forked from alumae/kaldi-offline-transcriber
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Makefile
401 lines (310 loc) · 15.4 KB
/
Makefile
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
SHELL := /bin/bash
# Use this file to override various settings
-include Makefile.options
DO_MUSIC_DETECTION?=yes
# Set to 'yes' if you want to do speaker ID for trs files
# Assumes you have models for speaker ID
DO_SPEAKER_ID?=yes
SID_THRESHOLD?=25
# Where is Kaldi root directory?
KALDI_ROOT?=/home/speech/tools/kaldi-trunk
# Location of the Java binary
JAVA_BIN?=/usr/bin/java
# How many processes to use for one transcription task
njobs ?= 1
# How many threads to use in each process
nthreads ?= 1
PATH := utils:$(KALDI_ROOT)/src/bin:$(KALDI_ROOT)/tools/openfst/bin:$(KALDI_ROOT)/src/fstbin/:$(KALDI_ROOT)/src/gmmbin/:$(KALDI_ROOT)/src/featbin/:$(KALDI_ROOT)/src/lm/:$(KALDI_ROOT)/src/sgmmbin/:$(KALDI_ROOT)/src/sgmm2bin/:$(KALDI_ROOT)/src/fgmmbin/:$(KALDI_ROOT)/src/latbin/:$(KALDI_ROOT)/src/nnet2bin/:$(KALDI_ROOT)/src/online2bin/:$(KALDI_ROOT)/src/kwsbin:$(KALDI_ROOT)/src/lmbin:$(PATH):$(KALDI_ROOT)/src/ivectorbin:$(PATH)
export train_cmd=run.pl
export decode_cmd=run.pl
export cuda_cmd=run.pl
export mkgraph_cmd=run.pl
# Main language model (should be slightly pruned), used for rescoring
LM ?=language_model/pruned.vestlused-dev.splitw2.arpa.gz
# More aggressively pruned LM, used in decoding
PRUNED_LM ?=language_model/pruned6.vestlused-dev.splitw2.arpa.gz
COMPOUNDER_LM ?=language_model/compounder-pruned.vestlused-dev.splitw.arpa.gz
# Vocabulary in dict format (no pronouncation probs for now)
VOCAB?=language_model/vestlused-dev.splitw2.dict
LM_SCALE?=17
DO_PUNCTUATION?=no
ifeq "yes" "$(DO_PUNCTUATION)"
PUNCTUATE_SYNC_TXT_CMD?=(cd ~/tools/punctuator/src; python wrapper.py)
DOT_PUNCTUATED=.punctuated
endif
# Find out where this Makefile is located (this is not really needed)
where-am-i = $(lastword $(MAKEFILE_LIST))
THIS_DIR := $(shell dirname $(call where-am-i))
FINAL_PASS=nnet2_online_ivector_pruned_rescored_main
LD_LIBRARY_PATH=$(KALDI_ROOT)/tools/openfst/lib
.SECONDARY:
.DELETE_ON_ERROR:
export
# Call this (once) before using the system
.init: .kaldi .lang
.kaldi:
rm -f steps utils sid
ln -s $(KALDI_ROOT)/egs/wsj/s5/steps
ln -s $(KALDI_ROOT)/egs/wsj/s5/utils
ln -s $(KALDI_ROOT)/egs/sre08/v1/sid
mkdir -p src-audio
.lang: build/fst/data/prunedlm build/fst/nnet2_online_ivector/graph_prunedlm build/fst/data/largelm build/fst/data/compounderlm
# Convert dict and LM to FST format
build/fst/data/dict build/fst/data/prunedlm: $(PRUNED_LM) $(VOCAB)
rm -rf build/fst/data/dict build/fst/data/prunedlm
mkdir -p build/fst/data/dict build/fst/data/prunedlm
cp -r $(THIS_DIR)/kaldi-data/dict/* build/fst/data/dict
rm -f build/fst/data/dict/lexicon.txt build/fst/data/dict/lexiconp.txt
cat models/etc/filler16k.dict | egrep -v "^<.?s>" > build/fst/data/dict/lexicon.txt
cat $(VOCAB) | perl -npe 's/\(\d\)(\s)/\1/' >> build/fst/data/dict/lexicon.txt
utils/prepare_lang.sh build/fst/data/dict "++garbage++" build/fst/data/dict/tmp build/fst/data/prunedlm
gunzip -c $(PRUNED_LM) | \
grep -v '<s> <s>' | \
