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 |
(...skipping 43 matching lines...) Loading... |
54 | 54 |
55 // Add some dither to quiet frames. This avoids the ExtractFeatures skip a | 55 // Add some dither to quiet frames. This avoids the ExtractFeatures skip a |
56 // silence frame. Otherwise true VAD would drift with respect to the audio. | 56 // silence frame. Otherwise true VAD would drift with respect to the audio. |
57 // We only consider mono inputs. | 57 // We only consider mono inputs. |
58 static void DitherSilence(AudioFrame* frame) { | 58 static void DitherSilence(AudioFrame* frame) { |
59 ASSERT_EQ(1, frame->num_channels_); | 59 ASSERT_EQ(1, frame->num_channels_); |
60 const double kRmsSilence = 5; | 60 const double kRmsSilence = 5; |
61 const double sum_squared_silence = kRmsSilence * kRmsSilence * | 61 const double sum_squared_silence = kRmsSilence * kRmsSilence * |
62 frame->samples_per_channel_; | 62 frame->samples_per_channel_; |
63 double sum_squared = 0; | 63 double sum_squared = 0; |
64 for (int n = 0; n < frame->samples_per_channel_; n++) | 64 for (size_t n = 0; n < frame->samples_per_channel_; n++) |
65 sum_squared += frame->data_[n] * frame->data_[n]; | 65 sum_squared += frame->data_[n] * frame->data_[n]; |
66 if (sum_squared <= sum_squared_silence) { | 66 if (sum_squared <= sum_squared_silence) { |
67 for (int n = 0; n < frame->samples_per_channel_; n++) | 67 for (size_t n = 0; n < frame->samples_per_channel_; n++) |
68 frame->data_[n] = (rand() & 0xF) - 8; | 68 frame->data_[n] = (rand() & 0xF) - 8; |
69 } | 69 } |
70 } | 70 } |
71 | 71 |
72 class AgcStat { | 72 class AgcStat { |
73 public: | 73 public: |
74 AgcStat() | 74 AgcStat() |
75 : video_index_(0), | 75 : video_index_(0), |
76 activity_threshold_(kDefaultActivityThreshold), | 76 activity_threshold_(kDefaultActivityThreshold), |
77 audio_content_(Histogram::Create(kAgcAnalWindowSamples)), | 77 audio_content_(Histogram::Create(kAgcAnalWindowSamples)), |
78 audio_processing_(new VadAudioProc()), | 78 audio_processing_(new VadAudioProc()), |
79 vad_(new PitchBasedVad()), | 79 vad_(new PitchBasedVad()), |
80 standalone_vad_(StandaloneVad::Create()), | 80 standalone_vad_(StandaloneVad::Create()), |
81 audio_content_fid_(NULL) { | 81 audio_content_fid_(NULL) { |
82 for (int n = 0; n < kMaxNumFrames; n++) | 82 for (size_t n = 0; n < kMaxNumFrames; n++) |
83 video_vad_[n] = 0.5; | 83 video_vad_[n] = 0.5; |
84 } | 84 } |
85 | 85 |
86 ~AgcStat() { | 86 ~AgcStat() { |
87 if (audio_content_fid_ != NULL) { | 87 if (audio_content_fid_ != NULL) { |
88 fclose(audio_content_fid_); | 88 fclose(audio_content_fid_); |
89 } | 89 } |
90 } | 90 } |
91 | 91 |
92 void set_audio_content_file(FILE* audio_content_fid) { | 92 void set_audio_content_file(FILE* audio_content_fid) { |
(...skipping 16 matching lines...) Loading... |
109 frame.samples_per_channel_); | 109 frame.samples_per_channel_); |
110 } | 110 } |
111 if (features.num_frames > 0) { | 111 if (features.num_frames > 0) { |
112 double p[kMaxNumFrames] = {0.5, 0.5, 0.5, 0.5}; | 112 double p[kMaxNumFrames] = {0.5, 0.5, 0.5, 0.5}; |
