• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *  Copyright (c) 2012 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/pitch_based_vad.h"
12 
13 #include <assert.h>
14 #include <math.h>
15 #include <string.h>
16 
17 #include "webrtc/modules/audio_processing/vad/vad_circular_buffer.h"
18 #include "webrtc/modules/audio_processing/vad/common.h"
19 #include "webrtc/modules/audio_processing/vad/noise_gmm_tables.h"
20 #include "webrtc/modules/audio_processing/vad/voice_gmm_tables.h"
21 #include "webrtc/modules/include/module_common_types.h"
22 
23 namespace webrtc {
24 
25 static_assert(kNoiseGmmDim == kVoiceGmmDim,
26               "noise and voice gmm dimension not equal");
27 
28 // These values should match MATLAB counterparts for unit-tests to pass.
29 static const int kPosteriorHistorySize = 500;  // 5 sec of 10 ms frames.
30 static const double kInitialPriorProbability = 0.3;
31 static const int kTransientWidthThreshold = 7;
32 static const double kLowProbabilityThreshold = 0.2;
33 
LimitProbability(double p)34 static double LimitProbability(double p) {
35   const double kLimHigh = 0.99;
36   const double kLimLow = 0.01;
37 
38   if (p > kLimHigh)
39     p = kLimHigh;
40   else if (p < kLimLow)
41     p = kLimLow;
42   return p;
43 }
44 
PitchBasedVad()45 PitchBasedVad::PitchBasedVad()
46     : p_prior_(kInitialPriorProbability),
47       circular_buffer_(VadCircularBuffer::Create(kPosteriorHistorySize)) {
48   // Setup noise GMM.
49   noise_gmm_.dimension = kNoiseGmmDim;
50   noise_gmm_.num_mixtures = kNoiseGmmNumMixtures;
51   noise_gmm_.weight = kNoiseGmmWeights;
52   noise_gmm_.mean = &kNoiseGmmMean[0][0];
53   noise_gmm_.covar_inverse = &kNoiseGmmCovarInverse[0][0][0];
54 
55   // Setup voice GMM.
56   voice_gmm_.dimension = kVoiceGmmDim;
57   voice_gmm_.num_mixtures = kVoiceGmmNumMixtures;
58   voice_gmm_.weight = kVoiceGmmWeights;
59   voice_gmm_.mean = &kVoiceGmmMean[0][0];
60   voice_gmm_.covar_inverse = &kVoiceGmmCovarInverse[0][0][0];
61 }
62 
~PitchBasedVad()63 PitchBasedVad::~PitchBasedVad() {
64 }
65 
VoicingProbability(const AudioFeatures & features,double * p_combined)66 int PitchBasedVad::VoicingProbability(const AudioFeatures& features,
67                                       double* p_combined) {
68   double p;
69   double gmm_features[3];
70   double pdf_features_given_voice;
71   double pdf_features_given_noise;
72   // These limits are the same in matlab implementation 'VoicingProbGMM().'
73   const double kLimLowLogPitchGain = -2.0;
74   const double kLimHighLogPitchGain = -0.9;
75   const double kLimLowSpectralPeak = 200;
76   const double kLimHighSpectralPeak = 2000;
77   const double kEps = 1e-12;
78   for (size_t n = 0; n < features.num_frames; n++) {
79     gmm_features[0] = features.log_pitch_gain[n];
80     gmm_features[1] = features.spectral_peak[n];
81     gmm_features[2] = features.pitch_lag_hz[n];
82 
83     pdf_features_given_voice = EvaluateGmm(gmm_features, voice_gmm_);
84     pdf_features_given_noise = EvaluateGmm(gmm_features, noise_gmm_);
85 
86     if (features.spectral_peak[n] < kLimLowSpectralPeak ||
87         features.spectral_peak[n] > kLimHighSpectralPeak ||
88         features.log_pitch_gain[n] < kLimLowLogPitchGain) {
89       pdf_features_given_voice = kEps * pdf_features_given_noise;
90     } else if (features.log_pitch_gain[n] > kLimHighLogPitchGain) {
91       pdf_features_given_noise = kEps * pdf_features_given_voice;
92     }
93 
94     p = p_prior_ * pdf_features_given_voice /
95         (pdf_features_given_voice * p_prior_ +
96          pdf_features_given_noise * (1 - p_prior_));
97 
98     p = LimitProbability(p);
99 
100     // Combine pitch-based probability with standalone probability, before
101     // updating prior probabilities.
102     double prod_active = p * p_combined[n];
103     double prod_inactive = (1 - p) * (1 - p_combined[n]);
104     p_combined[n] = prod_active / (prod_active + prod_inactive);
105 
106     if (UpdatePrior(p_combined[n]) < 0)
107       return -1;
108     // Limit prior probability. With a zero prior probability the posterior
109     // probability is always zero.
110     p_prior_ = LimitProbability(p_prior_);
111   }
112   return 0;
113 }
114 
UpdatePrior(double p)115 int PitchBasedVad::UpdatePrior(double p) {
116   circular_buffer_->Insert(p);
117   if (circular_buffer_->RemoveTransient(kTransientWidthThreshold,
118                                         kLowProbabilityThreshold) < 0)
119     return -1;
120   p_prior_ = circular_buffer_->Mean();
121   return 0;
122 }
123 
124 }  // namespace webrtc
125