OLD | NEW |
1 /* | 1 /* |
2 * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved. | 2 * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license | 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 | 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 | 6 * tree. An additional intellectual property rights grant can be found |
7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
9 */ | 9 */ |
10 | 10 |
11 #include "webrtc/modules/audio_processing/vad/vad_audio_proc.h" | 11 #include "webrtc/modules/audio_processing/agc/agc_audio_proc.h" |
12 | 12 |
13 #include <math.h> | 13 #include <math.h> |
14 #include <stdio.h> | 14 #include <stdio.h> |
15 | 15 |
16 #include "webrtc/common_audio/fft4g.h" | 16 #include "webrtc/common_audio/fft4g.h" |
17 #include "webrtc/modules/audio_processing/vad/vad_audio_proc_internal.h" | 17 #include "webrtc/modules/audio_processing/agc/agc_audio_proc_internal.h" |
18 #include "webrtc/modules/audio_processing/vad/pitch_internal.h" | 18 #include "webrtc/modules/audio_processing/agc/pitch_internal.h" |
19 #include "webrtc/modules/audio_processing/vad/pole_zero_filter.h" | 19 #include "webrtc/modules/audio_processing/agc/pole_zero_filter.h" |
20 extern "C" { | 20 extern "C" { |
21 #include "webrtc/modules/audio_coding/codecs/isac/main/source/codec.h" | 21 #include "webrtc/modules/audio_coding/codecs/isac/main/source/codec.h" |
22 #include "webrtc/modules/audio_coding/codecs/isac/main/source/lpc_analysis.h" | 22 #include "webrtc/modules/audio_coding/codecs/isac/main/source/lpc_analysis.h" |
23 #include "webrtc/modules/audio_coding/codecs/isac/main/source/pitch_estimator.h" | 23 #include "webrtc/modules/audio_coding/codecs/isac/main/source/pitch_estimator.h" |
24 #include "webrtc/modules/audio_coding/codecs/isac/main/source/structs.h" | 24 #include "webrtc/modules/audio_coding/codecs/isac/main/source/structs.h" |
25 } | 25 } |
26 #include "webrtc/modules/interface/module_common_types.h" | 26 #include "webrtc/modules/interface/module_common_types.h" |
27 | 27 |
28 namespace webrtc { | 28 namespace webrtc { |
29 | 29 |
30 // The following structures are declared anonymous in iSAC's structs.h. To | 30 // The following structures are declared anonymous in iSAC's structs.h. To |
31 // forward declare them, we use this derived class trick. | 31 // forward declare them, we use this derived class trick. |
32 struct VadAudioProc::PitchAnalysisStruct : public ::PitchAnalysisStruct {}; | 32 struct AgcAudioProc::PitchAnalysisStruct : public ::PitchAnalysisStruct {}; |
33 struct VadAudioProc::PreFiltBankstr : public ::PreFiltBankstr {}; | 33 struct AgcAudioProc::PreFiltBankstr : public ::PreFiltBankstr {}; |
34 | 34 |
35 static const float kFrequencyResolution = | 35 static const float kFrequencyResolution = kSampleRateHz / |
36 kSampleRateHz / static_cast<float>(VadAudioProc::kDftSize); | 36 static_cast<float>(AgcAudioProc::kDftSize); |
37 static const int kSilenceRms = 5; | 37 static const int kSilenceRms = 5; |
38 | 38 |
39 // TODO(turajs): Make a Create or Init for VadAudioProc. | 39 // TODO(turajs): Make a Create or Init for AgcAudioProc. |
