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 #if defined(OS_CHROMEOS)
  24 #include "base/sys_info.h"
  25 #endif
  26
  27 namespace content {
  28
  29 namespace {
  30
  31 using webrtc::AudioProcessing;
  32 using webrtc::NoiseSuppression;
  33
  34 const int kAudioProcessingNumberOfChannels = 1;
  35
  36 AudioProcessing::ChannelLayout MapLayout(media::ChannelLayout media_layout) {
  37   switch (media_layout) {
  38     case media::CHANNEL_LAYOUT_MONO:
  39       return AudioProcessing::kMono;
  40     case media::CHANNEL_LAYOUT_STEREO:
  41       return AudioProcessing::kStereo;
  42     case media::CHANNEL_LAYOUT_STEREO_AND_KEYBOARD_MIC:
  43       return AudioProcessing::kStereoAndKeyboard;
  44     default:
  45       NOTREACHED() << "Layout not supported: " << media_layout;
  46       return AudioProcessing::kMono;
  47   }
  48 }
  49
  50 // This is only used for playout data where only max two channels is supported.
  51 AudioProcessing::ChannelLayout ChannelsToLayout(int num_channels) {
  52   switch (num_channels) {
  53     case 1:
  54       return AudioProcessing::kMono;
  55     case 2:
  56       return AudioProcessing::kStereo;
  57     default:
  58       NOTREACHED() << "Channels not supported: " << num_channels;
  59       return AudioProcessing::kMono;
  60   }
  61 }
  62
  63 // Used by UMA histograms and entries shouldn't be re-ordered or removed.
  64 enum AudioTrackProcessingStates {
  65   AUDIO_PROCESSING_ENABLED = 0,
  66   AUDIO_PROCESSING_DISABLED,
  67   AUDIO_PROCESSING_IN_WEBRTC,
  68   AUDIO_PROCESSING_MAX
  69 };
  70
  71 void RecordProcessingState(AudioTrackProcessingStates state) {
  72   UMA_HISTOGRAM_ENUMERATION("Media.AudioTrackProcessingStates",
  73                             state, AUDIO_PROCESSING_MAX);
  74 }
  75
  76 bool IsDelayAgnosticAecEnabled() {
  77   // Note: It's important to query the field trial state first, to ensure that
  78   // UMA reports the correct group.
  79   const std::string group_name =
  80       base::FieldTrialList::FindFullName("UseDelayAgnosticAEC");
  81   base::CommandLine* command_line = base::CommandLine::ForCurrentProcess();
  82   if (command_line->HasSwitch(switches::kEnableDelayAgnosticAec))
  83     return true;
  84   if (command_line->HasSwitch(switches::kDisableDelayAgnosticAec))
  85     return false;
  86
  87   return (group_name == "Enabled" || group_name == "DefaultEnabled");
  88 }
  89
  90 bool IsBeamformingEnabled(const MediaAudioConstraints& audio_constraints) {
  91   return base::FieldTrialList::FindFullName("ChromebookBeamforming") ==
  92              "Enabled" ||
  93          audio_constraints.GetProperty(MediaAudioConstraints::kGoogBeamforming);
  94 }
  95
  96 void ConfigureBeamforming(webrtc::Config* config,
  97                           const std::string& geometry_str) {
  98   std::vector<webrtc::Point> geometry = ParseArrayGeometry(geometry_str);
  99 #if defined(OS_CHROMEOS)
 100   if (geometry.empty()) {
 101     const std::string& board = base::SysInfo::GetLsbReleaseBoard();
 102     if (board.find("nyan_kitty") != std::string::npos) {
 103       geometry.push_back(webrtc::Point(-0.03f, 0.f, 0.f));
 104       geometry.push_back(webrtc::Point(0.03f, 0.f, 0.f));
 105     } else if (board.find("peach_pi") != std::string::npos) {
 106       geometry.push_back(webrtc::Point(-0.025f, 0.f, 0.f));
 107       geometry.push_back(webrtc::Point(0.025f, 0.f, 0.f));
 108     } else if (board.find("samus") != std::string::npos) {
 109       geometry.push_back(webrtc::Point(-0.032f, 0.f, 0.f));
 110       geometry.push_back(webrtc::Point(0.032f, 0.f, 0.f));
 111     } else if (board.find("swanky") != std::string::npos) {
 112       geometry.push_back(webrtc::Point(-0.026f, 0.f, 0.f));
 113       geometry.push_back(webrtc::Point(0.026f, 0.f, 0.f));
 114     }
 115   }
 116 #endif
 117   config->Set<webrtc::Beamforming>(
 118       new webrtc::Beamforming(geometry.size() > 1, geometry));
 119 }
 120
 121 }  // namespace
 122
 123 // Wraps AudioBus to provide access to the array of channel pointers, since this
 124 // is the type webrtc::AudioProcessing deals in. The array is refreshed on every
 125 // channel_ptrs() call, and will be valid until the underlying AudioBus pointers
 126 // are changed, e.g. through calls to SetChannelData() or SwapChannels().
