content/renderer/media/media_stream_audio_processor.cc

   1 // Copyright 2013 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/media_stream_audio_processor.h"
   6
   7 #include "base/command_line.h"
   8 #include "base/metrics/field_trial.h"
   9 #include "base/metrics/histogram.h"
  10 #include "base/trace_event/trace_event.h"
  11 #include "content/public/common/content_switches.h"
  12 #include "content/renderer/media/media_stream_audio_processor_options.h"
  13 #include "content/renderer/media/rtc_media_constraints.h"
  14 #include "content/renderer/media/webrtc_audio_device_impl.h"
  15 #include "media/audio/audio_parameters.h"
  16 #include "media/base/audio_converter.h"
  17 #include "media/base/audio_fifo.h"
  18 #include "media/base/channel_layout.h"
  19 #include "third_party/WebKit/public/platform/WebMediaConstraints.h"
  20 #include "third_party/libjingle/source/talk/app/webrtc/mediaconstraintsinterface.h"
  21 #include "third_party/webrtc/modules/audio_processing/typing_detection.h"
  22
  23 namespace content {
  24
  25 namespace {
  26
  27 using webrtc::AudioProcessing;
  28 using webrtc::NoiseSuppression;
  29
  30 const int kAudioProcessingNumberOfChannels = 1;
  31
  32 AudioProcessing::ChannelLayout MapLayout(media::ChannelLayout media_layout) {
  33   switch (media_layout) {
  34     case media::CHANNEL_LAYOUT_MONO:
  35       return AudioProcessing::kMono;
  36     case media::CHANNEL_LAYOUT_STEREO:
  37       return AudioProcessing::kStereo;
  38     case media::CHANNEL_LAYOUT_STEREO_AND_KEYBOARD_MIC:
  39       return AudioProcessing::kStereoAndKeyboard;
  40     default:
  41       NOTREACHED() << "Layout not supported: " << media_layout;
  42       return AudioProcessing::kMono;
  43   }
  44 }
  45
  46 // This is only used for playout data where only max two channels is supported.
  47 AudioProcessing::ChannelLayout ChannelsToLayout(int num_channels) {
  48   switch (num_channels) {
  49     case 1:
  50       return AudioProcessing::kMono;
  51     case 2:
  52       return AudioProcessing::kStereo;
  53     default:
  54       NOTREACHED() << "Channels not supported: " << num_channels;
  55       return AudioProcessing::kMono;
  56   }
  57 }
  58
  59 // Used by UMA histograms and entries shouldn't be re-ordered or removed.
  60 enum AudioTrackProcessingStates {
  61   AUDIO_PROCESSING_ENABLED = 0,
  62   AUDIO_PROCESSING_DISABLED,
  63   AUDIO_PROCESSING_IN_WEBRTC,
  64   AUDIO_PROCESSING_MAX
  65 };
  66
  67 void RecordProcessingState(AudioTrackProcessingStates state) {
  68   UMA_HISTOGRAM_ENUMERATION("Media.AudioTrackProcessingStates",
  69                             state, AUDIO_PROCESSING_MAX);
  70 }
  71
  72 bool IsDelayAgnosticAecEnabled() {
  73   // Note: It's important to query the field trial state first, to ensure that
  74   // UMA reports the correct group.
  75   const std::string group_name =
  76       base::FieldTrialList::FindFullName("UseDelayAgnosticAEC");
  77   base::CommandLine* command_line = base::CommandLine::ForCurrentProcess();
  78   if (command_line->HasSwitch(switches::kEnableDelayAgnosticAec))
  79     return true;
  80   if (command_line->HasSwitch(switches::kDisableDelayAgnosticAec))
  81     return false;
  82
  83   return (group_name == "Enabled" || group_name == "DefaultEnabled");
  84 }
  85
  86 }  // namespace
  87
  88 // Wraps AudioBus to provide access to the array of channel pointers, since this
  89 // is the type webrtc::AudioProcessing deals in. The array is refreshed on every
  90 // channel_ptrs() call, and will be valid until the underlying AudioBus pointers
  91 // are changed, e.g. through calls to SetChannelData() or SwapChannels().