40 VadAudioProc::VadAudioProc() | 40 AgcAudioProc::AgcAudioProc() |
41 : audio_buffer_(), | 41 : audio_buffer_(), |
42 num_buffer_samples_(kNumPastSignalSamples), | 42 num_buffer_samples_(kNumPastSignalSamples), |
43 log_old_gain_(-2), | 43 log_old_gain_(-2), |
44 old_lag_(50), // Arbitrary but valid as pitch-lag (in samples). | 44 old_lag_(50), // Arbitrary but valid as pitch-lag (in samples). |
45 pitch_analysis_handle_(new PitchAnalysisStruct), | 45 pitch_analysis_handle_(new PitchAnalysisStruct), |
46 pre_filter_handle_(new PreFiltBankstr), | 46 pre_filter_handle_(new PreFiltBankstr), |
47 high_pass_filter_(PoleZeroFilter::Create(kCoeffNumerator, | 47 high_pass_filter_(PoleZeroFilter::Create( |
48 kFilterOrder, | 48 kCoeffNumerator, kFilterOrder, kCoeffDenominator, kFilterOrder)) { |
49 kCoeffDenominator, | |
50 kFilterOrder)) { | |
51 static_assert(kNumPastSignalSamples + kNumSubframeSamples == | 49 static_assert(kNumPastSignalSamples + kNumSubframeSamples == |
52 sizeof(kLpcAnalWin) / sizeof(kLpcAnalWin[0]), | 50 sizeof(kLpcAnalWin) / sizeof(kLpcAnalWin[0]), |
53 "lpc analysis window incorrect size"); | 51 "lpc analysis window incorrect size"); |
54 static_assert(kLpcOrder + 1 == sizeof(kCorrWeight) / sizeof(kCorrWeight[0]), | 52 static_assert(kLpcOrder + 1 == sizeof(kCorrWeight) / sizeof(kCorrWeight[0]), |
55 "correlation weight incorrect size"); | 53 "correlation weight incorrect size"); |
56 | 54 |
57 // TODO(turajs): Are we doing too much in the constructor? | 55 // TODO(turajs): Are we doing too much in the constructor? |
58 float data[kDftSize]; | 56 float data[kDftSize]; |
59 // Make FFT to initialize. | 57 // Make FFT to initialize. |
60 ip_[0] = 0; | 58 ip_[0] = 0; |
61 WebRtc_rdft(kDftSize, 1, data, ip_, w_fft_); | 59 WebRtc_rdft(kDftSize, 1, data, ip_, w_fft_); |
62 // TODO(turajs): Need to initialize high-pass filter. | 60 // TODO(turajs): Need to initialize high-pass filter. |
63 | 61 |
64 // Initialize iSAC components. | 62 // Initialize iSAC components. |
65 WebRtcIsac_InitPreFilterbank(pre_filter_handle_.get()); | 63 WebRtcIsac_InitPreFilterbank(pre_filter_handle_.get()); |
66 WebRtcIsac_InitPitchAnalysis(pitch_analysis_handle_.get()); | 64 WebRtcIsac_InitPitchAnalysis(pitch_analysis_handle_.get()); |
67 } | 65 } |
68 | 66 |
69 VadAudioProc::~VadAudioProc() { | 67 AgcAudioProc::~AgcAudioProc() {} |
70 } | |
71 | 68 |
72 void VadAudioProc::ResetBuffer() { | 69 void AgcAudioProc::ResetBuffer() { |
73 memcpy(audio_buffer_, &audio_buffer_[kNumSamplesToProcess], | 70 memcpy(audio_buffer_, &audio_buffer_[kNumSamplesToProcess], |
74 sizeof(audio_buffer_[0]) * kNumPastSignalSamples); | 71 sizeof(audio_buffer_[0]) * kNumPastSignalSamples); |
75 num_buffer_samples_ = kNumPastSignalSamples; | 72 num_buffer_samples_ = kNumPastSignalSamples; |
76 } | 73 } |
77 | 74 |
78 int VadAudioProc::ExtractFeatures(const int16_t* frame, | 75 int AgcAudioProc::ExtractFeatures(const int16_t* frame, |
79 int length, | 76 int length, |
80 AudioFeatures* features) { | 77 AudioFeatures* features) { |