 127 //
 128 // All methods are called on one of the capture or render audio threads
 129 // exclusively.
 130 class MediaStreamAudioBus {
 131  public:
 132   MediaStreamAudioBus(int channels, int frames)
 133       : bus_(media::AudioBus::Create(channels, frames)),
 134         channel_ptrs_(new float*[channels]) {
 135     // May be created in the main render thread and used in the audio threads.
 136     thread_checker_.DetachFromThread();
 137   }
 138
 139   media::AudioBus* bus() {
 140     DCHECK(thread_checker_.CalledOnValidThread());
 141     return bus_.get();
 142   }
 143
 144   float* const* channel_ptrs() {
 145     DCHECK(thread_checker_.CalledOnValidThread());
 146     for (int i = 0; i < bus_->channels(); ++i) {
 147       channel_ptrs_[i] = bus_->channel(i);
 148     }
 149     return channel_ptrs_.get();
 150   }
 151
 152  private:
 153   base::ThreadChecker thread_checker_;
 154   scoped_ptr<media::AudioBus> bus_;
 155   scoped_ptr<float*[]> channel_ptrs_;
 156 };
 157
 158 // Wraps AudioFifo to provide a cleaner interface to MediaStreamAudioProcessor.
 159 // It avoids the FIFO when the source and destination frames match. All methods
 160 // are called on one of the capture or render audio threads exclusively. If
 161 // |source_channels| is larger than |destination_channels|, only the first
 162 // |destination_channels| are kept from the source.
 163 class MediaStreamAudioFifo {
 164  public:
 165   MediaStreamAudioFifo(int source_channels,
 166                        int destination_channels,
 167                        int source_frames,
 168                        int destination_frames,
 169                        int sample_rate)
 170      : source_channels_(source_channels),
 171        source_frames_(source_frames),
 172        sample_rate_(sample_rate),
 173        destination_(
 174            new MediaStreamAudioBus(destination_channels, destination_frames)),
 175        data_available_(false) {
 176     DCHECK_GE(source_channels, destination_channels);
 177     DCHECK_GT(sample_rate_, 0);
 178
 179     if (source_channels > destination_channels) {
 180       audio_source_intermediate_ =
 181           media::AudioBus::CreateWrapper(destination_channels);
 182     }
 183
 184     if (source_frames != destination_frames) {
 185       // Since we require every Push to be followed by as many Consumes as
 186       // possible, twice the larger of the two is a (probably) loose upper bound
 187       // on the FIFO size.
 188       const int fifo_frames = 2 * std::max(source_frames, destination_frames);
 189       fifo_.reset(new media::AudioFifo(destination_channels, fifo_frames));
 190     }
 191
 192     // May be created in the main render thread and used in the audio threads.
