Binding media stream audio track to speech recognition [browser]

content/browser/speech/speech_recognizer_impl.cc

Index: content/browser/speech/speech_recognizer_impl.cc
diff --git a/content/browser/speech/speech_recognizer_impl.cc b/content/browser/speech/speech_recognizer_impl.cc
index f6fb3900a5e03cdfa08e0c89542c663021e288c4..821f2e1176be3a9da44f84331647fcacd46a569f 100644
--- a/content/browser/speech/speech_recognizer_impl.cc
+++ b/content/browser/speech/speech_recognizer_impl.cc
@@ -6,11 +6,16 @@
 
 #include "base/basictypes.h"
 #include "base/bind.h"
+#include "base/memory/shared_memory.h"
+#include "base/threading/thread.h"
+#include "base/threading/thread_restrictions.h"
 #include "base/time/time.h"
 #include "content/browser/browser_main_loop.h"
 #include "content/browser/media/media_internals.h"
 #include "content/browser/speech/audio_buffer.h"
 #include "content/browser/speech/google_one_shot_remote_engine.h"
+#include "content/browser/speech/speech_recognition_audio_receiver.h"
+#include "content/public/browser/browser_message_filter.h"
 #include "content/public/browser/speech_recognition_event_listener.h"
 #include "media/base/audio_converter.h"
 #include "net/url_request/url_request_context_getter.h"
@@ -41,6 +46,7 @@ class SpeechRecognizerImpl::OnDataConverter
   // Converts input audio |data| bus into an AudioChunk where the input format
   // is given by |input_parameters_| and the output format by
   // |output_parameters_|.
+  // Skips conversion on |audio_converter_| when input and output formats match.
   scoped_refptr<AudioChunk> Convert(const AudioBus* data);
 
  private:
@@ -49,13 +55,14 @@ class SpeechRecognizerImpl::OnDataConverter
 
   // Handles resampling, buffering, and channel mixing between input and output
   // parameters.
-  AudioConverter audio_converter_;
+  scoped_ptr<AudioConverter> audio_converter_;
 
   scoped_ptr<AudioBus> input_bus_;
   scoped_ptr<AudioBus> output_bus_;
   const AudioParameters input_parameters_;
   const AudioParameters output_parameters_;
   bool waiting_for_input_;
+  bool need_conversion_;
   scoped_ptr<uint8[]> converted_data_;
 
   DISALLOW_COPY_AND_ASSIGN(OnDataConverter);
@@ -115,30 +122,40 @@ COMPILE_ASSERT(SpeechRecognizerImpl::kNumBitsPerAudioSample % 8 == 0,
 
 SpeechRecognizerImpl::OnDataConverter::OnDataConverter(
     const AudioParameters& input_params, const AudioParameters& output_params)
-    : audio_converter_(input_params, output_params, false),
-      input_bus_(AudioBus::Create(input_params)),
+    : input_bus_(AudioBus::Create(input_params)),
       output_bus_(AudioBus::Create(output_params)),
       input_parameters_(input_params),
       output_parameters_(output_params),
       waiting_for_input_(false),
       converted_data_(new uint8[output_parameters_.GetBytesPerBuffer()]) {
-  audio_converter_.AddInput(this);
+  if (!input_parameters_.Equals(output_parameters_)) {
+    audio_converter_.reset(new AudioConverter(input_params,
+                                              output_params,
+                                              false));
+    audio_converter_->AddInput(this);
+  }
 }
 
 SpeechRecognizerImpl::OnDataConverter::~OnDataConverter() {
   // It should now be safe to unregister the converter since no more OnData()
   // callbacks are outstanding at this point.
-  audio_converter_.RemoveInput(this);
+  if (audio_converter_.get())
+    audio_converter_->RemoveInput(this);
 }
 
 scoped_refptr<AudioChunk> SpeechRecognizerImpl::OnDataConverter::Convert(
     const AudioBus* data) {
   CHECK_EQ(data->frames(), input_parameters_.frames_per_buffer());
 