81 features->num_frames = 0; | 78 features->num_frames = 0; |
82 if (length != kNumSubframeSamples) { | 79 if (length != kNumSubframeSamples) { |
83 return -1; | 80 return -1; |
84 } | 81 } |
85 | 82 |
86 // High-pass filter to remove the DC component and very low frequency content. | 83 // High-pass filter to remove the DC component and very low frequency content. |
87 // We have experienced that this high-pass filtering improves voice/non-voiced | 84 // We have experienced that this high-pass filtering improves voice/non-voiced |
88 // classification. | 85 // classification. |
89 if (high_pass_filter_->Filter(frame, kNumSubframeSamples, | 86 if (high_pass_filter_->Filter(frame, kNumSubframeSamples, |
90 &audio_buffer_[num_buffer_samples_]) != 0) { | 87 &audio_buffer_[num_buffer_samples_]) != 0) { |
91 return -1; | 88 return -1; |
92 } | 89 } |
93 | 90 |
94 num_buffer_samples_ += kNumSubframeSamples; | 91 num_buffer_samples_ += kNumSubframeSamples; |
95 if (num_buffer_samples_ < kBufferLength) { | 92 if (num_buffer_samples_ < kBufferLength) { |
96 return 0; | 93 return 0; |
97 } | 94 } |
98 assert(num_buffer_samples_ == kBufferLength); | 95 assert(num_buffer_samples_ == kBufferLength); |
99 features->num_frames = kNum10msSubframes; | 96 features->num_frames = kNum10msSubframes; |
100 features->silence = false; | 97 features->silence = false; |
101 | 98 |
102 Rms(features->rms, kMaxNumFrames); | 99 Rms(features->rms, kMaxNumFrames); |
103 for (int i = 0; i < kNum10msSubframes; ++i) { | 100 for (int i = 0; i < kNum10msSubframes; ++i) { |
104 if (features->rms[i] < kSilenceRms) { | 101 if (features->rms[i] < kSilenceRms) { |
105 // PitchAnalysis can cause NaNs in the pitch gain if it's fed silence. | 102 // PitchAnalysis can cause NaNs in the pitch gain if it's fed silence. |
106 // Bail out here instead. | 103 // Bail out here instead. |
107 features->silence = true; | 104 features->silence = true; |
108 ResetBuffer(); | 105 ResetBuffer(); |
109 return 0; | 106 return 0; |
110 } | 107 } |
111 } | 108 } |
112 | 109 |
113 PitchAnalysis(features->log_pitch_gain, features->pitch_lag_hz, | 110 PitchAnalysis(features->log_pitch_gain, features->pitch_lag_hz, |
114 kMaxNumFrames); | 111 kMaxNumFrames); |
115 FindFirstSpectralPeaks(features->spectral_peak, kMaxNumFrames); | 112 FindFirstSpectralPeaks(features->spectral_peak, kMaxNumFrames); |
116 ResetBuffer(); | 113 ResetBuffer(); |
117 return 0; | 114 return 0; |
118 } | 115 } |
119 | 116 |
120 // Computes |kLpcOrder + 1| correlation coefficients. | 117 // Computes |kLpcOrder + 1| correlation coefficients. |
121 void VadAudioProc::SubframeCorrelation(double* corr, | 118 void AgcAudioProc::SubframeCorrelation(double* corr, int length_corr, |
122 int length_corr, | |
123 int subframe_index) { | 119 int subframe_index) { |
124 assert(length_corr >= kLpcOrder + 1); | 120 assert(length_corr >= kLpcOrder + 1); |
125 double windowed_audio[kNumSubframeSamples + kNumPastSignalSamples]; | 121 double windowed_audio[kNumSubframeSamples + kNumPastSignalSamples]; |