 193     thread_checker_.DetachFromThread();
 194   }
 195
 196   void Push(const media::AudioBus& source, base::TimeDelta audio_delay) {
 197     DCHECK(thread_checker_.CalledOnValidThread());
 198     DCHECK_EQ(source.channels(), source_channels_);
 199     DCHECK_EQ(source.frames(), source_frames_);
 200
 201     const media::AudioBus* source_to_push = &source;
 202
 203     if (audio_source_intermediate_) {
 204       for (int i = 0; i < destination_->bus()->channels(); ++i) {
 205         audio_source_intermediate_->SetChannelData(
 206             i,
 207             const_cast<float*>(source.channel(i)));
 208       }
 209       audio_source_intermediate_->set_frames(source.frames());
 210       source_to_push = audio_source_intermediate_.get();
 211     }
 212
 213     if (fifo_) {
 214       CHECK_LT(fifo_->frames(), destination_->bus()->frames());
 215       next_audio_delay_ = audio_delay +
 216           fifo_->frames() * base::TimeDelta::FromSeconds(1) / sample_rate_;
 217       fifo_->Push(source_to_push);
 218     } else {
 219       CHECK(!data_available_);
 220       source_to_push->CopyTo(destination_->bus());
 221       next_audio_delay_ = audio_delay;
 222       data_available_ = true;
 223     }
 224   }
 225
 226   // Returns true if there are destination_frames() of data available to be
 227   // consumed, and otherwise false.
 228   bool Consume(MediaStreamAudioBus** destination,
 229                base::TimeDelta* audio_delay) {
 230     DCHECK(thread_checker_.CalledOnValidThread());
 231
 232     if (fifo_) {
 233       if (fifo_->frames() < destination_->bus()->frames())
 234         return false;
 235
 236       fifo_->Consume(destination_->bus(), 0, destination_->bus()->frames());
 237       *audio_delay = next_audio_delay_;
 238       next_audio_delay_ -=
 239           destination_->bus()->frames() * base::TimeDelta::FromSeconds(1) /
 240               sample_rate_;
 241     } else {
 242       if (!data_available_)
 243         return false;
 244       *audio_delay = next_audio_delay_;
 245       // The data was already copied to |destination_| in this case.
 246       data_available_ = false;
 247     }
 248
 249     *destination = destination_.get();
 250     return true;
 251   }
 252
 253  private:
 254   base::ThreadChecker thread_checker_;
 255   const int source_channels_;  // For a DCHECK.
 256   const int source_frames_;  // For a DCHECK.
 257   const int sample_rate_;
 258   scoped_ptr<media::AudioBus> audio_source_intermediate_;
 259   scoped_ptr<MediaStreamAudioBus> destination_;
 260   scoped_ptr<media::AudioFifo> fifo_;
 261
 262   // When using |fifo_|, this is the audio delay of the first sample to be
 263   // consumed next from the FIFO.  When not using |fifo_|, this is the audio
 264   // delay of the first sample in |destination_|.
 265   base::TimeDelta next_audio_delay_;
 266
 267   // True when |destination_| contains the data to be returned by the next call
 268   // to Consume().  Only used when the FIFO is disabled.
 269   bool data_available_;
 270 };
 271
 272 MediaStreamAudioProcessor::MediaStreamAudioProcessor(
 273     const blink::WebMediaConstraints& constraints,
 274     int effects,
 275     WebRtcPlayoutDataSource* playout_data_source)
 276     : render_delay_ms_(0),
 277       playout_data_source_(playout_data_source),
 278       audio_mirroring_(false),
 279       typing_detected_(false),
 280       stopped_(false) {
 281   capture_thread_checker_.DetachFromThread();
 282   render_thread_checker_.DetachFromThread();
 283   InitializeAudioProcessingModule(constraints, effects);
 284
 285   aec_dump_message_filter_ = AecDumpMessageFilter::Get();
 286   // In unit tests not creating a message filter, |aec_dump_message_filter_|
 287   // will be NULL. We can just ignore that. Other unit tests and browser tests
 288   // ensure that we do get the filter when we should.