grep -v '</s> <s>' | \
grep -v '</s> </s>' | \
arpa2fst - | fstprint | \
utils/eps2disambig.pl | utils/s2eps.pl | fstcompile --isymbols=build/fst/data/prunedlm/words.txt \
--osymbols=build/fst/data/prunedlm/words.txt --keep_isymbols=false --keep_osymbols=false | \
fstrmepsilon > build/fst/data/prunedlm/G.fst
fstisstochastic build/fst/data/prunedlm/G.fst || echo "Warning: LM not stochastic"
build/fst/data/largelm: build/fst/data/prunedlm $(LM)
rm -rf $@
mkdir -p $@
utils/build_const_arpa_lm.sh \
$(LM) build/fst/data/prunedlm $@
build/fst/data/compounderlm: $(COMPOUNDER_LM) $(VOCAB)
rm -rf $@
mkdir -p $@
cat $(VOCAB) | perl -npe 's/(\(\d\))?\s.+//' | uniq | python scripts/make-compounder-symbols.py > $@/words.txt
zcat $(COMPOUNDER_LM) | \
grep -v '<s> <s>' | \
grep -v '</s> <s>' | \
grep -v '</s> </s>' | \
arpa2fst - | fstprint | \
utils/s2eps.pl | fstcompile --isymbols=$@/words.txt --osymbols=$@/words.txt > $@/G.fst
# override some paths in nnet2 conf files
build/fst/nnet2_online_ivector/final.mdl:
rm -rf `dirname $@`
mkdir -p `dirname $@`
cp -r $(THIS_DIR)/kaldi-data/nnet2_online_ivector/* `dirname $@`
perl -i -npe 's/=.*online\//=build\/fst\/nnet2_online_ivector\//' build/fst/nnet2_online_ivector/conf/*.conf
build/fst/%/final.mdl:
rm -rf `dirname $@`
mkdir -p `dirname $@`
cp -r $(THIS_DIR)/kaldi-data/$*/* `dirname $@`
build/fst/%/graph_prunedlm: build/fst/data/prunedlm build/fst/%/final.mdl
rm -rf $@
utils/mkgraph.sh build/fst/data/prunedlm build/fst/$* $@
build/audio/base/%.wav: src-audio/%.wav
mkdir -p `dirname $@`
sox $^ -c 1 -2 build/audio/base/$*.wav rate -v 16k
build/audio/base/%.wav: src-audio/%.mp3
mkdir -p `dirname $@`
ffmpeg -i $^ -f sox - | sox -t sox - -c 1 -2 $@ rate -v 16k
build/audio/base/%.wav: src-audio/%.ogg
mkdir -p `dirname $@`
sox $^ -c 1 build/audio/base/$*.wav rate -v 16k
build/audio/base/%.wav: src-audio/%.mp2
mkdir -p `dirname $@`
sox $^ -c 1 build/audio/base/$*.wav rate -v 16k
build/audio/base/%.wav: src-audio/%.m4a
mkdir -p `dirname $@`
ffmpeg -i $^ -f sox - | sox -t sox - -c 1 -2 $@ rate -v 16k
build/audio/base/%.wav: src-audio/%.mp4
mkdir -p `dirname $@`
sox $^ -c 1 build/audio/base/$*.wav rate -v 16k
build/audio/base/%.wav: src-audio/%.flac
mkdir -p `dirname $@`
sox $^ -c 1 build/audio/base/$*.wav rate -v 16k
build/audio/base/%.wav: src-audio/%.amr
mkdir -p `dirname $@`
amrnb-decoder $^ $@.tmp.raw
sox -s -2 -c 1 -r 8000 $@.tmp.raw -c 1 build/audio/base/$*.wav rate -v 16k
rm $@.tmp.raw
build/audio/base/%.wav: src-audio/%.mpg
mkdir -p `dirname $@`
ffmpeg -i $^ -f sox - | sox -t sox - -c 1 -2 build/audio/base/$*.wav rate -v 16k
# Speaker diarization
build/diarization/%/show.seg: build/audio/base/%.wav
rm -rf `dirname $@`
mkdir -p `dirname $@`
echo "$* 1 0 1000000000 U U U 1" > `dirname $@`/show.uem.seg;
if [ $(DO_MUSIC_DETECTION) = yes ]; then diarization_opts="-m"; fi; \
./scripts/diarization.sh $$diarization_opts $^ `dirname $@`/show.uem.seg
build/audio/segmented/%: build/diarization/%/show.seg
rm -rf $@
mkdir -p $@
cat $^ | cut -f 3,4,8 -d " " | \
while read LINE ; do \
start=`echo $$LINE | cut -f 1 -d " " | perl -npe '$$_=$$_/100.0'`; \