126 int buffer_index = subframe_index * kNumSubframeSamples; | 122 int buffer_index = subframe_index * kNumSubframeSamples; |
127 | 123 |
128 for (int n = 0; n < kNumSubframeSamples + kNumPastSignalSamples; n++) | 124 for (int n = 0; n < kNumSubframeSamples + kNumPastSignalSamples; n++) |
129 windowed_audio[n] = audio_buffer_[buffer_index++] * kLpcAnalWin[n]; | 125 windowed_audio[n] = audio_buffer_[buffer_index++] * kLpcAnalWin[n]; |
130 | 126 |
131 WebRtcIsac_AutoCorr(corr, windowed_audio, | 127 WebRtcIsac_AutoCorr(corr, windowed_audio, kNumSubframeSamples + |
132 kNumSubframeSamples + kNumPastSignalSamples, kLpcOrder); | 128 kNumPastSignalSamples, kLpcOrder); |
133 } | 129 } |
134 | 130 |
135 // Compute |kNum10msSubframes| sets of LPC coefficients, one per 10 ms input. | 131 // Compute |kNum10msSubframes| sets of LPC coefficients, one per 10 ms input. |
136 // The analysis window is 15 ms long and it is centered on the first half of | 132 // The analysis window is 15 ms long and it is centered on the first half of |
137 // each 10ms sub-frame. This is equivalent to computing LPC coefficients for the | 133 // each 10ms sub-frame. This is equivalent to computing LPC coefficients for the |
138 // first half of each 10 ms subframe. | 134 // first half of each 10 ms subframe. |
139 void VadAudioProc::GetLpcPolynomials(double* lpc, int length_lpc) { | 135 void AgcAudioProc::GetLpcPolynomials(double* lpc, int length_lpc) { |
140 assert(length_lpc >= kNum10msSubframes * (kLpcOrder + 1)); | 136 assert(length_lpc >= kNum10msSubframes * (kLpcOrder + 1)); |
141 double corr[kLpcOrder + 1]; | 137 double corr[kLpcOrder + 1]; |
142 double reflec_coeff[kLpcOrder]; | 138 double reflec_coeff[kLpcOrder]; |
143 for (int i = 0, offset_lpc = 0; i < kNum10msSubframes; | 139 for (int i = 0, offset_lpc = 0; i < kNum10msSubframes; |
144 i++, offset_lpc += kLpcOrder + 1) { | 140 i++, offset_lpc += kLpcOrder + 1) { |
145 SubframeCorrelation(corr, kLpcOrder + 1, i); | 141 SubframeCorrelation(corr, kLpcOrder + 1, i); |
146 corr[0] *= 1.0001; | 142 corr[0] *= 1.0001; |
147 // This makes Lev-Durb a bit more stable. | 143 // This makes Lev-Durb a bit more stable. |
148 for (int k = 0; k < kLpcOrder + 1; k++) { | 144 for (int k = 0; k < kLpcOrder + 1; k++) { |
149 corr[k] *= kCorrWeight[k]; | 145 corr[k] *= kCorrWeight[k]; |
150 } | 146 } |
151 WebRtcIsac_LevDurb(&lpc[offset_lpc], reflec_coeff, corr, kLpcOrder); | 147 WebRtcIsac_LevDurb(&lpc[offset_lpc], reflec_coeff, corr, kLpcOrder); |
152 } | 148 } |
153 } | 149 } |
154 | 150 |
155 // Fit a second order curve to these 3 points and find the location of the | 151 // Fit a second order curve to these 3 points and find the location of the |
156 // extremum. The points are inverted before curve fitting. | 152 // extremum. The points are inverted before curve fitting. |
157 static float QuadraticInterpolation(float prev_val, | 153 static float QuadraticInterpolation(float prev_val, float curr_val, |
158 float curr_val, | |
159 float next_val) { | 154 float next_val) { |
