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, const base::SharedMemoryHandle memory, | |
18 scoped_ptr<base::SyncSocket> socket, OnStoppedCB on_stopped_cb) | |
19 : track_(track), | |
20 shared_memory_(memory, false), | |
21 socket_(socket.Pass()), | |
22 output_params_(params), | |
23 track_stopped_(false), | |
24 buffer_index_(0), | |
25 on_stopped_cb_(on_stopped_cb) { | |
26 DCHECK(socket_.get()); | |
27 DCHECK(main_render_thread_checker_.CalledOnValidThread()); | |
28 DCHECK(params.IsValid()); | |
29 DCHECK(IsSupportedTrack(track)); | |
30 const size_t memory_length = media::AudioBus::CalculateMemorySize(params) + | |
31 sizeof(media::AudioInputBufferParameters); | |
32 CHECK(shared_memory_.Map(memory_length)); | |
33 | |
34 // Buffer index for sync with client is |params.size| on the shared memory. | |
35 uint8* ptr = static_cast<uint8*>(shared_memory_.memory()); | |
36 media::AudioInputBuffer* buffer = | |
37 reinterpret_cast<media::AudioInputBuffer*>(ptr); | |
38 peer_buffer_index_ = &(buffer->params.size); | |
39 | |
40 // Client must manage his own counter and reset it. | |
41 DCHECK_EQ(0U, *peer_buffer_index_); | |
42 output_bus_ = media::AudioBus::WrapMemory(params, buffer->audio); | |
43 | |
44 // Connect this audio sink to the track | |
45 MediaStreamAudioSink::AddToAudioTrack(this, track_); | |
46 } | |
47 | |
48 SpeechRecognitionAudioSink::~SpeechRecognitionAudioSink() { | |
49 DCHECK(main_render_thread_checker_.CalledOnValidThread()); | |
50 if (audio_converter_.get()) | |
51 audio_converter_->RemoveInput(this); | |
52 | |
53 // Notify the track before this sink goes away. | |
54 if (!track_stopped_) | |
55 MediaStreamAudioSink::RemoveFromAudioTrack(this, track_); | |
56 } | |
57 | |
58 // static | |
59 bool SpeechRecognitionAudioSink::IsSupportedTrack( | |
60 const blink::WebMediaStreamTrack& track) { | |
61 if (track.source().type() != blink::WebMediaStreamSource::TypeAudio) | |
62 return false; | |
63 | |
64 MediaStreamAudioSource* native_source = | |
65 static_cast<MediaStreamAudioSource*>(track.source().extraData()); | |
66 if (!native_source) | |
67 return false; | |
68 | |
69 const StreamDeviceInfo& device_info = native_source->device_info(); | |
70 // Purposely only support tracks from an audio device. Dissallow WebAudio. | |
71 return (device_info.device.type == content::MEDIA_DEVICE_AUDIO_CAPTURE); | |
72 } | |
73 | |
74 void SpeechRecognitionAudioSink::OnSetFormat( | |
75 const media::AudioParameters& input_params) { | |
76 DCHECK(input_params.IsValid()); | |
77 DCHECK_LE( | |
78 input_params.frames_per_buffer() * 1000 / input_params.sample_rate(), | |
79 output_params_.frames_per_buffer() * 1000 / output_params_.sample_rate()); | |
80 | |
81 // We need detach the thread here because it will be a new capture thread | |
82 // calling OnSetFormat() and OnData() if the source is restarted. | |
83 capture_thread_checker_.DetachFromThread(); | |
84 | |
85 input_params_ = input_params; | |
86 fifo_buffer_size_ = | |
87 std::ceil(output_params_.frames_per_buffer() * | |
88 static_cast<double>(input_params_.sample_rate()) / | |
89 output_params_.sample_rate()); | |
no longer working on chromium
2014/09/29 09:28:57
nit, indentation should look:
std::ceil(output_par
burnik
2014/09/29 10:24:31
Done.