len=`echo $$LINE | cut -f 2 -d " " | perl -npe '$$_=$$_/100.0'`; \
sp_id=`echo $$LINE | cut -f 3 -d " "`; \
timeformatted=`echo "$$start $$len" | perl -ne '@t=split(); $$start=$$t[0]; $$len=$$t[1]; $$end=$$start+$$len; printf("%08.3f-%08.3f\n", $$start,$$end);'` ; \
sox build/audio/base/$*.wav --norm $@/$*_$${timeformatted}_$${sp_id}.wav trim $$start $$len ; \
done
build/audio/segmented/%: build/diarization/%/show.seg
rm -rf $@
mkdir -p $@
cat $^ | cut -f 3,4,8 -d " " | \
while read LINE ; do \
start=`echo $$LINE | cut -f 1 -d " " | perl -npe '$$_=$$_/100.0'`; \
len=`echo $$LINE | cut -f 2 -d " " | perl -npe '$$_=$$_/100.0'`; \
sp_id=`echo $$LINE | cut -f 3 -d " "`; \
timeformatted=`echo "$$start $$len" | perl -ne '@t=split(); $$start=$$t[0]; $$len=$$t[1]; $$end=$$start+$$len; printf("%08.3f-%08.3f\n", $$start,$$end);'` ; \
sox build/audio/base/$*.wav --norm $@/$*_$${timeformatted}_$${sp_id}.wav trim $$start $$len ; \
done
build/trans/%/wav.scp: build/audio/segmented/%
mkdir -p `dirname $@`
/bin/ls $</*.wav | \
perl -npe 'chomp; $$orig=$$_; s/.*\/(.*)_(\d+\.\d+-\d+\.\d+)_(S\d+)\.wav/\1-\3---\2/; $$_=$$_ . " $$orig\n";' | LC_ALL=C sort > $@
build/trans/%/utt2spk: build/trans/%/wav.scp
cat $^ | perl -npe 's/\s+.*//; s/((.*)---.*)/\1 \2/' > $@
build/trans/%/spk2utt: build/trans/%/utt2spk
utils/utt2spk_to_spk2utt.pl $^ > $@
# MFCC calculation
build/trans/%/mfcc: build/trans/%/spk2utt
rm -rf $@
rm -f build/trans/$*/cmvn.scp
steps/make_mfcc.sh --mfcc-config conf/mfcc_hires.conf --cmd "$$train_cmd" --nj $(njobs) \
build/trans/$* build/trans/$*/exp/make_mfcc $@ || exit 1
steps/compute_cmvn_stats.sh build/trans/$* build/trans/$*/exp/make_mfcc $@ || exit 1
### Do 1-pass decoding using nnet2 online models
### We use but use steps/nnet2/decode.sh since it is multithreaded
build/trans/%/nnet2_online_ivector_pruned/decode/log: build/fst/nnet2_online_ivector/final.mdl build/fst/nnet2_online_ivector/graph_prunedlm build/trans/%/spk2utt build/trans/%/mfcc
rm -rf build/trans/$*/nnet2_online_ivector_pruned
mkdir -p build/trans/$*/nnet2_online_ivector_pruned
steps/online/nnet2/extract_ivectors_online.sh --cmd "$$decode_cmd" --nj $(njobs) \
build/trans/$* build/fst/nnet2_online_ivector/ivector_extractor build/trans/$*/nnet2_online/ivectors
(cd build/trans/$*/nnet2_online_ivector_pruned; for f in ../../../fst/nnet2_online_ivector/*; do ln -s $$f; done)
steps/nnet2/decode.sh --num-threads $(nthreads) --config conf/decode.conf --skip-scoring true --cmd "$$decode_cmd" --nj $(njobs) \
--online-ivector-dir build/trans/$*/nnet2_online/ivectors \
build/fst/nnet2_online_ivector/graph_prunedlm build/trans/$* `dirname $@` || exit 1;
(cd build/trans/$*/nnet2_online_ivector_pruned; ln -s ../../../fst/nnet2_online_ivector/graph_prunedlm graph)
# Rescore nnet2 lattices with a larger language model
build/trans/%/nnet2_online_ivector_pruned_rescored_main/decode/log: build/trans/%/nnet2_online_ivector_pruned/decode/log build/fst/data/largelm
rm -rf build/trans/$*/nnet2_online_ivector_pruned_rescored_main
mkdir -p build/trans/$*/nnet2_online_ivector_pruned_rescored_main