160 // Doing the interpolation in |1 / A(z)|^2. | 155 // Doing the interpolation in |1 / A(z)|^2. |
161 float fractional_index = 0; | 156 float fractional_index = 0; |
162 next_val = 1.0f / next_val; | 157 next_val = 1.0f / next_val; |
163 prev_val = 1.0f / prev_val; | 158 prev_val = 1.0f / prev_val; |
164 curr_val = 1.0f / curr_val; | 159 curr_val = 1.0f / curr_val; |
165 | 160 |
166 fractional_index = | 161 fractional_index = -(next_val - prev_val) * 0.5f / (next_val + prev_val - |
167 -(next_val - prev_val) * 0.5f / (next_val + prev_val - 2.f * curr_val); | 162 2.f * curr_val); |
168 assert(fabs(fractional_index) < 1); | 163 assert(fabs(fractional_index) < 1); |
169 return fractional_index; | 164 return fractional_index; |
170 } | 165 } |
171 | 166 |
172 // 1 / A(z), where A(z) is defined by |lpc| is a model of the spectral envelope | 167 // 1 / A(z), where A(z) is defined by |lpc| is a model of the spectral envelope |
173 // of the input signal. The local maximum of the spectral envelope corresponds | 168 // of the input signal. The local maximum of the spectral envelope corresponds |
174 // with the local minimum of A(z). It saves complexity, as we save one | 169 // with the local minimum of A(z). It saves complexity, as we save one |
175 // inversion. Furthermore, we find the first local maximum of magnitude squared, | 170 // inversion. Furthermore, we find the first local maximum of magnitude squared, |
176 // to save on one square root. | 171 // to save on one square root. |
177 void VadAudioProc::FindFirstSpectralPeaks(double* f_peak, int length_f_peak) { | 172 void AgcAudioProc::FindFirstSpectralPeaks(double* f_peak, int length_f_peak) { |
178 assert(length_f_peak >= kNum10msSubframes); | 173 assert(length_f_peak >= kNum10msSubframes); |
179 double lpc[kNum10msSubframes * (kLpcOrder + 1)]; | 174 double lpc[kNum10msSubframes * (kLpcOrder + 1)]; |
180 // For all sub-frames. | 175 // For all sub-frames. |
181 GetLpcPolynomials(lpc, kNum10msSubframes * (kLpcOrder + 1)); | 176 GetLpcPolynomials(lpc, kNum10msSubframes * (kLpcOrder + 1)); |
182 | 177 |
183 const int kNumDftCoefficients = kDftSize / 2 + 1; | 178 const int kNumDftCoefficients = kDftSize / 2 + 1; |
184 float data[kDftSize]; | 179 float data[kDftSize]; |
185 | 180 |
186 for (int i = 0; i < kNum10msSubframes; i++) { | 181 for (int i = 0; i < kNum10msSubframes; i++) { |
187 // Convert to float with zero pad. | 182 // Convert to float with zero pad. |
188 memset(data, 0, sizeof(data)); | 183 memset(data, 0, sizeof(data)); |
189 for (int n = 0; n < kLpcOrder + 1; n++) { | 184 for (int n = 0; n < kLpcOrder + 1; n++) { |
190 data[n] = static_cast<float>(lpc[i * (kLpcOrder + 1) + n]); | 185 data[n] = static_cast<float>(lpc[i * (kLpcOrder + 1) + n]); |
191 } | 186 } |
192 // Transform to frequency domain. | 187 // Transform to frequency domain. |
193 WebRtc_rdft(kDftSize, 1, data, ip_, w_fft_); | 188 WebRtc_rdft(kDftSize, 1, data, ip_, w_fft_); |
194 | 189 |
195 int index_peak = 0; | 190 int index_peak = 0; |
196 float prev_magn_sqr = data[0] * data[0]; | 191 float prev_magn_sqr = data[0] * data[0]; |