| |
90 DCHECK_GE(fifo_buffer_size_, input_params_.frames_per_buffer()); | |
91 | |
92 // Allows for some delays on the endpoint client. | |
93 static const int kNumberOfBuffersInFifo = 2; | |
94 int frames_in_fifo = kNumberOfBuffersInFifo * fifo_buffer_size_; | |
95 fifo_.reset(new media::AudioFifo(input_params.channels(), frames_in_fifo)); | |
96 input_bus_ = media::AudioBus::Create(input_params.channels(), | |
97 input_params.frames_per_buffer()); | |
98 | |
99 // Create the audio converter with |disable_fifo| as false so that the | |
100 // converter will request input_params.frames_per_buffer() each time. | |
101 // This will not increase the complexity as there is only one client to | |
102 // the converter. | |
103 audio_converter_.reset( | |
104 new media::AudioConverter(input_params, output_params_, false)); | |
no longer working on chromium
2014/09/29 09:28:56
AudioConverter has a better performance when the f
burnik
2014/09/29 10:24:31
No, it doesn't work.
| |
105 audio_converter_->AddInput(this); | |
106 } | |
107 | |
108 void SpeechRecognitionAudioSink::OnReadyStateChanged( | |
109 blink::WebMediaStreamSource::ReadyState state) { | |
110 DCHECK(main_render_thread_checker_.CalledOnValidThread()); | |
111 DCHECK(!track_stopped_); | |
112 | |
113 if (state == blink::WebMediaStreamSource::ReadyStateEnded) { | |
114 track_stopped_ = true; | |
115 | |
116 if (!on_stopped_cb_.is_null()) | |
117 on_stopped_cb_.Run(); | |
118 } | |
119 } | |
120 | |
121 void SpeechRecognitionAudioSink::OnData(const int16* audio_data, | |
122 int sample_rate, | |
123 int number_of_channels, | |
124 int number_of_frames) { | |
125 DCHECK(capture_thread_checker_.CalledOnValidThread()); | |
126 DCHECK(peer_buffer_index_); | |
127 DCHECK_EQ(input_bus_->frames(), number_of_frames); | |
128 DCHECK_EQ(input_bus_->channels(), number_of_channels); | |
129 if (fifo_->frames() + number_of_frames > fifo_->max_frames()) { | |
130 // This would indicate a serious issue with the browser process or the | |
131 // SyncSocket and/or SharedMemory. We stop delivering any data to the peer. | |
132 NOTREACHED() << "Audio FIFO overflow"; | |
133 return; | |
134 } | |
135 // TODO(xians): A better way to handle the interleaved and deinterleaved | |
136 // format switching, see issue/317710. | |
137 input_bus_->FromInterleaved(audio_data, number_of_frames, | |
138 sizeof(audio_data[0])); | |
139 | |
140 fifo_->Push(input_bus_.get()); | |
141 // Wait for FIFO to have at least |fifo_buffer_size_| frames ready. | |
142 if (fifo_->frames() < fifo_buffer_size_) | |
143 return; | |
144 | |
145 // Make sure the previous output buffer was consumed by client before we send | |
146 // the next buffer. |peer_buffer_index_| is pointing to shared memory. | |
147 // The client must write to it (incrementing by 1) once the the buffer was | |
148 // consumed. This is intentional not to block this audio capturing thread. | |
149 if (buffer_index_ != (*peer_buffer_index_)) { | |
150 DLOG(WARNING) << "Buffer synchronization lag"; | |
151 return; | |
152 } | |
153 | |
154 audio_converter_->Convert(output_bus_.get()); | |
155 | |
156 // Notify client to consume buffer |buffer_index_| on |output_bus_|. | |
157 const size_t bytes_sent = | |
158 socket_->Send(&buffer_index_, sizeof(buffer_index_)); | |
159 if (bytes_sent != sizeof(buffer_index_)) { | |
160 // The send ocasionally fails if the user changes his input audio device. | |
161 DVLOG(1) << "Failed sending buffer index to peer"; | |
162 // We have discarded this buffer, but could still recover on the next one. | |
163 return; | |
164 } | |
165 | |
166 // Count the sent buffer. We expect the client to do the same on his end. | |
167 ++buffer_index_; | |
168 } | |
169 | |
170 double SpeechRecognitionAudioSink::ProvideInput( | |
171 media::AudioBus* audio_bus, base::TimeDelta buffer_delay) { | |
172 DCHECK(capture_thread_checker_.CalledOnValidThread()); | |
173 if (fifo_->frames() >= audio_bus->frames()) | |
174 fifo_->Consume(audio_bus, 0, audio_bus->frames()); | |
175 else | |
176 audio_bus->Zero(); | |
177 | |
178 // Return volume greater than zero to indicate we have more data. | |
179 return 1.0; | |
180 } | |
181 | |
182 } // namespace content | |
OLD | NEW |