(cd build/trans/$*/nnet2_online_ivector_pruned_rescored_main; for f in ../../../fst/nnet2_online_ivector/*; do ln -s $$f; done)
steps/lmrescore_const_arpa.sh \
build/fst/data/prunedlm build/fst/data/largelm \
build/trans/$* \
build/trans/$*/nnet2_online_ivector_pruned/decode build/trans/$*/nnet2_online_ivector_pruned_rescored_main/decode || exit 1;
cp -r --preserve=links build/trans/$*/nnet2_online_ivector_pruned/graph build/trans/$*/nnet2_online_ivector_pruned_rescored_main/
%/decode/.ctm: %/decode/log
steps/get_ctm.sh `dirname $*` $*/graph $*/decode
touch -m $@
build/trans/%.segmented.splitw2.ctm: build/trans/%/decode/.ctm
cat build/trans/$*/decode/score_$(LM_SCALE)/`dirname $*`.ctm | perl -npe 's/(.*)-(S\d+)---(\S+)/\1_\3_\2/' > $@
%.with-compounds.ctm: %.splitw2.ctm build/fst/data/compounderlm
scripts/compound-ctm.py \
"scripts/compounder.py build/fst/data/compounderlm/G.fst build/fst/data/compounderlm/words.txt" \
< $*.splitw2.ctm > $@
%.segmented.ctm: %.segmented.with-compounds.ctm
cat $^ | grep -v "++" | grep -v "\[sil\]" | grep -v -e " $$" | perl -npe 's/\+//g' > $@
%.synced.ctm: %.segmented.ctm
cat $^ | python scripts/unsegment-ctm.py | LC_ALL=C sort -k 1,1 -k 3,3n -k 4,4n > $@
%.with-compounds.synced.ctm: %.segmented.with-compounds.ctm
cat $^ | python scripts/unsegment-ctm.py | LC_ALL=C sort -k 1,1 -k 3,3n -k 4,4n > $@
%.ctm: %.synced.ctm
cat $^ | grep -v "<" > $@
%.with-sil.ctm: %.ctm
cat $^ | ./scripts/ctm2with-sil-ctm.py > $@
%.punctuated.synced.txt: %.synced.with-sil.ctm
cat $^ | cut -f 5 -d " " | perl -npe 's/\n/ /' | $(PUNCTUATE_SYNC_TXT_CMD) > $@
%.synced.txt: %.synced.with-sil.ctm
cat $^ | cut -f 5 -d " " | perl -npe 's/\n/ /; s/<sil=\S+>//' > $@
%.hyp: %.segmented.ctm
cat $^ | python scripts/segmented-ctm-to-hyp.py > $@
ifeq "yes" "$(DO_SPEAKER_ID)"
ifeq "yes" "$(DO_MUSIC_DETECTION)"
build/trans/%/$(FINAL_PASS).trs: build/trans/%/$(FINAL_PASS)$(DOT_PUNCTUATED).synced.txt build/sid/%/sid-result.txt build/diarization/%/show.seg
cat build/trans/$*/$(FINAL_PASS)$(DOT_PUNCTUATED).synced.txt | python scripts/synced-txt-to-trs.py --fid $* --sid build/sid/$*/sid-result.txt --pms build/diarization/$*/show.pms.seg > $@
else
build/trans/%/$(FINAL_PASS).trs: build/trans/%/$(FINAL_PASS)$(DOT_PUNCTUATED).synced.txt build/sid/%/sid-result.txt
cat build/trans/$*/$(FINAL_PASS)$(DOT_PUNCTUATED).synced.txt | python scripts/synced-txt-to-trs.py --fid $* --sid build/sid/$*/sid-result.txt > $@
endif
else
ifeq "yes" "$(DO_MUSIC_DETECTION)"
build/trans/%/$(FINAL_PASS).trs: build/trans/%/$(FINAL_PASS)$(DOT_PUNCTUATED).synced.txt build/diarization/%/show.seg
cat build/trans/$*/$(FINAL_PASS)$(DOT_PUNCTUATED).synced.txt | python scripts/synced-txt-to-trs.py --fid $* --pms build/diarization/$*/show.pms.seg > $@
else
build/trans/%/$(FINAL_PASS).trs: build/trans/%/$(FINAL_PASS)$(DOT_PUNCTUATED).synced.txt
cat build/trans/$*/$(FINAL_PASS)$(DOT_PUNCTUATED).synced.txt | python scripts/synced-txt-to-trs.py --fid $* > $@
endif
endif
%.sbv: %.hyp
cat $^ | python scripts/hyp2sbv.py > $@
%.txt: %.trs
cat $^ | grep -v "^<" > $@
%.punctuated.hyp: %.hyp
cat $^ | perl -npe 's/ \(\S+\)$$//' | hidden-ngram -hidden-vocab $(PUNCTUATOR_HIDDEN_VOCAB) -order 4 -lm $(PUNCTUATOR_LM) -text - -keep-unk | \
