-
Notifications
You must be signed in to change notification settings - Fork 342
/
decode_ctc.sh
executable file
·94 lines (74 loc) · 3.43 KB
/
decode_ctc.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
#!/bin/bash
# Copyright 2012 Johns Hopkins University (Author: Daniel Povey)
# 2015 Yajie Miao (Carnegie Mellon University)
# Apache 2.0
# Decode the CTC-trained model. Currently we are using the simplest best-path decoding. Lattice-based
# decoding and other formats of decoding outputs (e.g., CTM) will be added in our future development.
## Begin configuration section
stage=0
nj=16
cmd=run.pl
num_threads=1
max_active=7000 # max-active
beam=15.0 # beam used
skip_scoring=false # whether to skip WER scoring
acoustic_scales="0.5 0.6 0.7 0.8" # the acoustic scales to be used
# feature configurations; will be read from the training dir if not provided
norm_vars=
add_deltas=
## End configuration section
echo "$0 $@" # Print the command line for logging
[ -f ./path.sh ] && . ./path.sh;
. parse_options.sh || exit 1;
if [ $# != 3 ]; then
echo "Wrong #arguments ($#, expected 3)"
echo "Usage: steps/decode_ctc.sh [options] <graph-dir> <data-dir> <decode-dir>"
echo " e.g.: steps/decode_ctc.sh data/lang data/test exp/train_l4_c320/decode"
echo "main options (for others, see top of script file)"
echo " --stage # starts from which stage"
echo " --nj <nj> # number of parallel jobs"
echo " --cmd <cmd> # command to run in parallel with"
echo " --acoustic_scales # default 0.5 0.6 0.7 0.8 ... the values of acoustic scales to be used"
exit 1;
fi
graphdir=$1
data=$2
dir=`echo $3 | sed 's:/$::g'` # remove any trailing slash.
srcdir=`dirname $dir`; # assume model directory one level up from decoding directory.
sdata=$data/split$nj;
thread_string=
[ $num_threads -gt 1 ] && thread_string="-parallel --num-threads=$num_threads"
[ -z "$add_deltas" ] && add_deltas=`cat $srcdir/add_deltas 2>/dev/null`
[ -z "$norm_vars" ] && norm_vars=`cat $srcdir/norm_vars 2>/dev/null`
mkdir -p $dir/log
split_data.sh $data $nj || exit 1;
echo $nj > $dir/num_jobs
# Check if necessary files exist.
for f in $graphdir/TLG.fst $srcdir/label.counts $data/feats.scp; do
[ ! -f $f ] && echo "$0: no such file $f" && exit 1;
done
## Set up the features
echo "$0: feature: norm_vars(${norm_vars}) add_deltas(${add_deltas})"
feats="ark,s,cs:apply-cmvn --norm-vars=$norm_vars --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- |"
$add_deltas && feats="$feats add-deltas ark:- ark:- |"
##
# Decode for each of the acoustic scales
for ascale in $acoustic_scales; do
echo "$0: decoding with acoustic scale $ascale"
$cmd JOB=1:$nj $dir/log/decode.$ascale.JOB.log \
net-output-extract --class-frame-counts=$srcdir/label.counts --apply-log=true $srcdir/final.nnet "$feats" ark:- \| \
decode-faster --beam=$beam --max-active=$max_active --acoustic-scale=$ascale --word-symbol-table=$graphdir/words.txt \
--allow-partial=true $graphdir/TLG.fst ark:- ark,t:$dir/trans.$ascale.JOB
cat $dir/trans.$ascale.* > $dir/trans.$ascale
rm -f $dir/trans.$ascale.*
done
# Scoring
cat $data/text | sed 's:<UNK>::g' | sed 's:<NOISE>::g' | sed 's:<SPOKEN_NOISE>::g' > $dir/text_filt
if ! $skip_scoring ; then
for ascale in $acoustic_scales; do
cat $dir/trans.$ascale | utils/int2sym.pl -f 2- $graphdir/words.txt | \
sed 's:<UNK>::g' | sed 's:<NOISE>::g' | sed 's:<SPOKEN_NOISE>::g' | \
compute-wer --text --mode=present ark:$dir/text_filt ark,p:- >& $dir/wer_$ascale || exit 1;
done
fi
exit 0;