformImplBackgroundWorker& operator=(
+      const TtsPlatformImplBackgroundWorker&) = delete;
+  ~TtsPlatformImplBackgroundWorker() = default;
+
+  void Initialize();
+
+  void ProcessSpeech(int utterance_id,
+                     const std::string& lang,
+                     const VoiceData& voice,
+                     const UtteranceContinuousParameters& params,
+                     base::OnceCallback<void(bool)> on_speak_finished,
+                     const std::string& parsed_utterance);
+
+  void StopSpeaking(bool paused);
+  void Pause();
+  void Resume();
+  void Shutdown();
+
+  // This function is called after being notified by the speech synthetizer that
+  // there are TTS notifications are available and should be they should be
+  // processed.
+  void OnSpeechEvent(int utterance_id);
+
+  // Send an TTS event notification to the TTS controller.
+  void SendTtsEvent(int utterance_id,
+                    TtsEventType event_type,
+                    int char_index,
+                    int length = -1);
+
+ private:
+  void GetVoices(std::vector<VoiceData>* voices);
+
+  void SetVoiceFromName(const std::string& name);
+
+  // These apply to the current utterance only that is currently being processed
+  // on the worker thread. TTS events are dispatched by TtsEventSink to this
+  // class and update the current speaking state of the utterance.
+  std::string last_voice_name_;
+  ULONG stream_number_ = 0u;
+  int utterance_id_ = kInvalidUtteranceId;
+  size_t utterance_char_position_ = 0u;
+  size_t utterance_prefix_length_ = 0u;
+  size_t utterance_length_ = 0u;
+
+  // The COM class ISpVoice lives within the COM MTA apartment (worker pool).
+  // This interface can not be called on the UI thread since UI thread is
+  // COM STA.
+  Microsoft::WRL::ComPtr<ISpVoice> speech_synthesizer_;
+  Microsoft::WRL::ComPtr<TtsEventSink> tts_event_sink_;
+};
 
 class TtsPlatformImplWin : public TtsPlatformImpl {
  public:
@@ -38,7 +145,7 @@ class TtsPlatformImplWin : public TtsPlatformImpl {
   TtsPlatformImplWin& operator=(const TtsPlatformImplWin&) = delete;
 
   bool PlatformImplSupported() override { return true; }
-  bool PlatformImplInitialized() override { return true; }
+  bool PlatformImplInitialized() override;
 
   void Speak(int utterance_id,
              const std::string& utterance,
@@ -59,19 +166,18 @@ class TtsPlatformImplWin : public TtsPlatformImpl {
 
   void Shutdown() override;
 
+  void OnInitializeComplete(bool success, std::vector<VoiceData> voices);
+  void OnSpeakScheduled(base::OnceCallback<void(bool)> on_speak_finished,
+                        bool success);
+  void OnSpeakFinished(int utterance_id);
+
   // Get the single instance of this class.
   static TtsPlatformImplWin* GetInstance();
 
-  static void __stdcall SpeechEventCallback(WPARAM w_param, LPARAM l_param);
-
  private:
   friend base::NoDestructor<TtsPlatformImplWin>;
   TtsPlatformImplWin();
 
-  void OnSpeechEvent();
-
-  void SetVoiceFromName(const std::string& name);
-
   void ProcessSpeech(int utterance_id,
                      const std::string& lang,
                      const VoiceData& voice,
@@ -79,52 +185,63 @@ class TtsPlatformImplWin : public TtsPlatformImpl {
                      base::OnceCallback<void(bool)> on_speak_finished,
                      const std::string& parsed_utterance);
 
-  Microsoft::WRL::ComPtr<ISpVoice> speech_synthesizer_;
-
-  // These apply to the current utterance only.
-  std::wstring utterance_;
-  int utterance_id_ = 0;
-  int prefix_len_ = 0;
-  ULONG stream_number_ = 0u;
-  int char_position_ = 0;
-  int char_length_ = 0;
+  // These variables hold the platform state.
   bool paused_ = false;
-  std::string last_voice_name_;
+  bool is_speaking_ = false;
+  int utterance_id_ = kInvalidUtteranceId;
+  bool platform_initialized_ = false;
+  std::vector<VoiceData> voices_;
+
+  // Hold the state and the code of the background implementation.
+  scoped_refptr<base::SequencedTaskRunner> worker_task_runner_;
+  base::SequenceBound<TtsPlatformImplBackgroundWorker> worker_;
 };
 
