OLD | NEW |
(Empty) | |
| 1 // Copyright 2014 The Chromium Authors. All rights reserved. |
| 2 // Use of this source code is governed by a BSD-style license that can be |
| 3 // found in the LICENSE file. |
| 4 |
| 5 #include "content/renderer/media/speech_recognition_audio_sink.h" |
| 6 |
| 7 #include "base/logging.h" |
| 8 #include "base/memory/shared_memory.h" |
| 9 #include "base/time/time.h" |
| 10 #include "media/audio/audio_parameters.h" |
| 11 #include "media/base/audio_fifo.h" |
| 12 |
| 13 namespace content { |
| 14 |
| 15 SpeechRecognitionAudioSink::SpeechRecognitionAudioSink( |
| 16 const blink::WebMediaStreamTrack& track, |
| 17 const media::AudioParameters& params, |
| 18 const base::SharedMemoryHandle memory, |
| 19 scoped_ptr<base::SyncSocket> socket, |
| 20 const OnStoppedCB& on_stopped_cb) |
| 21 : track_(track), |
| 22 shared_memory_(memory, false), |
| 23 socket_(socket.Pass()), |
| 24 output_params_(params), |
| 25 track_stopped_(false), |
| 26 buffer_index_(0), |
| 27 on_stopped_cb_(on_stopped_cb) { |
| 28 DCHECK(socket_.get()); |
| 29 DCHECK(main_render_thread_checker_.CalledOnValidThread()); |
| 30 DCHECK(params.IsValid()); |
| 31 DCHECK(IsSupportedTrack(track)); |
| 32 const size_t memory_length = media::AudioBus::CalculateMemorySize(params) + |
| 33 sizeof(media::AudioInputBufferParameters); |
| 34 CHECK(shared_memory_.Map(memory_length)); |
| 35 |
| 36 // Peer's buffer index is accessed via |params.size| in shared memory. |
| 37 uint8* ptr = static_cast<uint8*>(shared_memory_.memory()); |
| 38 media::AudioInputBuffer* buffer = |
| 39 reinterpret_cast<media::AudioInputBuffer*>(ptr); |
| 40 peer_buffer_index_ = &(buffer->params.size); |
| 41 |
| 42 // The peer must manage their own counter and reset it. |
| 43 DCHECK_EQ(0U, *peer_buffer_index_); |
| 44 output_bus_ = media::AudioBus::WrapMemory(params, buffer->audio); |
| 45 |
| 46 // Connect this audio sink to the track |
| 47 MediaStreamAudioSink::AddToAudioTrack(this, track_); |
| 48 } |
| 49 |
| 50 SpeechRecognitionAudioSink::~SpeechRecognitionAudioSink() { |
| 51 DCHECK(main_render_thread_checker_.CalledOnValidThread()); |
| 52 if (audio_converter_.get()) |
| 53 audio_converter_->RemoveInput(this); |
| 54 |
| 55 // Notify the track before this sink goes away. |
| 56 if (!track_stopped_) |
| 57 MediaStreamAudioSink::RemoveFromAudioTrack(this, track_); |
| 58 } |
| 59 |
| 60 // static |
| 61 bool SpeechRecognitionAudioSink::IsSupportedTrack( |
| 62 const blink::WebMediaStreamTrack& track) { |
| 63 if (track.source().type() != blink::WebMediaStreamSource::TypeAudio) |
| 64 return false; |
| 65 |
| 66 MediaStreamAudioSource* native_source = |
| 67 static_cast<MediaStreamAudioSource*>(track.source().extraData()); |
| 68 if (!native_source) |
| 69 return false; |
| 70 |
| 71 const StreamDeviceInfo& device_info = native_source->device_info(); |
| 72 // Purposely only support tracks from an audio device. Dissallow WebAudio. |
| 73 return (device_info.device.type == content::MEDIA_DEVICE_AUDIO_CAPTURE); |
| 74 } |
| 75 |
| 76 void SpeechRecognitionAudioSink::OnSetFormat( |
| 77 const media::AudioParameters& input_params) { |
| 78 DCHECK(input_params.IsValid()); |
| 79 DCHECK_LE( |
| 80 input_params.frames_per_buffer() * 1000 / input_params.sample_rate(), |
| 81 output_params_.frames_per_buffer() * 1000 / output_params_.sample_rate()); |
| 82 |
| 83 // Detach the thread here because it will be a new capture thread |
| 84 // calling OnSetFormat() and OnData() if the source is restarted. |
| 85 capture_thread_checker_.DetachFromThread(); |
| 86 |
| 87 input_params_ = input_params; |
| 88 fifo_buffer_size_ = |
| 89 std::ceil(output_params_.frames_per_buffer() * |
| 90 static_cast<double>(input_params_.sample_rate()) / |
| 91 output_params_.sample_rate()); |
| 92 DCHECK_GE(fifo_buffer_size_, input_params_.frames_per_buffer()); |
| 93 |
| 94 // Allows for some delays on the peer. |
