-
Notifications
You must be signed in to change notification settings - Fork 10
/
longaudio_alignment.sh
executable file
·155 lines (151 loc) · 7.77 KB
/
longaudio_alignment.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
#!/bin/bash
# Copyright 2017 Speech Lab, EE Dept., IITM (Author: Srinivas Venkattaramanujam)
. ./path.sh # ensure kaldi, IRSTLM and sctk are in path
. ./cmd.sh
. ./longaudio_vars.sh
set -e
stage=0
working_dir=data/working_dir
new_dir=${data_dir}_segmented
create_dir="false"
while [[ $# -gt 1 ]]
do
arg=$1
case $arg in
--working-dir)
working_dir=$2
shift
shift
;;
--stage)
stage=$2
shift
shift
;;
--create-dir)
create_dir=$2
shift
shift
;;
esac
done;
segment_store=$working_dir/segments_store
log_dir=$working_dir/log
mkdir -p $working_dir
mkdir -p $log_dir
mkdir -p $segment_store
echo "Taking backup of $data_dir to ${data_dir}.laa.bkp"
rm -rf ${data_dir}.laa.bkp || echo "" > $log_dir/output.log 2>$log_dir/err.log || exit 1
cp -r $data_dir ${data_dir}.laa.bkp > $log_dir/output.log 2>$log_dir/err.log || exit
echo "Params: working_dir=$working_dir stage=$stage log directory=$log_dir"
if [ $stage -ge 1 ]; then
# mfcc and cmvn
(rm $data_dir/segments || echo "") >> $log_dir/output.log 2>&1
(mv $data_dir/text $data_dir/text_1 > $log_dir/output.log 2>$log_dir/err.log) || exit 1
echo "Making feats"
scripts/make-feats.sh $data_dir/wav.scp $working_dir $log_dir 2> $log_dir/err.log
# VAD and segmentation based on VAD
#head -1 $data_dir/feats.scp
echo "Doing VAD"
(compute-vad scp:$data_dir/feats.scp ark,t:- 2> $log_dir/err.log || exit 1) | cut -d' ' -f2- | tr -d ' '|tr -d '[' | tr -d ']' > $working_dir/vad.ark || exit 1
echo "Making segments using VAD"
# split_vad.py considers even one frame of 0 (silence) as potential breakpoint. But you might want to change it
(scripts/split_vad.py $working_dir/vad.ark 2> ${log_dir}/err.log || exit 1) | sort > $data_dir/segments
cp $data_dir/segments $working_dir/segments 2> $log_dir/err.log || exit 1
echo "Computing features for segments obtained using VAD"
scripts/make-feats.sh $data_dir/segments $working_dir $log_dir 2>${log_dir}/err.log
# prepare text file
echo "Preparing text files: text_actual"
(cat $data_dir/text_1 2> $log_dir/err.log || exit 1) | cut -d' ' -f2- | sed 's/^ \+//g' | sed 's/ \+$//g' | tr -s ' ' > $working_dir/text_actual
echo "Preparing text files: lm_text"
(cat $working_dir/text_actual 2> $log_dir/err.log || exit 1) | sed -e 's:^:<s> :' -e 's:$: </s>:' > $working_dir/lm_text
echo "Preparing text files: initializing WORD_TIMINGS file with all -1 -1"
(scripts/sym2int.py ${lang_dir}/words.txt $working_dir/text_actual 2> $log_dir/err.log || exit 1) | tr ' ' '\n' | sed 's/$/ -1 -1/g' > $working_dir/WORD_TIMINGS
echo "Preparation of text files over"
echo "Preparing trigram LM"
scripts/build-trigram.sh $working_dir $working_dir/lm_text >> $log_dir/output.log 2> $log_dir/err.log || exit 1
echo "Trigram LM created using $working_dir/lm_text"
# build graph and decode
echo "Executing build-graph-decode-hyp.sh"
num_lines=`wc -l $data_dir/feats.scp | cut -d' ' -f1` # min of num_lines and 20 for num_jobs
scripts/build-graph-decode-hyp.sh $(($num_lines>20?20:$num_lines)) decode $working_dir $log_dir 2> $log_dir/err.log || exit 1
echo "iter 0 decode over"
# create a status file which specifies which segments are done and pending and save timing information for each aligned word
num_text_words=`wc -w $working_dir/text_ints | cut -d' ' -f1`
text_end_index=$((num_text_words-1))
audio_duration=`(wav-to-duration --read-entire-file scp:$data_dir/wav.scp ark,t:- 2>> $log_dir/output.log) | cut -d' ' -f2`
scripts/make-status-and-word-timings.sh $working_dir $working_dir 0 $text_end_index 0.00 $audio_duration $log_dir 2> $log_dir/err.log || (echo "Failed: make-status-and-word-timings.sh" && exit 1)
fi
if [ $stage -ge 2 ]; then
segment_id=`wc -l $working_dir/segments | cut -d' ' -f1`
for x in `seq 1 $((num_iters-1))`;do