-// static
-TtsPlatformImpl* TtsPlatformImpl::GetInstance() {
-  return TtsPlatformImplWin::GetInstance();
+HRESULT TtsEventSink::Notify() {
+  worker_task_runner_->PostTask(
+      FROM_HERE, base::BindOnce(&TtsPlatformImplBackgroundWorker::OnSpeechEvent,
+                                base::Unretained(worker_), GetUtteranceId()));
+  return S_OK;
 }
 
-void TtsPlatformImplWin::Speak(
-    int utterance_id,
-    const std::string& utterance,
-    const std::string& lang,
-    const VoiceData& voice,
-    const UtteranceContinuousParameters& params,
-    base::OnceCallback<void(bool)> on_speak_finished) {
-  // Parse SSML and process speech.
-  TtsController::GetInstance()->StripSSML(
-      utterance, base::BindOnce(&TtsPlatformImplWin::ProcessSpeech,
-                                base::Unretained(this), utterance_id, lang,
-                                voice, params, std::move(on_speak_finished)));
+//
+// TtsPlatformImplBackgroundWorker
+//
+
+void TtsPlatformImplBackgroundWorker::Initialize() {
+  bool success = false;
+  std::vector<VoiceData> voices;
+
+  ::CoCreateInstance(CLSID_SpVoice, nullptr, CLSCTX_ALL,
+                     IID_PPV_ARGS(&speech_synthesizer_));
+  if (speech_synthesizer_.Get()) {
+    ULONGLONG event_mask =
+        SPFEI(SPEI_START_INPUT_STREAM) | SPFEI(SPEI_TTS_BOOKMARK) |
+        SPFEI(SPEI_WORD_BOUNDARY) | SPFEI(SPEI_SENTENCE_BOUNDARY) |
+        SPFEI(SPEI_END_INPUT_STREAM);
+    speech_synthesizer_->SetInterest(event_mask, event_mask);
+    speech_synthesizer_->SetNotifySink(tts_event_sink_.Get());
+
+    GetVoices(&voices);
+
+    success = true;
+  }
+
+  GetUIThreadTaskRunner({})->PostTask(
+      FROM_HERE,
+      base::BindOnce(&TtsPlatformImplWin::OnInitializeComplete,
+                     base::Unretained(TtsPlatformImplWin::GetInstance()),
+                     success, std::move(voices)));
 }
 
-void TtsPlatformImplWin::ProcessSpeech(
+void TtsPlatformImplBackgroundWorker::ProcessSpeech(
     int utterance_id,
     const std::string& lang,
     const VoiceData& voice,
     const UtteranceContinuousParameters& params,
     base::OnceCallback<void(bool)> on_speak_finished,
     const std::string& parsed_utterance) {
-  std::wstring prefix;
-  std::wstring suffix;
-
-  if (!speech_synthesizer_.Get()) {
-    std::move(on_speak_finished).Run(false);
-    return;
-  }
+  DCHECK(speech_synthesizer_.Get());
 
   SetVoiceFromName(voice.name);
 
@@ -137,6 +254,8 @@ void TtsPlatformImplWin::ProcessSpeech(
     speech_synthesizer_->SetRate(static_cast<int32_t>(10 * log10(params.rate)));
   }
 
+  std::wstring prefix;
+  std::wstring suffix;
   if (params.pitch >= 0.0) {
     // The TTS api allows a range of -10 to 10 for speech pitch:
     // https://docs.microsoft.com/en-us/previous-versions/windows/desktop/ms720500(v%3Dvs.85)
@@ -158,70 +277,125 @@ void TtsPlatformImplWin::ProcessSpeech(
 