 289   if (aec_dump_message_filter_.get())
 290     aec_dump_message_filter_->AddDelegate(this);
 291 }
 292
 293 MediaStreamAudioProcessor::~MediaStreamAudioProcessor() {
 294   DCHECK(main_thread_checker_.CalledOnValidThread());
 295   Stop();
 296 }
 297
 298 void MediaStreamAudioProcessor::OnCaptureFormatChanged(
 299     const media::AudioParameters& input_format) {
 300   DCHECK(main_thread_checker_.CalledOnValidThread());
 301   // There is no need to hold a lock here since the caller guarantees that
 302   // there is no more PushCaptureData() and ProcessAndConsumeData() callbacks
 303   // on the capture thread.
 304   InitializeCaptureFifo(input_format);
 305
 306   // Reset the |capture_thread_checker_| since the capture data will come from
 307   // a new capture thread.
 308   capture_thread_checker_.DetachFromThread();
 309 }
 310
 311 void MediaStreamAudioProcessor::PushCaptureData(
 312     const media::AudioBus& audio_source,
 313     base::TimeDelta capture_delay) {
 314   DCHECK(capture_thread_checker_.CalledOnValidThread());
 315
 316   capture_fifo_->Push(audio_source, capture_delay);
 317 }
 318
 319 bool MediaStreamAudioProcessor::ProcessAndConsumeData(
 320     int volume,
 321     bool key_pressed,
 322     media::AudioBus** processed_data,
 323     base::TimeDelta* capture_delay,
 324     int* new_volume) {
 325   DCHECK(capture_thread_checker_.CalledOnValidThread());
 326   DCHECK(processed_data);
 327   DCHECK(capture_delay);
 328   DCHECK(new_volume);
 329
 330   TRACE_EVENT0("audio", "MediaStreamAudioProcessor::ProcessAndConsumeData");
 331
 332   MediaStreamAudioBus* process_bus;
 333   if (!capture_fifo_->Consume(&process_bus, capture_delay))
 334     return false;
 335
 336   // Use the process bus directly if audio processing is disabled.
 337   MediaStreamAudioBus* output_bus = process_bus;
 338   *new_volume = 0;
 339   if (audio_processing_) {
 340     output_bus = output_bus_.get();
 341     *new_volume = ProcessData(process_bus->channel_ptrs(),
 342                               process_bus->bus()->frames(), *capture_delay,
 343                               volume, key_pressed, output_bus->channel_ptrs());
 344   }
 345
 346   // Swap channels before interleaving the data.
 347   if (audio_mirroring_ &&
 348       output_format_.channel_layout() == media::CHANNEL_LAYOUT_STEREO) {
 349     // Swap the first and second channels.
 350     output_bus->bus()->SwapChannels(0, 1);
 351   }
 352
 353   *processed_data = output_bus->bus();
 354
 355   return true;
 356 }
 357
 358 void MediaStreamAudioProcessor::Stop() {
 359   DCHECK(main_thread_checker_.CalledOnValidThread());
 360   if (stopped_)
 361     return;
 362
 363   stopped_ = true;
 364
 365   if (aec_dump_message_filter_.get()) {
 366     aec_dump_message_filter_->RemoveDelegate(this);
 367     aec_dump_message_filter_ = NULL;
 368   }
 369
 370   if (!audio_processing_.get())
 371     return;
 372
 373   audio_processing_.get()->UpdateHistogramsOnCallEnd();
 374   StopEchoCancellationDump(audio_processing_.get());
 375
 376   if (playout_data_source_) {
 377     playout_data_source_->RemovePlayoutSink(this);
 378     playout_data_source_ = NULL;
 379   }
 380 }
 381
 382 const media::AudioParameters& MediaStreamAudioProcessor::InputFormat() const {
 383   return input_format_;
 384 }
 385
 386 const media::AudioParameters& MediaStreamAudioProcessor::OutputFormat() const {
 387   return output_format_;
 388 }
 389