113 if (FLAGS_standalone_vad) { | 113 if (FLAGS_standalone_vad) { |
114 standalone_vad_->GetActivity(p, kMaxNumFrames); | 114 standalone_vad_->GetActivity(p, kMaxNumFrames); |
115 } | 115 } |
116 // TODO(turajs) combining and limiting are used in the source files as | 116 // TODO(turajs) combining and limiting are used in the source files as |
117 // well they can be moved to utility. | 117 // well they can be moved to utility. |
118 // Combine Video and stand-alone VAD. | 118 // Combine Video and stand-alone VAD. |
119 for (int n = 0; n < features.num_frames; n++) { | 119 for (size_t n = 0; n < features.num_frames; n++) { |
120 double p_active = p[n] * video_vad_[n]; | 120 double p_active = p[n] * video_vad_[n]; |
121 double p_passive = (1 - p[n]) * (1 - video_vad_[n]); | 121 double p_passive = (1 - p[n]) * (1 - video_vad_[n]); |
122 p[n] = p_active / (p_active + p_passive); | 122 p[n] = p_active / (p_active + p_passive); |
123 // Limit probabilities. | 123 // Limit probabilities. |
124 p[n] = std::min(std::max(p[n], 0.01), 0.99); | 124 p[n] = std::min(std::max(p[n], 0.01), 0.99); |
125 } | 125 } |
126 if (vad_->VoicingProbability(features, p) < 0) | 126 if (vad_->VoicingProbability(features, p) < 0) |
127 return -1; | 127 return -1; |
128 for (int n = 0; n < features.num_frames; n++) { | 128 for (size_t n = 0; n < features.num_frames; n++) { |
129 audio_content_->Update(features.rms[n], p[n]); | 129 audio_content_->Update(features.rms[n], p[n]); |
130 double ac = audio_content_->AudioContent(); | 130 double ac = audio_content_->AudioContent(); |
131 if (audio_content_fid_ != NULL) { | 131 if (audio_content_fid_ != NULL) { |
132 fwrite(&ac, sizeof(ac), 1, audio_content_fid_); | 132 fwrite(&ac, sizeof(ac), 1, audio_content_fid_); |
133 } | 133 } |
134 if (ac > kAgcAnalWindowSamples * activity_threshold_) { | 134 if (ac > kAgcAnalWindowSamples * activity_threshold_) { |
135 combined_vad[n] = 1; | 135 combined_vad[n] = 1; |
136 } else { | 136 } else { |
137 combined_vad[n] = 0; | 137 combined_vad[n] = 0; |
138 } | 138 } |
139 } | 139 } |
140 video_index_ = 0; | 140 video_index_ = 0; |
141 } | 141 } |
142 return features.num_frames; | 142 return static_cast<int>(features.num_frames); |
143 } | 143 } |
144 | 144 |
145 void Reset() { | 145 void Reset() { |
146 audio_content_->Reset(); | 146 audio_content_->Reset(); |
147 } | 147 } |
148 | 148 |
149 void SetActivityThreshold(double activity_threshold) { | 149 void SetActivityThreshold(double activity_threshold) { |
150 activity_threshold_ = activity_threshold; | 150 activity_threshold_ = activity_threshold; |
151 } | 151 } |
152 | 152 |
(...skipping 86 matching lines...) Loading... |
239 double p_video = 0.5; | 239 double p_video = 0.5; |
240 int total_active = 0; | 240 int total_active = 0; |
241 int total_passive = 0; | 241 int total_passive = 0; |
242 int total_false_positive = 0; | 242 int total_false_positive = 0; |
243 int total_missed_detection = 0; | 243 int total_missed_detection = 0; |
244 int onset_adaptation = 0; | 244 int onset_adaptation = 0; |
