Line data    Source code

       1             : /*
       2             :  *  Copyright (c) 2015 The WebRTC project authors. All Rights Reserved.
       3             :  *
       4             :  *  Use of this source code is governed by a BSD-style license
       5             :  *  that can be found in the LICENSE file in the root of the source
       6             :  *  tree. An additional intellectual property rights grant can be found
       7             :  *  in the file PATENTS.  All contributing project authors may
       8             :  *  be found in the AUTHORS file in the root of the source tree.
       9             :  */
      10             : 
      11             : #include "webrtc/modules/audio_processing/vad/voice_activity_detector.h"
      12             : 
      13             : #include <algorithm>
      14             : 
      15             : #include "webrtc/base/checks.h"
      16             : 
      17             : namespace webrtc {
      18             : namespace {
      19             : 
      20             : const size_t kMaxLength = 320;
      21             : const size_t kNumChannels = 1;
      22             : 
      23             : const double kDefaultVoiceValue = 1.0;
      24             : const double kNeutralProbability = 0.5;
      25             : const double kLowProbability = 0.01;
      26             : 
      27             : }  // namespace
      28             : 
      29           0 : VoiceActivityDetector::VoiceActivityDetector()
      30             :     : last_voice_probability_(kDefaultVoiceValue),
      31           0 :       standalone_vad_(StandaloneVad::Create()) {
      32           0 : }
      33             : 
      34             : VoiceActivityDetector::~VoiceActivityDetector() = default;
      35             : 
      36             : // Because ISAC has a different chunk length, it updates
      37             : // |chunkwise_voice_probabilities_| and |chunkwise_rms_| when there is new data.
      38             : // Otherwise it clears them.
      39           0 : void VoiceActivityDetector::ProcessChunk(const int16_t* audio,
      40             :                                          size_t length,
      41             :                                          int sample_rate_hz) {
      42           0 :   RTC_DCHECK_EQ(length, sample_rate_hz / 100);
      43           0 :   RTC_DCHECK_LE(length, kMaxLength);
      44             :   // Resample to the required rate.
      45           0 :   const int16_t* resampled_ptr = audio;
      46           0 :   if (sample_rate_hz != kSampleRateHz) {
      47           0 :     RTC_CHECK_EQ(
      48             :         resampler_.ResetIfNeeded(sample_rate_hz, kSampleRateHz, kNumChannels),
      49           0 :         0);
      50           0 :     resampler_.Push(audio, length, resampled_, kLength10Ms, length);
      51           0 :     resampled_ptr = resampled_;
      52             :   }
      53           0 :   RTC_DCHECK_EQ(length, kLength10Ms);
      54             : 
      55             :   // Each chunk needs to be passed into |standalone_vad_|, because internally it
      56             :   // buffers the audio and processes it all at once when GetActivity() is
      57             :   // called.
      58           0 :   RTC_CHECK_EQ(standalone_vad_->AddAudio(resampled_ptr, length), 0);
      59             : 
      60           0 :   audio_processing_.ExtractFeatures(resampled_ptr, length, &features_);
      61             : 
      62           0 :   chunkwise_voice_probabilities_.resize(features_.num_frames);
      63           0 :   chunkwise_rms_.resize(features_.num_frames);
      64           0 :   std::copy(features_.rms, features_.rms + chunkwise_rms_.size(),
      65           0 :             chunkwise_rms_.begin());
      66           0 :   if (features_.num_frames > 0) {
      67           0 :     if (features_.silence) {
      68             :       // The other features are invalid, so set the voice probabilities to an
      69             :       // arbitrary low value.
      70           0 :       std::fill(chunkwise_voice_probabilities_.begin(),
      71           0 :                 chunkwise_voice_probabilities_.end(), kLowProbability);
      72             :     } else {
      73           0 :       std::fill(chunkwise_voice_probabilities_.begin(),
      74           0 :                 chunkwise_voice_probabilities_.end(), kNeutralProbability);
      75           0 :       RTC_CHECK_GE(
      76             :           standalone_vad_->GetActivity(&chunkwise_voice_probabilities_[0],
      77             :                                        chunkwise_voice_probabilities_.size()),
      78           0 :           0);
      79           0 :       RTC_CHECK_GE(pitch_based_vad_.VoicingProbability(
      80             :                        features_, &chunkwise_voice_probabilities_[0]),
      81           0 :                    0);
      82             :     }
      83           0 :     last_voice_probability_ = chunkwise_voice_probabilities_.back();
      84             :   }
      85           0 : }
      86             : 
      87             : }  // namespace webrtc