  92 //
  93 // All methods are called on one of the capture or render audio threads
  94 // exclusively.
  95 class MediaStreamAudioBus {
  96  public:
  97   MediaStreamAudioBus(int channels, int frames)
  98       : bus_(media::AudioBus::Create(channels, frames)),
  99         channel_ptrs_(new float*[channels]) {
 100     // May be created in the main render thread and used in the audio threads.
 101     thread_checker_.DetachFromThread();
 102   }
 103
 104   media::AudioBus* bus() {
 105     DCHECK(thread_checker_.CalledOnValidThread());
 106     return bus_.get();
 107   }
 108
 109   float* const* channel_ptrs() {
 110     DCHECK(thread_checker_.CalledOnValidThread());
 111     for (int i = 0; i < bus_->channels(); ++i) {
 112       channel_ptrs_[i] = bus_->channel(i);
 113     }
 114     return channel_ptrs_.get();
 115   }
 116
 117  private:
 118   base::ThreadChecker thread_checker_;
 119   scoped_ptr<media::AudioBus> bus_;
 120   scoped_ptr<float*[]> channel_ptrs_;
 121 };
 122
 123 // Wraps AudioFifo to provide a cleaner interface to MediaStreamAudioProcessor.
 124 // It avoids the FIFO when the source and destination frames match. All methods
 125 // are called on one of the capture or render audio threads exclusively. If
 126 // |source_channels| is larger than |destination_channels|, only the first
 127 // |destination_channels| are kept from the source.
 128 class MediaStreamAudioFifo {
 129  public:
 130   MediaStreamAudioFifo(int source_channels,
 131                        int destination_channels,
 132                        int source_frames,
 133                        int destination_frames,
 134                        int sample_rate)
 135      : source_channels_(source_channels),
 136        source_frames_(source_frames),
 137        sample_rate_(sample_rate),
 138        destination_(
 139            new MediaStreamAudioBus(destination_channels, destination_frames)),
 140        data_available_(false) {
 141     DCHECK_GE(source_channels, destination_channels);
 142     DCHECK_GT(sample_rate_, 0);
 143
 144     if (source_channels > destination_channels) {
 145       audio_source_intermediate_ =
 146           media::AudioBus::CreateWrapper(destination_channels);
 147     }
 148
 149     if (source_frames != destination_frames) {
 150       // Since we require every Push to be followed by as many Consumes as
 151       // possible, twice the larger of the two is a (probably) loose upper bound
 152       // on the FIFO size.
 153       const int fifo_frames = 2 * std::max(source_frames, destination_frames);
 154       fifo_.reset(new media::AudioFifo(destination_channels, fifo_frames));
 155     }
 156
 157     // May be created in the main render thread and used in the audio threads.
 158     thread_checker_.DetachFromThread();
 159   }
 160
 161   void Push(const media::AudioBus& source, base::TimeDelta audio_delay) {
 162     DCHECK(thread_checker_.CalledOnValidThread());
 163     DCHECK_EQ(source.channels(), source_channels_);
 164     DCHECK_EQ(source.frames(), source_frames_);
 165
 166     const media::AudioBus* source_to_push = &source;
 167
 168     if (audio_source_intermediate_) {
 169       for (int i = 0; i < destination_->bus()->channels(); ++i) {
 170         audio_source_intermediate_->SetChannelData(
 171             i,
 172             const_cast<float*>(source.channel(i)));
 173       }
 174       audio_source_intermediate_->set_frames(source.frames());
 175       source_to_push = audio_source_intermediate_.get();
 176     }
 177
 178     if (fifo_) {
 179       CHECK_LT(fifo_->frames(), destination_->bus()->frames());
 180       next_audio_delay_ = audio_delay +
 181           fifo_->frames() * base::TimeDelta::FromSeconds(1) / sample_rate_;
 182       fifo_->Push(source_to_push);
 183     } else {
 184       CHECK(!data_available_);
 185       source_to_push->CopyTo(destination_->bus());
 186       next_audio_delay_ = audio_delay;
 187       data_available_ = true;
 188     }
 189   }
 190
 191   // Returns true if there are destination_frames() of data available to be
 192   // consumed, and otherwise false.