245 int num_onsets = 0; | 245 int num_onsets = 0; |
246 bool onset = false; | 246 bool onset = false; |
247 uint8_t previous_true_vad = 0; | 247 uint8_t previous_true_vad = 0; |
248 int num_not_adapted = 0; | 248 int num_not_adapted = 0; |
249 int true_vad_index = 0; | 249 size_t true_vad_index = 0; |
250 bool in_false_positive_region = false; | 250 bool in_false_positive_region = false; |
251 int total_false_positive_duration = 0; | 251 int total_false_positive_duration = 0; |
252 bool video_adapted = false; | 252 bool video_adapted = false; |
253 while (kSamplesToRead == fread(frame.data_, sizeof(int16_t), | 253 while (kSamplesToRead == fread(frame.data_, sizeof(int16_t), |
254 kSamplesToRead, pcm_fid)) { | 254 kSamplesToRead, pcm_fid)) { |
255 assert(true_vad_index < kMaxNumFrames); | 255 assert(true_vad_index < kMaxNumFrames); |
256 ASSERT_EQ(1u, fread(&true_vad[true_vad_index], sizeof(*true_vad), 1, | 256 ASSERT_EQ(1u, fread(&true_vad[true_vad_index], sizeof(*true_vad), 1, |
257 true_vad_fid)) | 257 true_vad_fid)) |
258 << "Size mismatch between True-VAD and the PCM file.\n"; | 258 << "Size mismatch between True-VAD and the PCM file.\n"; |
259 if (video_vad_fid != NULL) { | 259 if (video_vad_fid != NULL) { |
(...skipping 25 matching lines...) Loading... |
285 agc_stat.Reset(); | 285 agc_stat.Reset(); |
286 } | 286 } |
287 true_vad_index++; | 287 true_vad_index++; |
288 | 288 |
289 DitherSilence(&frame); | 289 DitherSilence(&frame); |
290 | 290 |
291 ret_val = agc_stat.AddAudio(frame, p_video, agc_vad); | 291 ret_val = agc_stat.AddAudio(frame, p_video, agc_vad); |
292 ASSERT_GE(ret_val, 0); | 292 ASSERT_GE(ret_val, 0); |
293 | 293 |
294 if (ret_val > 0) { | 294 if (ret_val > 0) { |
295 ASSERT_EQ(true_vad_index, ret_val); | 295 ASSERT_EQ(true_vad_index, static_cast<size_t>(ret_val)); |
296 for (int n = 0; n < ret_val; n++) { | 296 for (int n = 0; n < ret_val; n++) { |
297 if (true_vad[n] == 1) { | 297 if (true_vad[n] == 1) { |
298 total_active++; | 298 total_active++; |
299 if (previous_true_vad == 0) { | 299 if (previous_true_vad == 0) { |
300 num_onsets++; | 300 num_onsets++; |
301 onset = true; | 301 onset = true; |
302 } | 302 } |
303 if (agc_vad[n] == 0) { | 303 if (agc_vad[n] == 0) { |
304 total_missed_detection++; | 304 total_missed_detection++; |
305 if (onset) | 305 if (onset) |
(...skipping 69 matching lines...) Loading... |
375 " one probability per frame.\n" | 375 " one probability per frame.\n" |
376 "\nUsage:\n\n" | 376 "\nUsage:\n\n" |
377 "activity_metric input_pcm [options]\n" | 377 "activity_metric input_pcm [options]\n" |
378 "where 'input_pcm' is the input audio sampled at 16 kHz in 16 bits " | 378 "where 'input_pcm' is the input audio sampled at 16 kHz in 16 bits " |
379 "format.\n\n"; | 379 "format.\n\n"; |
380 google::SetUsageMessage(kUsage); | 380 google::SetUsageMessage(kUsage); |
381 google::ParseCommandLineFlags(&argc, &argv, true); | 381 google::ParseCommandLineFlags(&argc, &argv, true); |
382 webrtc::void_main(argc, argv); | 382 webrtc::void_main(argc, argv); |
383 return 0; | 383 return 0; |
384 } | 384 } |
OLD | NEW |