   // TODO(dmazzoni): convert SSML to SAPI xml. http://crbug.com/88072
 
-  utterance_ = base::UTF8ToWide(parsed_utterance);
+  std::wstring utterance = base::UTF8ToWide(parsed_utterance);
+  std::wstring merged_utterance = prefix + utterance + suffix;
+
   utterance_id_ = utterance_id;
-  char_position_ = 0;
-  char_length_ = 0;
-  std::wstring merged_utterance = prefix + utterance_ + suffix;
-  prefix_len_ = prefix.size();
+  utterance_char_position_ = 0;
+  utterance_length_ = utterance.size();
+  utterance_prefix_length_ = prefix.size();
+
+  tts_event_sink_->SetUtteranceId(utterance_id);
 
   HRESULT result = speech_synthesizer_->Speak(merged_utterance.c_str(),
                                               SPF_ASYNC, &stream_number_);
-  std::move(on_speak_finished).Run((result == S_OK));
+  bool success = (result == S_OK);
+  GetUIThreadTaskRunner({})->PostTask(
+      FROM_HERE, base::BindOnce(std::move(on_speak_finished), success));
 }
 
-bool TtsPlatformImplWin::StopSpeaking() {
+void TtsPlatformImplBackgroundWorker::StopSpeaking(bool paused) {
   if (speech_synthesizer_.Get()) {
-    // Clear the stream number so that any further events relating to this
-    // utterance are ignored.
-    stream_number_ = 0;
+    // Block notifications from the current utterance.
+    tts_event_sink_->SetUtteranceId(kInvalidUtteranceId);
+    utterance_id_ = kInvalidUtteranceId;
 
-    if (IsSpeaking()) {
     // Stop speech by speaking nullptr with the purge flag.
-      speech_synthesizer_->Speak(nullptr, SPF_ASYNC | SPF_PURGEBEFORESPEAK,
-                                 NULL);
-    }
-    if (paused_) {
+    speech_synthesizer_->Speak(nullptr, SPF_PURGEBEFORESPEAK, nullptr);
+
+    // Ensures the synthesizer is not paused after a stop.
+    if (paused)
       speech_synthesizer_->Resume();
-      paused_ = false;
-    }
   }
-  return true;
 }
 
-void TtsPlatformImplWin::Pause() {
-  if (speech_synthesizer_.Get() && utterance_id_ && !paused_) {
+void TtsPlatformImplBackgroundWorker::Pause() {
+  if (speech_synthesizer_.Get()) {
     speech_synthesizer_->Pause();
-    paused_ = true;
-    TtsController::GetInstance()->OnTtsEvent(utterance_id_, TTS_EVENT_PAUSE,
-                                             char_position_, -1, "");
+    SendTtsEvent(utterance_id_, TTS_EVENT_PAUSE, utterance_char_position_);
   }
 }
 
-void TtsPlatformImplWin::Resume() {
-  if (speech_synthesizer_.Get() && utterance_id_ && paused_) {
+void TtsPlatformImplBackgroundWorker::Resume() {
+  if (speech_synthesizer_.Get()) {
     speech_synthesizer_->Resume();
-    paused_ = false;
-    TtsController::GetInstance()->OnTtsEvent(utterance_id_, TTS_EVENT_RESUME,
-                                             char_position_, -1, "");
+    SendTtsEvent(utterance_id_, TTS_EVENT_RESUME, utterance_char_position_);
   }
 }
 