| 95 static const int kNumberOfBuffersInFifo = 2; |
| 96 int frames_in_fifo = kNumberOfBuffersInFifo * fifo_buffer_size_; |
| 97 fifo_.reset(new media::AudioFifo(input_params.channels(), frames_in_fifo)); |
| 98 input_bus_ = media::AudioBus::Create(input_params.channels(), |
| 99 input_params.frames_per_buffer()); |
| 100 |
| 101 // Create the audio converter with |disable_fifo| as false so that the |
| 102 // converter will request input_params.frames_per_buffer() each time. |
| 103 // This will not increase the complexity as there is only one client to |
| 104 // the converter. |
| 105 audio_converter_.reset( |
| 106 new media::AudioConverter(input_params, output_params_, false)); |
| 107 audio_converter_->AddInput(this); |
| 108 } |
| 109 |
| 110 void SpeechRecognitionAudioSink::OnReadyStateChanged( |
| 111 blink::WebMediaStreamSource::ReadyState state) { |
| 112 DCHECK(main_render_thread_checker_.CalledOnValidThread()); |
| 113 DCHECK(!track_stopped_); |
| 114 |
| 115 if (state == blink::WebMediaStreamSource::ReadyStateEnded) { |
| 116 track_stopped_ = true; |
| 117 |
| 118 if (!on_stopped_cb_.is_null()) |
| 119 on_stopped_cb_.Run(); |
| 120 } |
| 121 } |
| 122 |
| 123 void SpeechRecognitionAudioSink::OnData(const int16* audio_data, |
| 124 int sample_rate, |
| 125 int number_of_channels, |
| 126 int number_of_frames) { |
| 127 DCHECK(capture_thread_checker_.CalledOnValidThread()); |
| 128 DCHECK(peer_buffer_index_); |
| 129 DCHECK_EQ(input_bus_->frames(), number_of_frames); |
| 130 DCHECK_EQ(input_bus_->channels(), number_of_channels); |
| 131 if (fifo_->frames() + number_of_frames > fifo_->max_frames()) { |
| 132 // This would indicate a serious issue with the browser process or the |
| 133 // SyncSocket and/or SharedMemory. We stop delivering any data to the peer. |
| 134 NOTREACHED() << "Audio FIFO overflow"; |
| 135 return; |
| 136 } |
| 137 // TODO(xians): A better way to handle the interleaved and deinterleaved |
| 138 // format switching, see issue/317710. |
| 139 input_bus_->FromInterleaved(audio_data, number_of_frames, |
| 140 sizeof(audio_data[0])); |
| 141 |
| 142 fifo_->Push(input_bus_.get()); |
| 143 // Wait for FIFO to have at least |fifo_buffer_size_| frames ready. |
| 144 if (fifo_->frames() < fifo_buffer_size_) |
| 145 return; |
| 146 |
| 147 // Make sure the previous output buffer was consumed by the peer before we |
| 148 // send the next buffer. |peer_buffer_index_| is pointing to shared memory. |
| 149 // The peer must write to it (incrementing by 1) once the the buffer was |
| 150 // consumed. This is intentional not to block this audio capturing thread. |
| 151 if (buffer_index_ != (*peer_buffer_index_)) { |
| 152 DLOG(WARNING) << "Buffer synchronization lag"; |
| 153 return; |
| 154 } |
| 155 |
| 156 audio_converter_->Convert(output_bus_.get()); |
| 157 |
| 158 // Notify peer to consume buffer |buffer_index_| on |output_bus_|. |
| 159 const size_t bytes_sent = |
| 160 socket_->Send(&buffer_index_, sizeof(buffer_index_)); |
| 161 if (bytes_sent != sizeof(buffer_index_)) { |
| 162 // The send ocasionally fails if the user changes their input audio device. |
| 163 DVLOG(1) << "Failed sending buffer index to peer"; |
| 164 // We have discarded this buffer, but could still recover on the next one. |
| 165 return; |
| 166 } |
| 167 |
| 168 // Count the sent buffer. We expect the peer to do the same on their end. |
| 169 ++buffer_index_; |
| 170 } |
| 171 |
| 172 double SpeechRecognitionAudioSink::ProvideInput( |
| 173 media::AudioBus* audio_bus, base::TimeDelta buffer_delay) { |
| 174 DCHECK(capture_thread_checker_.CalledOnValidThread()); |
| 175 if (fifo_->frames() >= audio_bus->frames()) |
| 176 fifo_->Consume(audio_bus, 0, audio_bus->frames()); |
| 177 else |
| 178 audio_bus->Zero(); |
| 179 |
| 180 // Return volume greater than zero to indicate we have more data. |
| 181 return 1.0; |
| 182 } |
| 183 |
| 184 } // namespace content |
OLD | NEW |