 390 void MediaStreamAudioProcessor::OnAecDumpFile(
 391     const IPC::PlatformFileForTransit& file_handle) {
 392   DCHECK(main_thread_checker_.CalledOnValidThread());
 393
 394   base::File file = IPC::PlatformFileForTransitToFile(file_handle);
 395   DCHECK(file.IsValid());
 396
 397   if (audio_processing_)
 398     StartEchoCancellationDump(audio_processing_.get(), file.Pass());
 399   else
 400     file.Close();
 401 }
 402
 403 void MediaStreamAudioProcessor::OnDisableAecDump() {
 404   DCHECK(main_thread_checker_.CalledOnValidThread());
 405   if (audio_processing_)
 406     StopEchoCancellationDump(audio_processing_.get());
 407 }
 408
 409 void MediaStreamAudioProcessor::OnIpcClosing() {
 410   DCHECK(main_thread_checker_.CalledOnValidThread());
 411   aec_dump_message_filter_ = NULL;
 412 }
 413
 414 void MediaStreamAudioProcessor::OnPlayoutData(media::AudioBus* audio_bus,
 415                                               int sample_rate,
 416                                               int audio_delay_milliseconds) {
 417   DCHECK(render_thread_checker_.CalledOnValidThread());
 418   DCHECK(audio_processing_->echo_control_mobile()->is_enabled() ^
 419          audio_processing_->echo_cancellation()->is_enabled());
 420
 421   TRACE_EVENT0("audio", "MediaStreamAudioProcessor::OnPlayoutData");
 422   DCHECK_LT(audio_delay_milliseconds,
 423             std::numeric_limits<base::subtle::Atomic32>::max());
 424   base::subtle::Release_Store(&render_delay_ms_, audio_delay_milliseconds);
 425
 426   InitializeRenderFifoIfNeeded(sample_rate, audio_bus->channels(),
 427                                audio_bus->frames());
 428
 429   render_fifo_->Push(
 430       *audio_bus, base::TimeDelta::FromMilliseconds(audio_delay_milliseconds));
 431   MediaStreamAudioBus* analysis_bus;
 432   base::TimeDelta audio_delay;
 433   while (render_fifo_->Consume(&analysis_bus, &audio_delay)) {
 434     // TODO(ajm): Should AnalyzeReverseStream() account for the |audio_delay|?
 435     audio_processing_->AnalyzeReverseStream(
 436         analysis_bus->channel_ptrs(),
 437         analysis_bus->bus()->frames(),
 438         sample_rate,
 439         ChannelsToLayout(audio_bus->channels()));
 440   }
 441 }
 442
 443 void MediaStreamAudioProcessor::OnPlayoutDataSourceChanged() {
 444   DCHECK(main_thread_checker_.CalledOnValidThread());
 445   // There is no need to hold a lock here since the caller guarantees that
 446   // there is no more OnPlayoutData() callback on the render thread.
 447   render_thread_checker_.DetachFromThread();
 448   render_fifo_.reset();
 449 }
 450
 451 void MediaStreamAudioProcessor::GetStats(AudioProcessorStats* stats) {
 452   stats->typing_noise_detected =
 453       (base::subtle::Acquire_Load(&typing_detected_) != false);
 454   GetAecStats(audio_processing_.get()->echo_cancellation(), stats);
 455 }
 456
 457 void MediaStreamAudioProcessor::InitializeAudioProcessingModule(
 458     const blink::WebMediaConstraints& constraints, int effects) {
 459   DCHECK(main_thread_checker_.CalledOnValidThread());
 460   DCHECK(!audio_processing_);
 461
 462   MediaAudioConstraints audio_constraints(constraints, effects);
 463
 464   // Audio mirroring can be enabled even though audio processing is otherwise
 465   // disabled.
 466   audio_mirroring_ = audio_constraints.GetProperty(
 467       MediaAudioConstraints::kGoogAudioMirroring);
 468
 469 #if defined(OS_IOS)
 470   // On iOS, VPIO provides built-in AGC and AEC.
