-
Notifications
You must be signed in to change notification settings - Fork 92
/
apply-supervised-mvdr.cc
253 lines (215 loc) · 9.3 KB
/
apply-supervised-mvdr.cc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
// src/apply-supervised-mvdr.cc
// wujian@2018
// Copyright 2018 Jian Wu
// See ../../COPYING for clarification regarding multiple authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
// WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
// MERCHANTABLITY OR NON-INFRINGEMENT.
// See the Apache 2 License for the specific language governing permissions and
// limitations under the License.
#include "include/stft.h"
#include "include/beamformer.h"
using namespace kaldi;
void ParseInputRspecifier(const std::string &input_rspecifier,
std::vector<std::string> *rspecifiers) {
size_t found = input_rspecifier.find_first_of(":", 0);
if (found == std::string::npos)
KALDI_ERR << "Wrong input-rspecifier format: " << input_rspecifier;
const std::string &decorator = input_rspecifier.substr(0, found);
std::vector<std::string> tmp;
SplitStringToVector(input_rspecifier.substr(found + 1), ",", false, &tmp);
for (std::string &s : tmp) rspecifiers->push_back(decorator + ":" + s);
}
int main(int argc, char *argv[]) {
try {
const char *usage =
"Do minimum variance distortionless response (MVDR) beamformer, "
"depending on TF mask\n"
"\n"
"Usage: apply-supervised-mvdr [options...] <mask-rspecifier> "
"<input-rspecifier> <target-wav-wspecifier>\n"
"\n"
"e.g.:\n"
" apply-supervised-mvdr --config=mask.conf scp:mask.scp "
"scp:CH1.scp,CH2.scp,CH3.scp scp:dst.scp\n";
ParseOptions po(usage);
ShortTimeFTOptions stft_options;
bool track_volumn = true, normalize_input = true;
std::string window = "hamming";
BaseFloat frame_shift = 256, frame_length = 1024;
int32 update_periods = 0, minimum_update_periods = 20;
po.Register("frame-shift", &frame_shift,
"Frame shift in number of samples");
po.Register("frame-length", &frame_length,
"Frame length in number of samples");
po.Register(
"window", &window,
"Type of window(\"hamming\"|\"hanning\"|\"blackman\"|\"rectangular\")");
po.Register("track-volumn", &track_volumn,
"If true, using average volumn of input channels as target's");
po.Register(
"normalize-input", &normalize_input,
"Scale samples into float in range [-1, 1], like MATLAB or librosa");
po.Register(
"update-periods", &update_periods,
"Number of frames to use for estimating psd of noise or target, "
"if zero, do beamforming offline");
po.Read(argc, argv);
int32 num_args = po.NumArgs();
if (num_args != 3) {
po.PrintUsage();
exit(1);
}
KALDI_ASSERT(update_periods >= 0);
if (update_periods < minimum_update_periods && update_periods > 0) {
KALDI_WARN << "Value of update_periods may be too small, ignore it";
}
std::string mask_rspecifier = po.GetArg(1), input_rspecifier = po.GetArg(2),
enhan_wspecifier = po.GetArg(3);
std::vector<std::string> rspecifiers;
ParseInputRspecifier(input_rspecifier, &rspecifiers);
int32 num_channels = rspecifiers.size();
// Construct wave reader.