-  data->CopyTo(input_bus_.get());
-
-  waiting_for_input_ = true;
-  audio_converter_.Convert(output_bus_.get());
+  // Do conversion if needed - only when |input_parameters_| differ from
+  // |output_parameters_|.
+  if (audio_converter_.get()) {
+    data->CopyTo(input_bus_.get());
+    waiting_for_input_ = true;
+    audio_converter_->Convert(output_bus_.get());
+  } else {
+    data->CopyTo(output_bus_.get());
+  }
 
   output_bus_->ToInterleaved(
       output_bus_->frames(), output_parameters_.bits_per_sample() / 8,
@@ -174,6 +191,8 @@ SpeechRecognizerImpl::SpeechRecognizerImpl(
     int session_id,
     bool continuous,
     bool provisional_results,
+    bool using_audio_receiver,
+    const base::ProcessHandle peer_process_handle,
     SpeechRecognitionEngine* engine)
     : SpeechRecognizer(listener, session_id),
       recognition_engine_(engine),
@@ -182,6 +201,8 @@ SpeechRecognizerImpl::SpeechRecognizerImpl(
           media::AudioLogFactory::AUDIO_INPUT_CONTROLLER)),
       is_dispatching_event_(false),
       provisional_results_(provisional_results),
+      using_audio_receiver_(using_audio_receiver),
+      peer_process_handle_(peer_process_handle),
       state_(STATE_IDLE) {
   DCHECK(recognition_engine_ != NULL);
   if (!continuous) {
@@ -213,7 +234,7 @@ SpeechRecognizerImpl::SpeechRecognizerImpl(
 // synchronous callbacks.
 
 void SpeechRecognizerImpl::StartRecognition(const std::string& device_id) {
-  DCHECK(!device_id.empty());
+  DCHECK(!device_id.empty() ^ using_audio_receiver_);
   device_id_ = device_id;
 
   BrowserThread::PostTask(BrowserThread::IO, FROM_HERE,
@@ -244,7 +265,8 @@ bool SpeechRecognizerImpl::IsCapturingAudio() const {
   DCHECK(BrowserThread::CurrentlyOn(BrowserThread::IO)); // See IsActive().
   const bool is_capturing_audio = state_ >= STATE_STARTING &&
                                   state_ <= STATE_RECOGNIZING;
-  DCHECK((is_capturing_audio && (audio_controller_.get() != NULL)) ||
+  DCHECK((is_capturing_audio &&
+          (using_audio_receiver_ || (audio_controller_.get() != NULL))) ||
          (!is_capturing_audio && audio_controller_.get() == NULL));
   return is_capturing_audio;
 }
@@ -261,6 +283,8 @@ SpeechRecognizerImpl::~SpeechRecognizerImpl() {
     audio_controller_->Close(
         base::Bind(&KeepAudioControllerRefcountedForDtor, audio_controller_));
     audio_log_->OnClosed(0);
+  } else if (audio_receiver_.get())  {
+    ResetAudioReceiver();
   }
 }
 
@@ -481,7 +505,7 @@ void SpeechRecognizerImpl::ProcessAudioPipeline(const AudioChunk& raw_audio) {
     endpointer_.ProcessAudio(raw_audio, &rms);
 
   if (route_to_vumeter) {
-    DCHECK(route_to_endpointer); // Depends on endpointer due to |rms|.
+    DCHECK(route_to_endpointer);  // Depends on endpointer due to |rms|.
     UpdateSignalAndNoiseLevels(rms, clip_detected);
   }
   if (route_to_sr_engine) {
@@ -528,6 +552,16 @@ SpeechRecognizerImpl::StartRecording(const FSMEventArgs&) {
       AudioParameters::AUDIO_PCM_LOW_LATENCY, kChannelLayout, kAudioSampleRate,
       kNumBitsPerAudioSample, frames_per_buffer);
 
