forked from kaldi-asr/kaldi
-
Notifications
You must be signed in to change notification settings - Fork 0
/
run.sh
executable file
·427 lines (356 loc) · 7.42 KB
/
run.sh
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
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
#!/bin/bash
. ./cmd.sh
. ./path.sh
stage=0
. ./utils/parse_options.sh
set -e
set -o pipefail
set u
# the location of the LDC corpus
datadir=/mnt/corpora/LDC2006S37/data
# acoustic models are trained on the heroico corpus
# testing is done on the usma corpus
# heroico consists of 2 parts: answers and recordings (recited)
answers_transcripts=$datadir/transcripts/heroico-answers.txt
recordings_transcripts=$datadir/transcripts/heroico-recordings.txt
# usma is all recited
usma_transcripts=$datadir/transcripts/usma-prompts.txt
tmpdir=data/local/tmp
# make acoustic model training lists
if [ $stage -le 0 ]; then
mkdir \
-p \
$tmpdir/heroico \
$tmpdir/usma
local/get_wav_list.sh \
$datadir
# make separate lists for heroico answers and recordings
export LC_ALL=en_US.UTF-8
cat \
$answers_transcripts \
| \
iconv -f ISO-8859-1 -t UTF-8 \
| \
sed -e s/// \
| \
local/heroico_answers_make_lists.pl
utils/fix_data_dir.sh \
$tmpdir/heroico/answers
cat \
$recordings_transcripts \
| \
iconv -f ISO-8859-1 -t UTF-8 \
| \
sed -e s/// \
| \
local/heroico_recordings_make_lists.pl
utils/fix_data_dir.sh \
$tmpdir/heroico/recordings
# consolidate heroico lists
mkdir -p $tmpdir/heroico/lists
for x in wav.scp utt2spk text; do
cat \
$tmpdir/heroico/answers/$x \
$tmpdir/heroico/recordings/$x \
| \
sed -e s/// \
| \
sort \
-k1,1 \
-u \
> \
$tmpdir/heroico/lists/$x
done
utils/fix_data_dir.sh \
$tmpdir/heroico/lists
fi
if [ $stage -le 1 ]; then
# make separate lists for usma native and nonnative
cat \
$usma_transcripts \
| \
iconv -f ISO-8859-1 -t UTF-8 \
| \
sed -e s/// \
| \
local/usma_native_make_lists.pl
cat \
$usma_transcripts \
| \
iconv -f ISO-8859-1 -t UTF-8 \
| \
sed -e s/// \
| \
local/usma_nonnative_make_lists.pl
for n in native nonnative; do
mkdir -p $tmpdir/usma/$n/lists
for x in wav.scp utt2spk text; do
sort \
$tmpdir/usma/$n/$x \
> \
$tmpdir/usma/$n/lists/$x
done
utils/fix_data_dir.sh \
$tmpdir/usma/$n/lists
done
mkdir -p data/train
mkdir -p $tmpdir/lists
# get training lists
for x in wav.scp utt2spk text; do
cat \
$tmpdir/heroico/answers/${x} \
$tmpdir/heroico/recordings/${x} \
| \
sed -e s/// \
> \
$tmpdir/lists/$x
sort \
$tmpdir/lists/$x \
> \
data/train/$x
done
utils/utt2spk_to_spk2utt.pl \
data/train/utt2spk \
| \
sort \
> \
data/train/spk2utt
utils/fix_data_dir.sh \
data/train
# make testing lists
mkdir \
-p \
data/test \
data/native \
data/nonnative \
$tmpdir/usma/lists
for x in wav.scp text utt2spk; do
# get testing lists
for n in native nonnative; do
cat \
$tmpdir/usma/$n/lists/$x \
>> \
$tmpdir/usma/lists/$x
done
sort \
$tmpdir/usma/lists/$x \
> \
data/test/$x
for n in native nonnative; do
sort \
$tmpdir/usma/$n/$x \
> \
data/$n/$x
done
done
# spk2utt
for n in native nonnative test; do
utils/utt2spk_to_spk2utt.pl \
data/$n/utt2spk \
| \
sort \
> \
data/$n/spk2utt
utils/fix_data_dir.sh \
data/$n
done
fi
if [ $stage -le 2 ]; then
# prepare a dictionary
mkdir -p data/local/dict
mkdir -p data/local/tmp/dict
# download the dictionary from openslr
if [ ! -f data/local/tmp/dict/santiago.tar.gz ]; then
wget \
-O data/local/tmp/dict/santiago.tar.gz \
https://www.openslr.org/resources/34/santiago.tar.gz
fi
if [ -e data/local/tmp/dict/santiago.tar ]; then
rm data/local/tmp/dict/santiago.tar
fi
gunzip data/local/tmp/dict/santiago.tar.gz
cd data/local/tmp/dict
tar -xvf santiago.tar
cd ../../../..