 193   bool Consume(MediaStreamAudioBus** destination,
 194                base::TimeDelta* audio_delay) {
 195     DCHECK(thread_checker_.CalledOnValidThread());
 196
 197     if (fifo_) {
 198       if (fifo_->frames() < destination_->bus()->frames())
 199         return false;
 200
 201       fifo_->Consume(destination_->bus(), 0, destination_->bus()->frames());
 202       *audio_delay = next_audio_delay_;
 203       next_audio_delay_ -=
 204           destination_->bus()->frames() * base::TimeDelta::FromSeconds(1) /
 205               sample_rate_;
 206     } else {
 207       if (!data_available_)
 208         return false;
 209       *audio_delay = next_audio_delay_;
 210       // The data was already copied to |destination_| in this case.
 211       data_available_ = false;
 212     }
 213
 214     *destination = destination_.get();
 215     return true;
 216   }
 217
 218  private:
 219   base::ThreadChecker thread_checker_;
 220   const int source_channels_;  // For a DCHECK.
 221   const int source_frames_;  // For a DCHECK.
 222   const int sample_rate_;
 223   scoped_ptr<media::AudioBus> audio_source_intermediate_;
 224   scoped_ptr<MediaStreamAudioBus> destination_;
 225   scoped_ptr<media::AudioFifo> fifo_;
 226
 227   // When using |fifo_|, this is the audio delay of the first sample to be
 228   // consumed next from the FIFO.  When not using |fifo_|, this is the audio
 229   // delay of the first sample in |destination_|.
 230   base::TimeDelta next_audio_delay_;
 231
 232   // True when |destination_| contains the data to be returned by the next call
 233   // to Consume().  Only used when the FIFO is disabled.
 234   bool data_available_;
 235 };
 236
 237 MediaStreamAudioProcessor::MediaStreamAudioProcessor(
 238     const blink::WebMediaConstraints& constraints,
 239     const MediaStreamDevice::AudioDeviceParameters& input_params,
 240     WebRtcPlayoutDataSource* playout_data_source)
 241     : render_delay_ms_(0),
 242       playout_data_source_(playout_data_source),
 243       audio_mirroring_(false),
 244       typing_detected_(false),
 245       stopped_(false) {
 246   capture_thread_checker_.DetachFromThread();
 247   render_thread_checker_.DetachFromThread();
 248   InitializeAudioProcessingModule(constraints, input_params);
 249
 250   aec_dump_message_filter_ = AecDumpMessageFilter::Get();
 251   // In unit tests not creating a message filter, |aec_dump_message_filter_|
 252   // will be NULL. We can just ignore that. Other unit tests and browser tests
 253   // ensure that we do get the filter when we should.
 254   if (aec_dump_message_filter_.get())
 255     aec_dump_message_filter_->AddDelegate(this);
 256 }
 257
 258 MediaStreamAudioProcessor::~MediaStreamAudioProcessor() {
 259   DCHECK(main_thread_checker_.CalledOnValidThread());
 260   Stop();
 261 }
 262
 263 void MediaStreamAudioProcessor::OnCaptureFormatChanged(
 264     const media::AudioParameters& input_format) {
 265   DCHECK(main_thread_checker_.CalledOnValidThread());
 266   // There is no need to hold a lock here since the caller guarantees that
 267   // there is no more PushCaptureData() and ProcessAndConsumeData() callbacks
 268   // on the capture thread.