perl -npe 's/ ,COMMA/,/g; s/ \.PERIOD/\./g' > $@.tmp
cat $^ | perl -npe 's/.*(\(\S+\))$$/ \1/' | paste $@.tmp - > $@
#rm $@.tmp
build/output/%.trs: build/trans/%/$(FINAL_PASS).trs
mkdir -p `dirname $@`
cp $^ $@
build/output/%.ctm: build/trans/%/$(FINAL_PASS).ctm
mkdir -p `dirname $@`
cp $^ $@
build/output/%.txt: build/trans/%/$(FINAL_PASS).txt
mkdir -p `dirname $@`
cp $^ $@
build/output/%.with-compounds.ctm: build/trans/%/$(FINAL_PASS).with-compounds.ctm
mkdir -p `dirname $@`
cp $^ $@
build/output/%.sbv: build/trans/%/$(FINAL_PASS).sbv
mkdir -p `dirname $@`
cp $^ $@
### Speaker ID stuff
# MFCC for Speaker ID, since the features for MFCC are different from speech recognition
build/sid/%/wav.scp: build/trans/%/wav.scp
mkdir -p `dirname $@`
ln $^ $@
build/sid/%/utt2spk : build/trans/%/utt2spk
mkdir -p `dirname $@`
ln $^ $@
build/sid/%/spk2utt : build/trans/%/spk2utt
mkdir -p `dirname $@`
ln $^ $@
build/sid/%/mfcc: build/sid/%/wav.scp build/sid/%/utt2spk build/sid/%/spk2utt
rm -rf $@
rm -f build/sid/$*/vad.scp
rm -f build/sid/$*/cmvn.scp
steps/make_mfcc.sh --mfcc-config conf/mfcc_sid.conf --cmd "$$train_cmd" --nj $(njobs) \
build/sid/$* build/sid/$*/exp/make_mfcc $@ || exit 1
steps/compute_cmvn_stats.sh build/sid/$* build/sid/$*/exp/make_mfcc $@ || exit 1
sid/compute_vad_decision.sh --nj $(njobs) --cmd "$$decode_cmd" \
build/sid/$* build/sid/$*/exp/make_vad $@ || exit 1
# i-vectors for each speaker in our audio file
build/sid/%/ivectors: build/sid/%/mfcc
rm -rf build/sid/$*/ivectors
sid/extract_ivectors.sh --cmd "$$decode_cmd" --nj $(njobs) \
$(THIS_DIR)/kaldi-data/extractor_2048_top500 build/sid/$* $@
# a cross product of train and test speakers
build/sid/%/sid-trials.txt: build/sid/%/ivectors
cut -f 1 -d " " $(THIS_DIR)/kaldi-data/ivectors_train_top500/spk_ivector.scp | \
while read a; do \
cut -f 1 -d " " build/sid/$*/ivectors/spk_ivector.scp | \
while read b; do \
echo "$$a $$b"; \
done ; \
done > $@
# similarity scores
build/sid/%/sid-scores.txt: build/sid/%/sid-trials.txt
ivector-plda-scoring \
"ivector-copy-plda --smoothing=0.1 $(THIS_DIR)/kaldi-data/ivectors_train_top500/plda - |" \
"ark:ivector-subtract-global-mean scp:$(THIS_DIR)/kaldi-data/ivectors_train_top500/spk_ivector.scp ark:- |" \
"ark:ivector-subtract-global-mean scp:build/sid/$*/ivectors/spk_ivector.scp ark:- |" \
build/sid/$*/sid-trials.txt $@
# pick speakers above the threshold
build/sid/%/sid-result.txt: build/sid/%/sid-scores.txt
cat build/sid/$*/sid-scores.txt | sort -u -k 2,2 -k 3,3nr | sort -u -k2,2 | \
awk 'int($$3)>=$(SID_THRESHOLD)' | perl -npe 's/(\S+) \S+-(S\d+) \S+/\2 \1/; s/-/ /g' | \
LC_ALL=C sort -k 2 | LC_ALL=C join -1 2 - $(THIS_DIR)/kaldi-data/ivectors_train_top500/speaker2names.txt | cut -f 2- -d " " > $@
# Meta-target that deletes all files created during processing a file. Call e.g. 'make .etteytlus2013.clean
.%.clean:
rm -rf build/audio/base/$*.wav build/audio/segmented/$* build/diarization/$* build/trans/$* build/sid/$*
# Also deletes the output files
.%.cleanest: .%.clean
rm -rf build/output/$*.{trs,txt,ctm,with-compounds.ctm,sbv}