197 float curr_magn_sqr = data[2] * data[2] + data[3] * data[3]; | 192 float curr_magn_sqr = data[2] * data[2] + data[3] * data[3]; |
198 float next_magn_sqr; | 193 float next_magn_sqr; |
199 bool found_peak = false; | 194 bool found_peak = false; |
200 for (int n = 2; n < kNumDftCoefficients - 1; n++) { | 195 for (int n = 2; n < kNumDftCoefficients - 1; n++) { |
201 next_magn_sqr = | 196 next_magn_sqr = data[2 * n] * data[2 * n] + |
202 data[2 * n] * data[2 * n] + data[2 * n + 1] * data[2 * n + 1]; | 197 data[2 * n + 1] * data[2 * n + 1]; |
203 if (curr_magn_sqr < prev_magn_sqr && curr_magn_sqr < next_magn_sqr) { | 198 if (curr_magn_sqr < prev_magn_sqr && curr_magn_sqr < next_magn_sqr) { |
204 found_peak = true; | 199 found_peak = true; |
205 index_peak = n - 1; | 200 index_peak = n - 1; |
206 break; | 201 break; |
207 } | 202 } |
208 prev_magn_sqr = curr_magn_sqr; | 203 prev_magn_sqr = curr_magn_sqr; |
209 curr_magn_sqr = next_magn_sqr; | 204 curr_magn_sqr = next_magn_sqr; |
210 } | 205 } |
211 float fractional_index = 0; | 206 float fractional_index = 0; |
212 if (!found_peak) { | 207 if (!found_peak) { |
213 // Checking if |kNumDftCoefficients - 1| is the local minimum. | 208 // Checking if |kNumDftCoefficients - 1| is the local minimum. |
214 next_magn_sqr = data[1] * data[1]; | 209 next_magn_sqr = data[1] * data[1]; |
215 if (curr_magn_sqr < prev_magn_sqr && curr_magn_sqr < next_magn_sqr) { | 210 if (curr_magn_sqr < prev_magn_sqr && curr_magn_sqr < next_magn_sqr) { |
216 index_peak = kNumDftCoefficients - 1; | 211 index_peak = kNumDftCoefficients - 1; |
217 } | 212 } |
218 } else { | 213 } else { |
219 // A peak is found, do a simple quadratic interpolation to get a more | 214 // A peak is found, do a simple quadratic interpolation to get a more |
220 // accurate estimate of the peak location. | 215 // accurate estimate of the peak location. |
221 fractional_index = | 216 fractional_index = QuadraticInterpolation(prev_magn_sqr, curr_magn_sqr, |
222 QuadraticInterpolation(prev_magn_sqr, curr_magn_sqr, next_magn_sqr); | 217 next_magn_sqr); |
223 } | 218 } |
224 f_peak[i] = (index_peak + fractional_index) * kFrequencyResolution; | 219 f_peak[i] = (index_peak + fractional_index) * kFrequencyResolution; |
225 } | 220 } |
226 } | 221 } |
227 | 222 |
228 // Using iSAC functions to estimate pitch gains & lags. | 223 // Using iSAC functions to estimate pitch gains & lags. |
229 void VadAudioProc::PitchAnalysis(double* log_pitch_gains, | 224 void AgcAudioProc::PitchAnalysis(double* log_pitch_gains, double* pitch_lags_hz, |
230 double* pitch_lags_hz, | |
231 int length) { | 225 int length) { |
232 // TODO(turajs): This can be "imported" from iSAC & and the next two | 226 // TODO(turajs): This can be "imported" from iSAC & and the next two |
233 // constants. | 227 // constants. |
234 assert(length >= kNum10msSubframes); | 228 assert(length >= kNum10msSubframes); |
235 const int kNumPitchSubframes = 4; | 229 const int kNumPitchSubframes = 4; |
236 double gains[kNumPitchSubframes]; | 230 double gains[kNumPitchSubframes]; |