 269   InitializeCaptureFifo(input_format);
 270
 271   // Reset the |capture_thread_checker_| since the capture data will come from
 272   // a new capture thread.
 273   capture_thread_checker_.DetachFromThread();
 274 }
 275
 276 void MediaStreamAudioProcessor::PushCaptureData(
 277     const media::AudioBus& audio_source,
 278     base::TimeDelta capture_delay) {
 279   DCHECK(capture_thread_checker_.CalledOnValidThread());
 280
 281   capture_fifo_->Push(audio_source, capture_delay);
 282 }
 283
 284 bool MediaStreamAudioProcessor::ProcessAndConsumeData(
 285     int volume,
 286     bool key_pressed,
 287     media::AudioBus** processed_data,
 288     base::TimeDelta* capture_delay,
 289     int* new_volume) {
 290   DCHECK(capture_thread_checker_.CalledOnValidThread());
 291   DCHECK(processed_data);
 292   DCHECK(capture_delay);
 293   DCHECK(new_volume);
 294
 295   TRACE_EVENT0("audio", "MediaStreamAudioProcessor::ProcessAndConsumeData");
 296
 297   MediaStreamAudioBus* process_bus;
 298   if (!capture_fifo_->Consume(&process_bus, capture_delay))
 299     return false;
 300
 301   // Use the process bus directly if audio processing is disabled.
 302   MediaStreamAudioBus* output_bus = process_bus;
 303   *new_volume = 0;
 304   if (audio_processing_) {
 305     output_bus = output_bus_.get();
 306     *new_volume = ProcessData(process_bus->channel_ptrs(),
 307                               process_bus->bus()->frames(), *capture_delay,
 308                               volume, key_pressed, output_bus->channel_ptrs());
 309   }
 310
 311   // Swap channels before interleaving the data.
 312   if (audio_mirroring_ &&
 313       output_format_.channel_layout() == media::CHANNEL_LAYOUT_STEREO) {
 314     // Swap the first and second channels.
 315     output_bus->bus()->SwapChannels(0, 1);
 316   }
 317
 318   *processed_data = output_bus->bus();
 319
 320   return true;
 321 }
 322
 323 void MediaStreamAudioProcessor::Stop() {
 324   DCHECK(main_thread_checker_.CalledOnValidThread());
 325   if (stopped_)
 326     return;
 327
 328   stopped_ = true;
 329
 330   if (aec_dump_message_filter_.get()) {
 331     aec_dump_message_filter_->RemoveDelegate(this);
 332     aec_dump_message_filter_ = NULL;
 333   }
 334
 335   if (!audio_processing_.get())
 336     return;
 337
 338   audio_processing_.get()->UpdateHistogramsOnCallEnd();
 339   StopEchoCancellationDump(audio_processing_.get());
 340
 341   if (playout_data_source_) {
 342     playout_data_source_->RemovePlayoutSink(this);
 343     playout_data_source_ = NULL;
 344   }
 345 }
 346
 347 const media::AudioParameters& MediaStreamAudioProcessor::InputFormat() const {
 348   return input_format_;
 349 }
 350
 351 const media::AudioParameters& MediaStreamAudioProcessor::OutputFormat() const {
 352   return output_format_;
 353 }
 354
 355 void MediaStreamAudioProcessor::OnAecDumpFile(
 356     const IPC::PlatformFileForTransit& file_handle) {
 357   DCHECK(main_thread_checker_.CalledOnValidThread());
 358
 359   base::File file = IPC::PlatformFileForTransitToFile(file_handle);
 360   DCHECK(file.IsValid());
 361
 362   if (audio_processing_)
 363     StartEchoCancellationDump(audio_processing_.get(), file.Pass());