-bool TtsPlatformImplWin::IsSpeaking() {
-  if (speech_synthesizer_.Get()) {
-    SPVOICESTATUS status;
-    HRESULT result = speech_synthesizer_->GetStatus(&status, NULL);
-    if (result == S_OK) {
-      if (status.dwRunningState == 0 ||  // 0 == waiting to speak
-          status.dwRunningState == SPRS_IS_SPEAKING) {
-        return true;
+void TtsPlatformImplBackgroundWorker::Shutdown() {
+  if (speech_synthesizer_)
+    speech_synthesizer_->SetNotifySink(nullptr);
+  if (tts_event_sink_) {
+    tts_event_sink_->SetUtteranceId(kInvalidUtteranceId);
+    utterance_id_ = kInvalidUtteranceId;
   }
+
+  tts_event_sink_ = nullptr;
+  speech_synthesizer_ = nullptr;
+}
+
+void TtsPlatformImplBackgroundWorker::OnSpeechEvent(int utterance_id) {
+  if (!speech_synthesizer_.Get())
+    return;
+
+  SPEVENT event;
+  while (S_OK == speech_synthesizer_->GetEvents(1, &event, nullptr)) {
+    // Ignore notifications that are not related to the current utterance.
+    if (event.ulStreamNum != stream_number_ ||
+        utterance_id_ == kInvalidUtteranceId || utterance_id != utterance_id_) {
+      continue;
+    }
+
+    switch (event.eEventId) {
+      case SPEI_START_INPUT_STREAM:
+        utterance_char_position_ = 0;
+        SendTtsEvent(utterance_id_, TTS_EVENT_START, utterance_char_position_);
+        break;
+      case SPEI_END_INPUT_STREAM:
+        GetUIThreadTaskRunner({})->PostTask(
+            FROM_HERE,
+            base::BindOnce(&TtsPlatformImplWin::OnSpeakFinished,
+                           base::Unretained(TtsPlatformImplWin::GetInstance()),
+                           utterance_id_));
+
+        utterance_char_position_ = utterance_length_;
+        SendTtsEvent(utterance_id_, TTS_EVENT_END, utterance_char_position_);
+        break;
+      case SPEI_TTS_BOOKMARK:
+        SendTtsEvent(utterance_id_, TTS_EVENT_MARKER, utterance_char_position_);
+        break;
+      case SPEI_WORD_BOUNDARY:
+        SendTtsEvent(utterance_id_, TTS_EVENT_WORD, utterance_char_position_,
+                     static_cast<ULONG>(event.wParam));
+
+        break;
+      case SPEI_SENTENCE_BOUNDARY:
+        SendTtsEvent(utterance_id_, TTS_EVENT_SENTENCE,
+                     utterance_char_position_);
+        break;
+      default:
+        break;
     }
   }
-  return false;
 }
 
-void TtsPlatformImplWin::GetVoices(std::vector<VoiceData>* out_voices) {
+void TtsPlatformImplBackgroundWorker::SendTtsEvent(int utterance_id,
+                                                   TtsEventType event_type,
+                                                   int char_index,
+                                                   int length) {
+  GetUIThreadTaskRunner({})->PostTask(
+      FROM_HERE, base::BindOnce(&TtsController::OnTtsEvent,
+                                base::Unretained(TtsController::GetInstance()),
+                                utterance_id, event_type, char_index, length,
+                                std::string()));
+}
+
+void TtsPlatformImplBackgroundWorker::GetVoices(
+    std::vector<VoiceData>* out_voices) {
+  if (!speech_synthesizer_.Get())
+    return;
+
   Microsoft::WRL::ComPtr<IEnumSpObjectTokens> voice_tokens;
   unsigned long voice_count;
   if (S_OK != SpEnumTokens(SPCAT_VOICES, NULL, NULL, &voice_tokens))
@@ -267,54 +441,11 @@ void TtsPlatformImplWin::GetVoices(std::vector<VoiceData>* out_voices) {
   }
 }
 