std::vector<RandomAccessTableReader<WaveHolder> > wav_reader(num_channels);
for (int32 c = 0; c < num_channels; c++) {
std::string &cur_ch = rspecifiers[c];
if (ClassifyRspecifier(cur_ch, NULL, NULL) == kNoRspecifier)
KALDI_ERR << cur_ch << " is not a rspecifier";
KALDI_ASSERT(wav_reader[c].Open(cur_ch));
}
// config stft options
stft_options.window = window;
stft_options.normalize_input = normalize_input;
stft_options.frame_shift = frame_shift;
stft_options.frame_length = frame_length;
ShortTimeFTComputer stft_computer(stft_options);
SequentialBaseFloatMatrixReader mask_reader(mask_rspecifier);
TableWriter<WaveHolder> wav_writer(enhan_wspecifier);
int32 num_done = 0, num_miss = 0, num_utts = 0;
for (; !mask_reader.Done(); mask_reader.Next()) {
std::string utt_key = mask_reader.Key();
const Matrix<BaseFloat> &target_mask = mask_reader.Value();
// init num_channels
// mstft: cache for realfft of each channel
std::vector<Matrix<BaseFloat> > mstft(num_channels);
std::vector<BaseFloat> mfreq(num_channels);
BaseFloat range = 0.0;
num_utts++;
int32 cur_ch = 0;
for (int32 c = 0; c < num_channels; c++) {
if (wav_reader[c].HasKey(utt_key)) {
const WaveData &wave_data = wav_reader[c].Value(utt_key);
const Matrix<BaseFloat> &wave_samp = wave_data.Data();
if (track_volumn) range += wave_samp.LargestAbsElem();
mfreq[cur_ch] = wave_data.SampFreq();
stft_computer.Compute(wave_samp, &mstft[cur_ch], NULL, NULL);
cur_ch++;
}
}
KALDI_VLOG(2) << "Processing " << cur_ch << " channels for " << utt_key;
// do not process if num_channels <= 1
if (cur_ch <= 1) {
num_miss++;
continue;
}
// check dimentions
// mstft[..].NumCols() == frame_length
int32 num_frames = mstft[0].NumRows(),
num_bins = mstft[0].NumCols() / 2 + 1;
BaseFloat target_freq = mfreq[0];
bool problem = false;
for (int32 c = 1; c < cur_ch; c++) {
if (mstft[c].NumCols() != (num_bins - 1) * 2 ||
mstft[c].NumRows() != num_frames) {
KALDI_WARN << "There is obvious length difference between"
<< "multiple channels, please check, skip for " << utt_key;
problem = true;
break;
}
if (target_freq != mfreq[c]) {
KALDI_WARN << "Sample frequency may be difference between"
<< "multiple channels, please check, skip for " << utt_key;
problem = true;
break;
}
}
if (problem) {
num_miss++;
continue;
}
// target_mask is a real matrix
if (target_mask.NumRows() != num_frames ||
target_mask.NumCols() != num_bins) {
KALDI_WARN << "Utterance " << utt_key
<< ": The shape of target mask is different from stft"
<< " (" << target_mask.NumRows() << " x "
<< target_mask.NumCols() << ") vs"
<< " (" << num_frames << " x " << num_bins << ")";
num_miss++;
continue;
}
CMatrix<BaseFloat> stft_reshape(num_frames, num_bins * cur_ch), src_stft;
for (int32 c = 0; c < cur_ch; c++) {
stft_reshape.ColRange(c * num_bins, num_bins).CopyFromRealfft(mstft[c]);
}
CMatrix<BaseFloat> noise_psd, target_psd, steer_vector, beam_weights,
enh_stft;
int32 num_segments =
(num_frames - minimum_update_periods) / update_periods + 1;
if (update_periods >= minimum_update_periods && num_segments > 1) {
KALDI_VLOG(1) << "Do mvdr beamforming, update power spectrum matrix "
"estimation per "
<< update_periods << " frames";
int32 duration = 0, start_from = 0;
CMatrix<BaseFloat> enh_stft_segment;
enh_stft.Resize(num_frames, num_bins);
for (int32 i = 0; i < num_segments; i++) {
start_from = i * update_periods;
duration = (i == num_segments - 1 ? num_frames - start_from
: update_periods);
TrimStft(num_bins, cur_ch,
stft_reshape.RowRange(start_from, duration), &src_stft);
EstimatePsd(src_stft, target_mask.RowRange(start_from, duration),
&target_psd, &noise_psd);
EstimateSteerVector(target_psd, &steer_vector);
ComputeMvdrBeamWeights(noise_psd, steer_vector, &beam_weights);
Beamform(src_stft, beam_weights, &enh_stft_segment);
enh_stft.RowRange(start_from, duration).CopyFromMat(enh_stft_segment);
}
} else {
KALDI_VLOG(1) << "Do mvdr beamforming offline";
TrimStft(num_bins, cur_ch, stft_reshape, &src_stft);
EstimatePsd(src_stft, target_mask, &target_psd, &noise_psd);
EstimateSteerVector(target_psd, &steer_vector);
ComputeMvdrBeamWeights(noise_psd, steer_vector, &beam_weights);
Beamform(src_stft, beam_weights, &enh_stft);
}
Matrix<BaseFloat> rstft, enhan_speech;
CastIntoRealfft(enh_stft, &rstft);
stft_computer.InverseShortTimeFT(rstft, &enhan_speech,
range / cur_ch - 1);
WaveData target_data(target_freq, enhan_speech);
wav_writer.Write(utt_key, target_data);
num_done++;
if (num_done % 100 == 0)
KALDI_LOG << "Processed " << num_utts << " utterances.";
KALDI_VLOG(2) << "Do mvdr beamforming for utterance-id " << utt_key
<< " done.";
}
KALDI_LOG << "Done " << num_done << " utterances out of " << num_utts
<< ", " << num_miss << " missing cause of some problems.";
return num_done == 0 ? 1 : 0;
} catch (const std::exception &e) {
std::cerr << e.what();
return -1;
}
return 0;
}