 364   else
 365     file.Close();
 366 }
 367
 368 void MediaStreamAudioProcessor::OnDisableAecDump() {
 369   DCHECK(main_thread_checker_.CalledOnValidThread());
 370   if (audio_processing_)
 371     StopEchoCancellationDump(audio_processing_.get());
 372 }
 373
 374 void MediaStreamAudioProcessor::OnIpcClosing() {
 375   DCHECK(main_thread_checker_.CalledOnValidThread());
 376   aec_dump_message_filter_ = NULL;
 377 }
 378
 379 void MediaStreamAudioProcessor::OnPlayoutData(media::AudioBus* audio_bus,
 380                                               int sample_rate,
 381                                               int audio_delay_milliseconds) {
 382   DCHECK(render_thread_checker_.CalledOnValidThread());
 383   DCHECK(audio_processing_->echo_control_mobile()->is_enabled() ^
 384          audio_processing_->echo_cancellation()->is_enabled());
 385
 386   TRACE_EVENT0("audio", "MediaStreamAudioProcessor::OnPlayoutData");
 387   DCHECK_LT(audio_delay_milliseconds,
 388             std::numeric_limits<base::subtle::Atomic32>::max());
 389   base::subtle::Release_Store(&render_delay_ms_, audio_delay_milliseconds);
 390
 391   InitializeRenderFifoIfNeeded(sample_rate, audio_bus->channels(),
 392                                audio_bus->frames());
 393
 394   render_fifo_->Push(
 395       *audio_bus, base::TimeDelta::FromMilliseconds(audio_delay_milliseconds));
 396   MediaStreamAudioBus* analysis_bus;
 397   base::TimeDelta audio_delay;
 398   while (render_fifo_->Consume(&analysis_bus, &audio_delay)) {
 399     // TODO(ajm): Should AnalyzeReverseStream() account for the |audio_delay|?
 400     audio_processing_->AnalyzeReverseStream(
 401         analysis_bus->channel_ptrs(),
 402         analysis_bus->bus()->frames(),
 403         sample_rate,
 404         ChannelsToLayout(audio_bus->channels()));
 405   }
 406 }
 407
 408 void MediaStreamAudioProcessor::OnPlayoutDataSourceChanged() {
 409   DCHECK(main_thread_checker_.CalledOnValidThread());
 410   // There is no need to hold a lock here since the caller guarantees that
 411   // there is no more OnPlayoutData() callback on the render thread.
 412   render_thread_checker_.DetachFromThread();
 413   render_fifo_.reset();
 414 }
 415
 416 void MediaStreamAudioProcessor::GetStats(AudioProcessorStats* stats) {
 417   stats->typing_noise_detected =
 418       (base::subtle::Acquire_Load(&typing_detected_) != false);
 419   GetAecStats(audio_processing_.get()->echo_cancellation(), stats);
 420 }
 421
 422 void MediaStreamAudioProcessor::InitializeAudioProcessingModule(
 423     const blink::WebMediaConstraints& constraints,
 424     const MediaStreamDevice::AudioDeviceParameters& input_params) {
 425   DCHECK(main_thread_checker_.CalledOnValidThread());
 426   DCHECK(!audio_processing_);
 427
 428   MediaAudioConstraints audio_constraints(constraints, input_params.effects);
 429
 430   // Audio mirroring can be enabled even though audio processing is otherwise
 431   // disabled.
 432   audio_mirroring_ = audio_constraints.GetProperty(
 433       MediaAudioConstraints::kGoogAudioMirroring);
 434
 435 #if defined(OS_IOS)
 436   // On iOS, VPIO provides built-in AGC and AEC.