-void TtsPlatformImplWin::OnSpeechEvent() {
-  if (!speech_synthesizer_.Get())
-    return;
-
-  TtsController* controller = TtsController::GetInstance();
-  SPEVENT event;
-  while (S_OK == speech_synthesizer_->GetEvents(1, &event, NULL)) {
-    if (event.ulStreamNum != stream_number_)
-      continue;
-
-    switch (event.eEventId) {
-      case SPEI_START_INPUT_STREAM:
-        controller->OnTtsEvent(utterance_id_, TTS_EVENT_START, 0, -1,
-                               std::string());
-        break;
-      case SPEI_END_INPUT_STREAM:
-        char_position_ = utterance_.size();
-        controller->OnTtsEvent(utterance_id_, TTS_EVENT_END, char_position_, 0,
-                               std::string());
-        break;
-      case SPEI_TTS_BOOKMARK:
-        controller->OnTtsEvent(utterance_id_, TTS_EVENT_MARKER, char_position_,
-                               -1, std::string());
-        break;
-      case SPEI_WORD_BOUNDARY:
-        char_position_ = static_cast<ULONG>(event.lParam) - prefix_len_;
-        char_length_ = static_cast<ULONG>(event.wParam);
-        controller->OnTtsEvent(utterance_id_, TTS_EVENT_WORD, char_position_,
-                               char_length_, std::string());
-        break;
-      case SPEI_SENTENCE_BOUNDARY:
-        char_position_ = static_cast<ULONG>(event.lParam) - prefix_len_;
-        controller->OnTtsEvent(utterance_id_, TTS_EVENT_SENTENCE,
-                               char_position_, -1, std::string());
-        break;
-      default:
-        break;
-    }
-  }
-}
-
-void TtsPlatformImplWin::SetVoiceFromName(const std::string& name) {
+void TtsPlatformImplBackgroundWorker::SetVoiceFromName(
+    const std::string& name) {
   if (name.empty() || name == last_voice_name_)
     return;
 
-  if (!speech_synthesizer_.Get())
-    return;
-
   last_voice_name_ = name;
 
   Microsoft::WRL::ComPtr<IEnumSpObjectTokens> voice_tokens;
@@ -339,24 +470,158 @@ void TtsPlatformImplWin::SetVoiceFromName(const std::string& name) {
   }
 }
 
+//
+// TtsPlatformImplWin
+//
+
+bool TtsPlatformImplWin::PlatformImplInitialized() {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  return platform_initialized_;
+}
+
+void TtsPlatformImplWin::Speak(
+    int utterance_id,
+    const std::string& utterance,
+    const std::string& lang,
+    const VoiceData& voice,
+    const UtteranceContinuousParameters& params,
+    base::OnceCallback<void(bool)> on_speak_finished) {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  DCHECK(platform_initialized_);
+
+  // Do not emit utterance if the platform is not ready.
+  if (paused_ || is_speaking_) {
+    std::move(on_speak_finished).Run(false);
+    return;
+  }
+
+  // Flag that a utterance is getting emitted. The |is_speaking_| flag will be
+  // set back to false when the utterance will be fully spoken, stopped or if
+  // the voice synthetizer was not able to emit it.
+  is_speaking_ = true;
+  utterance_id_ = utterance_id;
+
+  // Parse SSML and process speech.
+  TtsController::GetInstance()->StripSSML(
+      utterance,
+      base::BindOnce(&TtsPlatformImplWin::ProcessSpeech, base::Unretained(this),
+                     utterance_id, lang, voice, params,
+                     base::BindOnce(&TtsPlatformImplWin::OnSpeakScheduled,
+                                    base::Unretained(this),
+                                    std::move(on_speak_finished))));
+}
+
+bool TtsPlatformImplWin::StopSpeaking() {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+
+  worker_.Post(FROM_HERE, &TtsPlatformImplBackgroundWorker::StopSpeaking,
+               paused_);
+  paused_ = false;
+
+  is_speaking_ = false;
+  utterance_id_ = kInvalidUtteranceId;
+
+  return true;
+}
+
+void TtsPlatformImplWin::Pause() {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  DCHECK(platform_initialized_);
+
+  if (paused_ || !is_speaking_)
+    return;
+  worker_.Post(FROM_HERE, &TtsPlatformImplBackgroundWorker::Pause);
+  paused_ = true;
+}
+
+void TtsPlatformImplWin::Resume() {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  DCHECK(platform_initialized_);
+
+  if (!paused_ || !is_speaking_)
+    return;
+
+  worker_.Post(FROM_HERE, &TtsPlatformImplBackgroundWorker::Resume);
+  paused_ = false;
+}
+
+bool TtsPlatformImplWin::IsSpeaking() {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  DCHECK(platform_initialized_);
+  return is_speaking_ && !paused_;
+}
+
+void TtsPlatformImplWin::GetVoices(std::vector<VoiceData>* out_voices) {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  DCHECK(platform_initialized_);
+  out_voices->insert(out_voices->end(), voices_.begin(), voices_.end());
+}
+
 void TtsPlatformImplWin::Shutdown() {
   // This is required to ensures the object is released before the COM is
   // uninitialized. Otherwise, this is causing shutdown hangs.
-  speech_synthesizer_ = nullptr;
+  worker_.Post(FROM_HERE, &TtsPlatformImplBackgroundWorker::Shutdown);
 }
 