 471   const bool echo_cancellation = false;
 472   const bool goog_agc = false;
 473 #else
 474   const bool echo_cancellation =
 475       audio_constraints.GetEchoCancellationProperty();
 476   const bool goog_agc = audio_constraints.GetProperty(
 477       MediaAudioConstraints::kGoogAutoGainControl);
 478 #endif
 479
 480 #if defined(OS_IOS) || defined(OS_ANDROID)
 481   const bool goog_experimental_aec = false;
 482   const bool goog_typing_detection = false;
 483 #else
 484   const bool goog_experimental_aec = audio_constraints.GetProperty(
 485       MediaAudioConstraints::kGoogExperimentalEchoCancellation);
 486   const bool goog_typing_detection = audio_constraints.GetProperty(
 487       MediaAudioConstraints::kGoogTypingNoiseDetection);
 488 #endif
 489
 490   const bool goog_ns = audio_constraints.GetProperty(
 491       MediaAudioConstraints::kGoogNoiseSuppression);
 492   const bool goog_experimental_ns = audio_constraints.GetProperty(
 493       MediaAudioConstraints::kGoogExperimentalNoiseSuppression);
 494   const bool goog_beamforming = IsBeamformingEnabled(audio_constraints);
 495   const bool goog_high_pass_filter = audio_constraints.GetProperty(
 496       MediaAudioConstraints::kGoogHighpassFilter);
 497   // Return immediately if no goog constraint is enabled.
 498   if (!echo_cancellation && !goog_experimental_aec && !goog_ns &&
 499       !goog_high_pass_filter && !goog_typing_detection &&
 500       !goog_agc && !goog_experimental_ns && !goog_beamforming) {
 501     RecordProcessingState(AUDIO_PROCESSING_DISABLED);
 502     return;
 503   }
 504
 505   // Experimental options provided at creation.
 506   webrtc::Config config;
 507   if (goog_experimental_aec)
 508     config.Set<webrtc::ExtendedFilter>(new webrtc::ExtendedFilter(true));
 509   if (goog_experimental_ns)
 510     config.Set<webrtc::ExperimentalNs>(new webrtc::ExperimentalNs(true));
 511   if (IsDelayAgnosticAecEnabled())
 512     config.Set<webrtc::DelayAgnostic>(new webrtc::DelayAgnostic(true));
 513   if (goog_beamforming) {
 514     ConfigureBeamforming(&config,
 515                          audio_constraints.GetPropertyAsString(
 516                              MediaAudioConstraints::kGoogArrayGeometry));
 517   }
 518
 519   // Create and configure the webrtc::AudioProcessing.
 520   audio_processing_.reset(webrtc::AudioProcessing::Create(config));
 521
 522   // Enable the audio processing components.
 523   if (echo_cancellation) {
 524     EnableEchoCancellation(audio_processing_.get());
 525
 526     if (playout_data_source_)
 527       playout_data_source_->AddPlayoutSink(this);
 528
 529     // Prepare for logging echo information. If there are data remaining in
 530     // |echo_information_| we simply discard it.
 531     echo_information_.reset(new EchoInformation());
 532   }
 533
 534   if (goog_ns) {
 535     // The beamforming postfilter is effective at suppressing stationary noise,
 536     // so reduce the single-channel NS aggressiveness when enabled.
 537     const NoiseSuppression::Level ns_level =
 538         config.Get<webrtc::Beamforming>().enabled ? NoiseSuppression::kLow
 539                                                   : NoiseSuppression::kHigh;
 540
 541     EnableNoiseSuppression(audio_processing_.get(), ns_level);
 542   }
 543
 544   if (goog_high_pass_filter)
 545     EnableHighPassFilter(audio_processing_.get());
 546
 547   if (goog_typing_detection) {
 548     // TODO(xians): Remove this |typing_detector_| after the typing suppression
 549     // is enabled by default.