 437   const bool echo_cancellation = false;
 438   const bool goog_agc = false;
 439 #else
 440   const bool echo_cancellation =
 441       audio_constraints.GetEchoCancellationProperty();
 442   const bool goog_agc = audio_constraints.GetProperty(
 443       MediaAudioConstraints::kGoogAutoGainControl);
 444 #endif
 445
 446 #if defined(OS_IOS) || defined(OS_ANDROID)
 447   const bool goog_experimental_aec = false;
 448   const bool goog_typing_detection = false;
 449 #else
 450   const bool goog_experimental_aec = audio_constraints.GetProperty(
 451       MediaAudioConstraints::kGoogExperimentalEchoCancellation);
 452   const bool goog_typing_detection = audio_constraints.GetProperty(
 453       MediaAudioConstraints::kGoogTypingNoiseDetection);
 454 #endif
 455
 456   const bool goog_ns = audio_constraints.GetProperty(
 457       MediaAudioConstraints::kGoogNoiseSuppression);
 458   const bool goog_experimental_ns = audio_constraints.GetProperty(
 459       MediaAudioConstraints::kGoogExperimentalNoiseSuppression);
 460   const bool goog_beamforming = audio_constraints.GetProperty(
 461       MediaAudioConstraints::kGoogBeamforming);
 462   const bool goog_high_pass_filter = audio_constraints.GetProperty(
 463       MediaAudioConstraints::kGoogHighpassFilter);
 464   // Return immediately if no goog constraint is enabled.
 465   if (!echo_cancellation && !goog_experimental_aec && !goog_ns &&
 466       !goog_high_pass_filter && !goog_typing_detection &&
 467       !goog_agc && !goog_experimental_ns && !goog_beamforming) {
 468     RecordProcessingState(AUDIO_PROCESSING_DISABLED);
 469     return;
 470   }
 471
 472   // Experimental options provided at creation.
 473   webrtc::Config config;
 474   if (goog_experimental_aec)
 475     config.Set<webrtc::ExtendedFilter>(new webrtc::ExtendedFilter(true));
 476   if (goog_experimental_ns)
 477     config.Set<webrtc::ExperimentalNs>(new webrtc::ExperimentalNs(true));
 478   if (IsDelayAgnosticAecEnabled())
 479     config.Set<webrtc::DelayAgnostic>(new webrtc::DelayAgnostic(true));
 480   if (goog_beamforming) {
 481     const auto& geometry =
 482         GetArrayGeometryPreferringConstraints(audio_constraints, input_params);
 483
 484     // Only enable beamforming if we have at least two mics.
 485     config.Set<webrtc::Beamforming>(
 486         new webrtc::Beamforming(geometry.size() > 1, geometry));
 487   }
 488
 489   // Create and configure the webrtc::AudioProcessing.
 490   audio_processing_.reset(webrtc::AudioProcessing::Create(config));
 491
 492   // Enable the audio processing components.
 493   if (echo_cancellation) {
 494     EnableEchoCancellation(audio_processing_.get());
 495
 496     if (playout_data_source_)
 497       playout_data_source_->AddPlayoutSink(this);
 498
 499     // Prepare for logging echo information. If there are data remaining in
 500     // |echo_information_| we simply discard it.
 501     echo_information_.reset(new EchoInformation());
 502   }
 503
 504   if (goog_ns) {
 505     // The beamforming postfilter is effective at suppressing stationary noise,
 506     // so reduce the single-channel NS aggressiveness when enabled.
 507     const NoiseSuppression::Level ns_level =
 508         config.Get<webrtc::Beamforming>().enabled ? NoiseSuppression::kLow
 509                                                   : NoiseSuppression::kHigh;
 510
 511     EnableNoiseSuppression(audio_processing_.get(), ns_level);
 512   }
 513
 514   if (goog_high_pass_filter)
 515     EnableHighPassFilter(audio_processing_.get());
 516
 517   if (goog_typing_detection) {
 518     // TODO(xians): Remove this |typing_detector_| after the typing suppression
 519     // is enabled by default.
