1 /*
2 * Copyright (c) 2018 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 <array>
12 #include <string>
13 #include <vector>
14
15 #include "absl/flags/flag.h"
16 #include "absl/flags/parse.h"
17 #include "common_audio/resampler/push_sinc_resampler.h"
18 #include "common_audio/wav_file.h"
19 #include "modules/audio_processing/agc2/cpu_features.h"
20 #include "modules/audio_processing/agc2/rnn_vad/common.h"
21 #include "modules/audio_processing/agc2/rnn_vad/features_extraction.h"
22 #include "modules/audio_processing/agc2/rnn_vad/rnn.h"
23 #include "rtc_base/logging.h"
24 #include "rtc_base/numerics/safe_compare.h"
25
26 ABSL_FLAG(std::string, i, "", "Path to the input wav file");
27 ABSL_FLAG(std::string, f, "", "Path to the output features file");
28 ABSL_FLAG(std::string, o, "", "Path to the output VAD probabilities file");
29
30 namespace webrtc {
31 namespace rnn_vad {
32 namespace test {
33
main(int argc,char * argv[])34 int main(int argc, char* argv[]) {
35 absl::ParseCommandLine(argc, argv);
36 rtc::LogMessage::LogToDebug(rtc::LS_INFO);
37
38 // Open wav input file and check properties.
39 const std::string input_wav_file = absl::GetFlag(FLAGS_i);
40 WavReader wav_reader(input_wav_file);
41 if (wav_reader.num_channels() != 1) {
42 RTC_LOG(LS_ERROR) << "Only mono wav files are supported";
43 return 1;
44 }
45 if (wav_reader.sample_rate() % 100 != 0) {
46 RTC_LOG(LS_ERROR) << "The sample rate rate must allow 10 ms frames.";
47 return 1;
48 }
49 RTC_LOG(LS_INFO) << "Input sample rate: " << wav_reader.sample_rate();
50
51 // Init output files.
52 const std::string output_vad_probs_file = absl::GetFlag(FLAGS_o);
53 FILE* vad_probs_file = fopen(output_vad_probs_file.c_str(), "wb");
54 FILE* features_file = nullptr;
55 const std::string output_feature_file = absl::GetFlag(FLAGS_f);
56 if (!output_feature_file.empty()) {
57 features_file = fopen(output_feature_file.c_str(), "wb");
58 }
59
60 // Initialize.
61 const int frame_size_10ms =
62 rtc::CheckedDivExact(wav_reader.sample_rate(), 100);
63 std::vector<float> samples_10ms;
64 samples_10ms.resize(frame_size_10ms);
65 std::array<float, kFrameSize10ms24kHz> samples_10ms_24kHz;
66 PushSincResampler resampler(frame_size_10ms, kFrameSize10ms24kHz);
67 const AvailableCpuFeatures cpu_features = GetAvailableCpuFeatures();
68 FeaturesExtractor features_extractor(cpu_features);
69 std::array<float, kFeatureVectorSize> feature_vector;
70 RnnVad rnn_vad(cpu_features);
71
72 // Compute VAD probabilities.
73 while (true) {
74 // Read frame at the input sample rate.
75 const size_t read_samples =
76 wav_reader.ReadSamples(frame_size_10ms, samples_10ms.data());
77 if (rtc::SafeLt(read_samples, frame_size_10ms)) {
78 break; // EOF.
79 }
80 // Resample input.
81 resampler.Resample(samples_10ms.data(), samples_10ms.size(),
82 samples_10ms_24kHz.data(), samples_10ms_24kHz.size());
83
84 // Extract features and feed the RNN.
85 bool is_silence = features_extractor.CheckSilenceComputeFeatures(
86 samples_10ms_24kHz, feature_vector);
87 float vad_probability =
88 rnn_vad.ComputeVadProbability(feature_vector, is_silence);
89 // Write voice probability.
90 RTC_DCHECK_GE(vad_probability, 0.f);
91 RTC_DCHECK_GE(1.f, vad_probability);
92 fwrite(&vad_probability, sizeof(float), 1, vad_probs_file);
93 // Write features.
94 if (features_file) {
95 const float float_is_silence = is_silence ? 1.f : 0.f;
96 fwrite(&float_is_silence, sizeof(float), 1, features_file);
97 if (is_silence) {
98 // Do not write uninitialized values.
99 feature_vector.fill(0.f);
100 }
101 fwrite(feature_vector.data(), sizeof(float), kFeatureVectorSize,
102 features_file);
103 }
104 }
105
106 // Close output file(s).
107 fclose(vad_probs_file);
108 RTC_LOG(LS_INFO) << "VAD probabilities written to " << output_vad_probs_file;
109 if (features_file) {
110 fclose(features_file);
111 RTC_LOG(LS_INFO) << "features written to " << output_feature_file;
112 }
113
114 return 0;
115 }
116
117 } // namespace test
118 } // namespace rnn_vad
119 } // namespace webrtc
120
main(int argc,char * argv[])121 int main(int argc, char* argv[]) {
122 return webrtc::rnn_vad::test::main(argc, argv);
123 }
124