+  // Check if using audio receiver as source (audio data processed on renderer).
+  // If |StartRecordingUsingAudioReceiver()| fails, we abort the recognition.
+  if (using_audio_receiver_) {
+    return StartRecordingUsingAudioReceiver(output_parameters) ?
+        STATE_STARTING :
+        Abort(SpeechRecognitionError(SPEECH_RECOGNITION_ERROR_AUDIO));
+  }
+
+  // Get audio data from microphone on the browser.
+
   // Audio converter will receive audio based on these parameters as input.
   // On Windows we start by verifying that Core Audio is supported. If not,
   // the WaveIn API is used and we might as well avoid all audio conversations
@@ -584,6 +618,77 @@ SpeechRecognizerImpl::StartRecording(const FSMEventArgs&) {
   return STATE_STARTING;
 }
 
+bool SpeechRecognizerImpl::StartRecordingUsingAudioReceiver(
+    const media::AudioParameters& params) {
+  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::IO));
+  DCHECK(!audio_thread_.get());
+  DCHECK(!audio_receiver_.get());
+
+  audio_thread_.reset(new base::Thread("AudioReceiverThread"));
+  if (!audio_thread_->Start()) {
+    DLOG(ERROR) << "Cannot start new audio thread";
+    audio_thread_.reset();
+    return false;
+  }
+
+  // Start receiving audio, incoming data will be posted to |OnAudioTrackData|.
+  base::SharedMemoryHandle foreign_memory_handle;
+  base::SyncSocket::TransitDescriptor foreign_socket_descriptor;
+  audio_receiver_ = SpeechRecognitionAudioReceiver::Create(
+      params,
+      peer_process_handle_,
+      this,  // Trigger OnData() and OnError() events here.
+      audio_thread_->task_runner().get(),
+      &foreign_memory_handle,
+      &foreign_socket_descriptor);
+
+  if (!audio_receiver_.get()) {
+    ResetAudioReceiver();
+    return false;
+  }
+
+  // Reusing the audio converter only to create chunks from the bus.
+  // Same input and output format indicate no additional conversion here.
+  audio_converter_.reset(new OnDataConverter(params, params));
+
+  // The endpointer needs to estimate the environment/background noise before
+  // starting to treat the audio as user input. We wait in the state
+  // ESTIMATING_ENVIRONMENT until such interval has elapsed before switching
+  // to user input mode.
+  endpointer_.SetEnvironmentEstimationMode();
+
+  listener()->OnAudioReceiverReady(session_id(),
+                                   params,
+                                   foreign_memory_handle,
+                                   foreign_socket_descriptor);
+
+  // Start listening on the receiver's audio thread.
+  audio_receiver_->StartListening();
+
+  return true;
+}
+
+void SpeechRecognizerImpl::ResetAudioReceiver() {
+  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::IO));
+
+  // We stop listening not to post any new tasks to the audio thread.
+  if (audio_receiver_.get())
+    audio_receiver_->Stop();
+
+  if (!audio_thread_.get())
+    return;
+
+  // This synchronous operation is usually instantaneous since the receiver
+  // closes the socket on Stop() and no new tasks will be posted to the
+  // |audio_thread_|.
+  const bool previous_io_allowed = base::ThreadRestrictions::SetIOAllowed(true);
+  audio_thread_->Stop();
+  base::ThreadRestrictions::SetIOAllowed(previous_io_allowed);
+
+  audio_thread_.reset();
+  audio_receiver_.reset();
+}
+
 SpeechRecognizerImpl::FSMState
 SpeechRecognizerImpl::StartRecognitionEngine(const FSMEventArgs& event_args) {
   // This is the first audio packet captured, so the recognition engine is
@@ -769,6 +874,12 @@ SpeechRecognizerImpl::NotFeasible(const FSMEventArgs& event_args) {
 
 void SpeechRecognizerImpl::CloseAudioControllerAsynchronously() {
   DCHECK(IsCapturingAudio());
+
+  if (using_audio_receiver_) {
+    ResetAudioReceiver();
+    return;
+  }
+
   DVLOG(1) << "SpeechRecognizerImpl closing audio controller.";
   // Issues a Close on the audio controller, passing an empty callback. The only
   // purpose of such callback is to keep the audio controller refcounted until