 520     typing_detector_.reset(new webrtc::TypingDetection());
 521     EnableTypingDetection(audio_processing_.get(), typing_detector_.get());
 522   }
 523
 524   if (goog_agc)
 525     EnableAutomaticGainControl(audio_processing_.get());
 526
 527   RecordProcessingState(AUDIO_PROCESSING_ENABLED);
 528 }
 529
 530 void MediaStreamAudioProcessor::InitializeCaptureFifo(
 531     const media::AudioParameters& input_format) {
 532   DCHECK(main_thread_checker_.CalledOnValidThread());
 533   DCHECK(input_format.IsValid());
 534   input_format_ = input_format;
 535
 536   // TODO(ajm): For now, we assume fixed parameters for the output when audio
 537   // processing is enabled, to match the previous behavior. We should either
 538   // use the input parameters (in which case, audio processing will convert
 539   // at output) or ideally, have a backchannel from the sink to know what
 540   // format it would prefer.
 541 #if defined(OS_ANDROID)
 542   int audio_processing_sample_rate = AudioProcessing::kSampleRate16kHz;
 543 #else
 544   int audio_processing_sample_rate = AudioProcessing::kSampleRate48kHz;
 545 #endif
 546   const int output_sample_rate = audio_processing_ ?
 547                                  audio_processing_sample_rate :
 548                                  input_format.sample_rate();
 549   media::ChannelLayout output_channel_layout = audio_processing_ ?
 550       media::GuessChannelLayout(kAudioProcessingNumberOfChannels) :
 551       input_format.channel_layout();
 552
 553   // The output channels from the fifo is normally the same as input.
 554   int fifo_output_channels = input_format.channels();
 555
 556   // Special case for if we have a keyboard mic channel on the input and no
 557   // audio processing is used. We will then have the fifo strip away that
 558   // channel. So we use stereo as output layout, and also change the output
 559   // channels for the fifo.
 560   if (input_format.channel_layout() ==
 561           media::CHANNEL_LAYOUT_STEREO_AND_KEYBOARD_MIC &&
 562       !audio_processing_) {
 563     output_channel_layout = media::CHANNEL_LAYOUT_STEREO;
 564     fifo_output_channels = ChannelLayoutToChannelCount(output_channel_layout);
 565   }
 566
 567   // webrtc::AudioProcessing requires a 10 ms chunk size. We use this native
 568   // size when processing is enabled. When disabled we use the same size as
 569   // the source if less than 10 ms.
 570   //
 571   // TODO(ajm): This conditional buffer size appears to be assuming knowledge of
 572   // the sink based on the source parameters. PeerConnection sinks seem to want
 573   // 10 ms chunks regardless, while WebAudio sinks want less, and we're assuming
 574   // we can identify WebAudio sinks by the input chunk size. Less fragile would
 575   // be to have the sink actually tell us how much it wants (as in the above
 576   // todo).
 577   int processing_frames = input_format.sample_rate() / 100;
 578   int output_frames = output_sample_rate / 100;
 579   if (!audio_processing_ && input_format.frames_per_buffer() < output_frames) {
 580     processing_frames = input_format.frames_per_buffer();
 581     output_frames = processing_frames;
 582   }
 583
 584   output_format_ = media::AudioParameters(
 585       media::AudioParameters::AUDIO_PCM_LOW_LATENCY,
 586       output_channel_layout,
 587       output_sample_rate,
 588       16,
 589       output_frames);
 590
 591   capture_fifo_.reset(
 592       new MediaStreamAudioFifo(input_format.channels(),
 593                                fifo_output_channels,
 594                                input_format.frames_per_buffer(),
 595                                processing_frames,
 596                                input_format.sample_rate()));
 597
 598   if (audio_processing_) {
 599     output_bus_.reset(new MediaStreamAudioBus(output_format_.channels(),
 600                                               output_frames));
 601   }
 602 }
 603
 604 void MediaStreamAudioProcessor::InitializeRenderFifoIfNeeded(
 605     int sample_rate, int number_of_channels, int frames_per_buffer) {
 606   DCHECK(render_thread_checker_.CalledOnValidThread());
 607   if (render_fifo_.get() &&
 608       render_format_.sample_rate() == sample_rate &&
 609       render_format_.channels() == number_of_channels &&
 610       render_format_.frames_per_buffer() == frames_per_buffer) {
 611     // Do nothing if the |render_fifo_| has been setup properly.