 550     typing_detector_.reset(new webrtc::TypingDetection());
 551     EnableTypingDetection(audio_processing_.get(), typing_detector_.get());
 552   }
 553
 554   if (goog_agc)
 555     EnableAutomaticGainControl(audio_processing_.get());
 556
 557   RecordProcessingState(AUDIO_PROCESSING_ENABLED);
 558 }
 559
 560 void MediaStreamAudioProcessor::InitializeCaptureFifo(
 561     const media::AudioParameters& input_format) {
 562   DCHECK(main_thread_checker_.CalledOnValidThread());
 563   DCHECK(input_format.IsValid());
 564   input_format_ = input_format;
 565
 566   // TODO(ajm): For now, we assume fixed parameters for the output when audio
 567   // processing is enabled, to match the previous behavior. We should either
 568   // use the input parameters (in which case, audio processing will convert
 569   // at output) or ideally, have a backchannel from the sink to know what
 570   // format it would prefer.
 571 #if defined(OS_ANDROID)
 572   int audio_processing_sample_rate = AudioProcessing::kSampleRate16kHz;
 573 #else
 574   int audio_processing_sample_rate = AudioProcessing::kSampleRate48kHz;
 575 #endif
 576   const int output_sample_rate = audio_processing_ ?
 577                                  audio_processing_sample_rate :
 578                                  input_format.sample_rate();
 579   media::ChannelLayout output_channel_layout = audio_processing_ ?
 580       media::GuessChannelLayout(kAudioProcessingNumberOfChannels) :
 581       input_format.channel_layout();
 582
 583   // The output channels from the fifo is normally the same as input.
 584   int fifo_output_channels = input_format.channels();
 585
 586   // Special case for if we have a keyboard mic channel on the input and no
 587   // audio processing is used. We will then have the fifo strip away that
 588   // channel. So we use stereo as output layout, and also change the output
 589   // channels for the fifo.
 590   if (input_format.channel_layout() ==
 591           media::CHANNEL_LAYOUT_STEREO_AND_KEYBOARD_MIC &&
 592       !audio_processing_) {
 593     output_channel_layout = media::CHANNEL_LAYOUT_STEREO;
 594     fifo_output_channels = ChannelLayoutToChannelCount(output_channel_layout);
 595   }
 596
 597   // webrtc::AudioProcessing requires a 10 ms chunk size. We use this native
 598   // size when processing is enabled. When disabled we use the same size as
 599   // the source if less than 10 ms.
 600   //
 601   // TODO(ajm): This conditional buffer size appears to be assuming knowledge of
 602   // the sink based on the source parameters. PeerConnection sinks seem to want
 603   // 10 ms chunks regardless, while WebAudio sinks want less, and we're assuming
 604   // we can identify WebAudio sinks by the input chunk size. Less fragile would
 605   // be to have the sink actually tell us how much it wants (as in the above
 606   // TODO).
 607   int processing_frames = input_format.sample_rate() / 100;
 608   int output_frames = output_sample_rate / 100;
 609   if (!audio_processing_ && input_format.frames_per_buffer() < output_frames) {
 610     processing_frames = input_format.frames_per_buffer();
 611     output_frames = processing_frames;
 612   }
 613
 614   output_format_ = media::AudioParameters(
 615       media::AudioParameters::AUDIO_PCM_LOW_LATENCY,
 616       output_channel_layout,
 617       output_sample_rate,
 618       16,
 619       output_frames);
 620
 621   capture_fifo_.reset(
 622       new MediaStreamAudioFifo(input_format.channels(),
 623                                fifo_output_channels,
 624                                input_format.frames_per_buffer(),
 625                                processing_frames,
 626                                input_format.sample_rate()));
 627
 628   if (audio_processing_) {
 629     output_bus_.reset(new MediaStreamAudioBus(output_format_.channels(),
 630                                               output_frames));
 631   }
 632 }
 633
 634 void MediaStreamAudioProcessor::InitializeRenderFifoIfNeeded(
 635     int sample_rate, int number_of_channels, int frames_per_buffer) {
 636   DCHECK(render_thread_checker_.CalledOnValidThread());
 637   if (render_fifo_.get() &&
 638       render_format_.sample_rate() == sample_rate &&
 639       render_format_.channels() == number_of_channels &&
 640       render_format_.frames_per_buffer() == frames_per_buffer) {
 641     // Do nothing if the |render_fifo_| has been setup properly.