# echo "segment id is $segment_id"
# grep PENDING from status file
# for each PENDING entry, do; tc of segment_id
# make segment file, utt2spk, spk2utt
# make lm
# mkgraph
# decode
# get the decoded output and put in the TEMPSTATUS file
# merge TEMPSTATUS and STATUS
# repeat
echo "Doing iteration ${x}. Starting segment id: $segment_id"
while read y;do
echo $y >> $log_dir/output.log
mkdir -p $segment_store/${segment_id}
# make segments 10-15 seconds segments TODO
echo "segment_$segment_id key_1 `echo $y | cut -d' ' -f 1,2 `" > $data_dir/segments
scripts/make-feats.sh $data_dir/segments $working_dir $log_dir 2>${log_dir}/err.log
cp $data_dir/segments $segment_store/${segment_id}/segments
time_begin="`echo $y | cut -d' ' -f1`"
time_end="`echo $y | cut -d' ' -f2`"
word_begin_index=`echo $y | cut -d' ' -f4 `
word_begin_index=$((word_begin_index+1))
word_end_index=`echo $y | cut -d' ' -f5`
word_end_index=$((word_end_index+1))
word_string=`cat $working_dir/text_actual | cut -d' ' -f $word_begin_index-$word_end_index`
word_begin_index=$((word_begin_index-1))
word_end_index=$((word_end_index-1))
echo "<s> $word_string </s>" > $segment_store/${segment_id}/lm_text
echo "$word_string" > $segment_store/${segment_id}/text_actual
if [ $x -eq $((num_iters-2)) ]; then
scripts/build-transducer.sh $segment_store/${segment_id} $segment_store/${segment_id}/text_actual false >> $log_dir/output.log 2> $log_dir/err.log || exit 1
elif [ $x -eq $((num_iters-1)) ]; then
scripts/build-transducer.sh $segment_store/${segment_id} $segment_store/${segment_id}/text_actual true >> $log_dir/output.log 2> $log_dir/err.log || exit 1
else
scripts/build-trigram.sh $segment_store/${segment_id} $segment_store/${segment_id}/lm_text >> $log_dir/output.log 2> $log_dir/err.log || exit 1
fi
scripts/build-graph-decode-hyp.sh 1 decode_${segment_id} $segment_store/${segment_id} $log_dir 2> $log_dir/err.log || exit 1
scripts/make-status-and-word-timings.sh $working_dir $segment_store/${segment_id} \
$word_begin_index $word_end_index $time_begin $time_end $log_dir 2> $log_dir/err.log || (echo "Failed: make-status-and-word-timings.sh" && exit 1)
cat $segment_store/${segment_id}/ALIGNMENT_STATUS >> $working_dir/ALIGNMENT_STATUS.working.iter${x} # this file is appended with ALIGNMENT_STATUS of each segment of the iteration.
segment_id=$((segment_id+1))
done < <(cat $working_dir/ALIGNMENT_STATUS | grep PENDING)
cp $working_dir/ALIGNMENT_STATUS $working_dir/ALIGNMENT_STATUS.iter$((x-1))
cat $working_dir/ALIGNMENT_STATUS | grep 'DONE' > $working_dir/ALIGNMENT_STATUS.tmp
cat $working_dir/ALIGNMENT_STATUS.working.iter${x} >> $working_dir/ALIGNMENT_STATUS.tmp
cat $working_dir/ALIGNMENT_STATUS.tmp | sort -s -k 1,1n > $working_dir/ALIGNMENT_STATUS.tmp2
# clean up the alignment file so that ALIGNMENT_STATUS has DONE and PENDING in alternate lines
echo "Cleaning up Alignment Status" >> $log_dir/output.log
scripts/cleanup_status.py $working_dir/ALIGNMENT_STATUS.tmp2 > $working_dir/ALIGNMENT_STATUS
rm $working_dir/ALIGNMENT_STATUS.tmp*
rm $working_dir/ALIGNMENT_STATUS.working.iter${x} # might need for debugging
done;
fi
rm -r $model_dir/decode_* || echo ""
#echo "converting integer ids to words in"
utils/int2sym.pl -f 1 $lang_dir/words.txt $working_dir/WORD_TIMINGS > $working_dir/WORD_TIMINGS.words
rm -rf $data_dir
mv ${data_dir}.laa.bkp $data_dir
if [ $create_dir == "true" ]; then
echo "Creating $new_dir"
mkdir -p $new_dir
x=`echo "$data_dir" | rev | cut -d'/' -f1 | rev`
# the following script makes a segment with 10 words but if there is no timing info for the 10th word, we proceed until we find a word with known timing
scripts/timing_to_segment_and_text.py $working_dir/WORD_TIMINGS.words $x $new_dir/segments $new_dir/text `(wav-to-duration --read-entire-file scp:${data_dir}/wav.scp ark,t:- 2>> $log_dir/output.log) | cut -d' ' -f2`
echo "${x} `cat ${data_dir}/wav.scp|cut -d' ' -f2-`" > $new_dir/wav.scp
cut -d ' ' -f1 $new_dir/segments | sed "s/$/ $x/g" > $new_dir/utt2spk
cut -d ' ' -f1 $new_dir/segments | sed "s/^/$x /g" > $new_dir/spk2utt
fi
touch $new_dir/.done
echo 'Finished successfully'