local/prepare_dict.sh
# prepare the lang directory
utils/prepare_lang.sh \
data/local/dict \
"<UNK>" \
data/local/lang \
data/lang || exit 1;
fi
if [ $stage -le 3 ]; then
# prepare lm on training transcripts
local/prepare_lm.sh
utils/format_lm.sh \
data/lang \
data/local/lm/lm_threegram.arpa.gz \
data/local/dict/lexicon.txt \
data/lang_test
fi
if [ $stage -le 4 ]; then
# extract acoustic features
mkdir -p exp
for fld in native nonnative test train; do
if [ -e data/$fld/cmvn.scp ]; then
rm data/$fld/cmvn.scp
fi
steps/make_mfcc.sh \
--cmd "$train_cmd" \
--nj 4 \
data/$fld \
exp/make_mfcc/$fld \
mfcc || exit 1;
utils/fix_data_dir.sh \
data/$fld || exit 1;
steps/compute_cmvn_stats.sh \
data/$fld \
exp/make_mfcc\
mfcc || exit 1;
utils/fix_data_dir.sh \
data/$fld || exit 1;
done
fi
if [ $stage -le 5 ]; then
echo "monophone training"
steps/train_mono.sh \
--nj 4 \
--cmd "$train_cmd" \
data/train \
data/lang \
exp/mono || exit 1;
# align with monophones
steps/align_si.sh \
--nj 8 \
--cmd "$train_cmd" \
data/train \
data/lang \
exp/mono \
exp/mono_ali || exit 1;
fi
if [ $stage -le 6 ]; then
echo "Starting triphone training in exp/tri1"
steps/train_deltas.sh \
--cmd "$train_cmd" \
--cluster-thresh 100 \
1500 \
25000 \
data/train \
data/lang \
exp/mono_ali \
exp/tri1 || exit 1;
# align with triphones
steps/align_si.sh \
--nj 8 \
--cmd "$train_cmd" \
data/train \
data/lang \
exp/tri1 \
exp/tri1_ali
fi
if [ $stage -le 7 ]; then
echo "Starting (lda_mllt) triphone training in exp/tri2b"
steps/train_lda_mllt.sh \
--splice-opts "--left-context=3 --right-context=3" \
2000 \
30000 \
data/train \
data/lang \
exp/tri1_ali \
exp/tri2b
# align with lda and mllt adapted triphones
steps/align_si.sh \
--use-graphs true \
--nj 8 \
--cmd "$train_cmd" \
data/train \
data/lang \
exp/tri2b \
exp/tri2b_ali
echo "Starting (SAT) triphone training in exp/tri3b"
steps/train_sat.sh \
--cmd "$train_cmd" \
3100 \
50000 \
data/train \
data/lang \
exp/tri2b_ali \
exp/tri3b
# align with tri3b models
echo "Starting exp/tri3b_ali"
steps/align_fmllr.sh \
--nj 8 \
--cmd "$train_cmd" \
data/train \
data/lang \
exp/tri3b \
exp/tri3b_ali
fi
if [ $stage -le 8 ]; then
# train and test chain models
local/chain/run_tdnn.sh
fi
if [ $stage -le 9 ]; then
# evaluation
# make decoding graph for monophones
utils/mkgraph.sh \
data/lang \
exp/mono \
exp/mono/graph || exit 1;
# test monophones
for x in native nonnative test; do
steps/decode.sh \
--nj 8 \
exp/mono/graph \
data/$x \
exp/mono/decode_${x} || exit 1;
done
# test cd gmm hmm models
# make decoding graph for tri1
utils/mkgraph.sh \
data/lang \
exp/tri1 \
exp/tri1/graph || exit 1;
# decode test data with tri1 models
for x in native nonnative test; do
steps/decode.sh \
--nj 8 \
exp/tri1/graph \
data/$x \
exp/tri1/decode_${x} || exit 1;
done
# make decoding fst for tri2b models
utils/mkgraph.sh \
data/lang \
exp/tri2b \
exp/tri2b/graph || exit 1;
# decode test with tri2b models
for x in native nonnative test; do
steps/decode.sh \
--nj 8 \
exp/tri2b/graph \
data/$x \
exp/tri2b/decode_${x} || exit 1;
done
# make decoding graph for SAT models
utils/mkgraph.sh \
data/lang \
exp/tri3b \
exp/tri3b/graph || exit 1;
# decode test set with tri3b models
for x in native nonnative test; do
steps/decode_fmllr.sh \
--nj 8 \
--cmd "$decode_cmd" \
exp/tri3b/graph \
data/$x \
exp/tri3b/decode_${x}
done
fi