237 double lags[kNumPitchSubframes]; | 231 double lags[kNumPitchSubframes]; |
238 | 232 |
239 const int kNumSubbandFrameSamples = 240; | 233 const int kNumSubbandFrameSamples = 240; |
240 const int kNumLookaheadSamples = 24; | 234 const int kNumLookaheadSamples = 24; |
241 | 235 |
242 float lower[kNumSubbandFrameSamples]; | 236 float lower[kNumSubbandFrameSamples]; |
243 float upper[kNumSubbandFrameSamples]; | 237 float upper[kNumSubbandFrameSamples]; |
244 double lower_lookahead[kNumSubbandFrameSamples]; | 238 double lower_lookahead[kNumSubbandFrameSamples]; |
245 double upper_lookahead[kNumSubbandFrameSamples]; | 239 double upper_lookahead[kNumSubbandFrameSamples]; |
246 double lower_lookahead_pre_filter[kNumSubbandFrameSamples + | 240 double lower_lookahead_pre_filter[kNumSubbandFrameSamples + |
247 kNumLookaheadSamples]; | 241 kNumLookaheadSamples]; |
248 | 242 |
249 // Split signal to lower and upper bands | 243 // Split signal to lower and upper bands |
250 WebRtcIsac_SplitAndFilterFloat(&audio_buffer_[kNumPastSignalSamples], lower, | 244 WebRtcIsac_SplitAndFilterFloat(&audio_buffer_[kNumPastSignalSamples], |
251 upper, lower_lookahead, upper_lookahead, | 245 lower, upper, lower_lookahead, upper_lookahead, |
252 pre_filter_handle_.get()); | 246 pre_filter_handle_.get()); |
253 WebRtcIsac_PitchAnalysis(lower_lookahead, lower_lookahead_pre_filter, | 247 WebRtcIsac_PitchAnalysis(lower_lookahead, lower_lookahead_pre_filter, |
254 pitch_analysis_handle_.get(), lags, gains); | 248 pitch_analysis_handle_.get(), lags, gains); |
255 | 249 |
256 // Lags are computed on lower-band signal with sampling rate half of the | 250 // Lags are computed on lower-band signal with sampling rate half of the |
257 // input signal. | 251 // input signal. |
258 GetSubframesPitchParameters( | 252 GetSubframesPitchParameters(kSampleRateHz / 2, gains, lags, |
259 kSampleRateHz / 2, gains, lags, kNumPitchSubframes, kNum10msSubframes, | 253 kNumPitchSubframes, kNum10msSubframes, |
260 &log_old_gain_, &old_lag_, log_pitch_gains, pitch_lags_hz); | 254 &log_old_gain_, &old_lag_, |
| 255 log_pitch_gains, pitch_lags_hz); |
261 } | 256 } |
262 | 257 |
263 void VadAudioProc::Rms(double* rms, int length_rms) { | 258 void AgcAudioProc::Rms(double* rms, int length_rms) { |
264 assert(length_rms >= kNum10msSubframes); | 259 assert(length_rms >= kNum10msSubframes); |
265 int offset = kNumPastSignalSamples; | 260 int offset = kNumPastSignalSamples; |
266 for (int i = 0; i < kNum10msSubframes; i++) { | 261 for (int i = 0; i < kNum10msSubframes; i++) { |
267 rms[i] = 0; | 262 rms[i] = 0; |
268 for (int n = 0; n < kNumSubframeSamples; n++, offset++) | 263 for (int n = 0; n < kNumSubframeSamples; n++, offset++) |
269 rms[i] += audio_buffer_[offset] * audio_buffer_[offset]; | 264 rms[i] += audio_buffer_[offset] * audio_buffer_[offset]; |
270 rms[i] = sqrt(rms[i] / kNumSubframeSamples); | 265 rms[i] = sqrt(rms[i] / kNumSubframeSamples); |
271 } | 266 } |
272 } | 267 } |
273 | 268 |
274 } // namespace webrtc | 269 } // namespace webrtc |
OLD | NEW |