 612     return;
 613   }
 614
 615   render_format_ = media::AudioParameters(
 616       media::AudioParameters::AUDIO_PCM_LOW_LATENCY,
 617       media::GuessChannelLayout(number_of_channels),
 618       sample_rate,
 619       16,
 620       frames_per_buffer);
 621
 622   const int analysis_frames = sample_rate / 100;  // 10 ms chunks.
 623   render_fifo_.reset(
 624       new MediaStreamAudioFifo(number_of_channels,
 625                                number_of_channels,
 626                                frames_per_buffer,
 627                                analysis_frames,
 628                                sample_rate));
 629 }
 630
 631 int MediaStreamAudioProcessor::ProcessData(const float* const* process_ptrs,
 632                                            int process_frames,
 633                                            base::TimeDelta capture_delay,
 634                                            int volume,
 635                                            bool key_pressed,
 636                                            float* const* output_ptrs) {
 637   DCHECK(audio_processing_);
 638   DCHECK(capture_thread_checker_.CalledOnValidThread());
 639
 640   TRACE_EVENT0("audio", "MediaStreamAudioProcessor::ProcessData");
 641
 642   base::subtle::Atomic32 render_delay_ms =
 643       base::subtle::Acquire_Load(&render_delay_ms_);
 644   int64 capture_delay_ms = capture_delay.InMilliseconds();
 645   DCHECK_LT(capture_delay_ms,
 646             std::numeric_limits<base::subtle::Atomic32>::max());
 647   int total_delay_ms =  capture_delay_ms + render_delay_ms;
 648   if (total_delay_ms > 300) {
 649     LOG(WARNING) << "Large audio delay, capture delay: " << capture_delay_ms
 650                  << "ms; render delay: " << render_delay_ms << "ms";
 651   }
 652
 653   webrtc::AudioProcessing* ap = audio_processing_.get();
 654   ap->set_stream_delay_ms(total_delay_ms);
 655
 656   DCHECK_LE(volume, WebRtcAudioDeviceImpl::kMaxVolumeLevel);
 657   webrtc::GainControl* agc = ap->gain_control();
 658   int err = agc->set_stream_analog_level(volume);
 659   DCHECK_EQ(err, 0) << "set_stream_analog_level() error: " << err;
 660
 661   ap->set_stream_key_pressed(key_pressed);
 662
 663   err = ap->ProcessStream(process_ptrs,
 664                           process_frames,
 665                           input_format_.sample_rate(),
 666                           MapLayout(input_format_.channel_layout()),
 667                           output_format_.sample_rate(),
 668                           MapLayout(output_format_.channel_layout()),
 669                           output_ptrs);
 670   DCHECK_EQ(err, 0) << "ProcessStream() error: " << err;
 671
 672   if (typing_detector_) {
 673     webrtc::VoiceDetection* vad = ap->voice_detection();
 674     DCHECK(vad->is_enabled());
 675     bool detected = typing_detector_->Process(key_pressed,
 676                                               vad->stream_has_voice());
 677     base::subtle::Release_Store(&typing_detected_, detected);
 678   }
 679
 680   if (echo_information_) {
 681     echo_information_.get()->UpdateAecDelayStats(ap->echo_cancellation());
 682   }
 683
 684   // Return 0 if the volume hasn't been changed, and otherwise the new volume.
 685   return (agc->stream_analog_level() == volume) ?
 686       0 : agc->stream_analog_level();
 687 }
 688
 689 }  // namespace content