-TtsPlatformImplWin::TtsPlatformImplWin() {
-  ::CoCreateInstance(CLSID_SpVoice, nullptr, CLSCTX_ALL,
-                     IID_PPV_ARGS(&speech_synthesizer_));
-  if (speech_synthesizer_.Get()) {
-    ULONGLONG event_mask =
-        SPFEI(SPEI_START_INPUT_STREAM) | SPFEI(SPEI_TTS_BOOKMARK) |
-        SPFEI(SPEI_WORD_BOUNDARY) | SPFEI(SPEI_SENTENCE_BOUNDARY) |
-        SPFEI(SPEI_END_INPUT_STREAM);
-    speech_synthesizer_->SetInterest(event_mask, event_mask);
-    speech_synthesizer_->SetNotifyCallbackFunction(
-        TtsPlatformImplWin::SpeechEventCallback, 0, 0);
+void TtsPlatformImplWin::OnInitializeComplete(bool success,
+                                              std::vector<VoiceData> voices) {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+
+  if (success)
+    voices_ = std::move(voices);
+
+  platform_initialized_ = true;
+  TtsController::GetInstance()->VoicesChanged();
+}
+
+void TtsPlatformImplWin::OnSpeakScheduled(
+    base::OnceCallback<void(bool)> on_speak_finished,
+    bool success) {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  DCHECK(is_speaking_);
+
+  // If the utterance was not able to be emitted, stop the speaking. There
+  // won't be any asynchronous TTS event to confirm the end of the speech.
+  if (!success) {
+    is_speaking_ = false;
+    utterance_id_ = kInvalidUtteranceId;
   }
+
+  // Pass the results to our caller.
+  std::move(on_speak_finished).Run(success);
+}
+
+void TtsPlatformImplWin::OnSpeakFinished(int utterance_id) {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  if (utterance_id != utterance_id_)
+    return;
+
+  DCHECK(is_speaking_);
+  DCHECK_NE(utterance_id_, kInvalidUtteranceId);
+  is_speaking_ = false;
+  utterance_id_ = kInvalidUtteranceId;
+}
+
+void TtsPlatformImplWin::ProcessSpeech(
+    int utterance_id,
+    const std::string& lang,
+    const VoiceData& voice,
+    const UtteranceContinuousParameters& params,
+    base::OnceCallback<void(bool)> on_speak_finished,
+    const std::string& parsed_utterance) {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+
+  worker_.Post(FROM_HERE, &TtsPlatformImplBackgroundWorker::ProcessSpeech,
+               utterance_id, lang, voice, params, std::move(on_speak_finished),
+               parsed_utterance);
+}
+
+TtsPlatformImplWin::TtsPlatformImplWin()
+    : worker_task_runner_(
+          base::ThreadPool::CreateSequencedTaskRunner({base::MayBlock()})),
+      worker_(worker_task_runner_, worker_task_runner_) {
+  DCHECK(BrowserThread::CurrentlyOn(content::BrowserThread::UI));
+  worker_.Post(FROM_HERE, &TtsPlatformImplBackgroundWorker::Initialize);
 }
 
 // static
@@ -365,9 +630,11 @@ TtsPlatformImplWin* TtsPlatformImplWin::GetInstance() {
   return tts_platform.get();
 }
 
+}  // namespace
+
 // static
-void TtsPlatformImplWin::SpeechEventCallback(WPARAM w_param, LPARAM l_param) {
-  GetInstance()->OnSpeechEvent();
+TtsPlatformImpl* TtsPlatformImpl::GetInstance() {
+  return TtsPlatformImplWin::GetInstance();
 }
 
 }  // namespace content