 642     return;
 643   }
 644
 645   render_format_ = media::AudioParameters(
 646       media::AudioParameters::AUDIO_PCM_LOW_LATENCY,
 647       media::GuessChannelLayout(number_of_channels),
 648       sample_rate,
 649       16,
 650       frames_per_buffer);
 651
 652   const int analysis_frames = sample_rate / 100;  // 10 ms chunks.
 653   render_fifo_.reset(
 654       new MediaStreamAudioFifo(number_of_channels,
 655                                number_of_channels,
 656                                frames_per_buffer,
 657                                analysis_frames,
 658                                sample_rate));
 659 }
 660
 661 int MediaStreamAudioProcessor::ProcessData(const float* const* process_ptrs,
 662                                            int process_frames,
 663                                            base::TimeDelta capture_delay,
 664                                            int volume,
 665                                            bool key_pressed,
 666                                            float* const* output_ptrs) {
 667   DCHECK(audio_processing_);
 668   DCHECK(capture_thread_checker_.CalledOnValidThread());
 669
 670   TRACE_EVENT0("audio", "MediaStreamAudioProcessor::ProcessData");
 671
 672   base::subtle::Atomic32 render_delay_ms =
 673       base::subtle::Acquire_Load(&render_delay_ms_);
 674   int64 capture_delay_ms = capture_delay.InMilliseconds();
 675   DCHECK_LT(capture_delay_ms,
 676             std::numeric_limits<base::subtle::Atomic32>::max());
 677   int total_delay_ms =  capture_delay_ms + render_delay_ms;
 678   if (total_delay_ms > 300) {
 679     LOG(WARNING) << "Large audio delay, capture delay: " << capture_delay_ms
 680                  << "ms; render delay: " << render_delay_ms << "ms";
 681   }
 682
 683   webrtc::AudioProcessing* ap = audio_processing_.get();
 684   ap->set_stream_delay_ms(total_delay_ms);
 685
 686   DCHECK_LE(volume, WebRtcAudioDeviceImpl::kMaxVolumeLevel);
 687   webrtc::GainControl* agc = ap->gain_control();
 688   int err = agc->set_stream_analog_level(volume);
 689   DCHECK_EQ(err, 0) << "set_stream_analog_level() error: " << err;
 690
 691   ap->set_stream_key_pressed(key_pressed);
 692
 693   err = ap->ProcessStream(process_ptrs,
 694                           process_frames,
 695                           input_format_.sample_rate(),
 696                           MapLayout(input_format_.channel_layout()),
 697                           output_format_.sample_rate(),
 698                           MapLayout(output_format_.channel_layout()),
 699                           output_ptrs);
 700   DCHECK_EQ(err, 0) << "ProcessStream() error: " << err;
 701
 702   if (typing_detector_) {
 703     webrtc::VoiceDetection* vad = ap->voice_detection();
 704     DCHECK(vad->is_enabled());
 705     bool detected = typing_detector_->Process(key_pressed,
 706                                               vad->stream_has_voice());
 707     base::subtle::Release_Store(&typing_detected_, detected);
 708   }
 709
 710   if (echo_information_) {
 711     echo_information_.get()->UpdateAecDelayStats(ap->echo_cancellation());
 712   }
 713
 714   // Return 0 if the volume hasn't been changed, and otherwise the new volume.
 715   return (agc->stream_analog_level() == volume) ?
 716       0 : agc->stream_analog_level();
 717 }
 718
 719 }  // namespace content