make_denlats.sh
7.19 KB
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
#!/bin/bash
# Copyright 2012 Johns Hopkins University (Author: Daniel Povey). Apache 2.0.
# Create denominator lattices for MMI/MPE training.
# Creates its output in $dir/lat.*.gz
# Begin configuration section.
nj=4
cmd=run.pl
sub_split=1
beam=13.0
lattice_beam=7.0
acwt=0.1
max_active=5000
transform_dir=
max_mem=20000000 # This will stop the processes getting too large.
# This is in bytes, but not "real" bytes-- you have to multiply
# by something like 5 or 10 to get real bytes (not sure why so large)
num_threads=1
parallel_opts= # ignored now
# End configuration section.
echo "$0 $@" # Print the command line for logging
[ -f ./path.sh ] && . ./path.sh; # source the path.
. parse_options.sh || exit 1;
if [ $# != 4 ]; then
echo "Usage: steps/make_denlats.sh [options] <data-dir> <lang-dir> <src-dir> <exp-dir>"
echo " e.g.: steps/make_denlats.sh data/train data/lang exp/tri1 exp/tri1_denlats"
echo "Works for (delta|lda) features, and (with --transform-dir option) such features"
echo " plus transforms."
echo ""
echo "Main options (for others, see top of script file)"
echo " --config <config-file> # config containing options"
echo " --nj <nj> # number of parallel jobs"
echo " --cmd (utils/run.pl|utils/queue.pl <queue opts>) # how to run jobs."
echo " --sub-split <n-split> # e.g. 40; use this for "
echo " # large databases so your jobs will be smaller and"
echo " # will (individually) finish reasonably soon."
echo " --transform-dir <transform-dir> # directory to find fMLLR transforms."
echo " --num-threads <n> # number of threads per decoding job"
exit 1;
fi
data=$1
lang=$2
srcdir=$3
dir=$4
sdata=$data/split$nj
splice_opts=`cat $srcdir/splice_opts 2>/dev/null`
cmvn_opts=`cat $srcdir/cmvn_opts 2>/dev/null`
delta_opts=`cat $srcdir/delta_opts 2>/dev/null`
thread_string=
[ $num_threads -gt 1 ] && thread_string="-parallel --num-threads=$num_threads"
mkdir -p $dir/log
[[ -d $sdata && $data/feats.scp -ot $sdata ]] || split_data.sh $data $nj || exit 1;
echo $nj > $dir/num_jobs
utils/lang/check_phones_compatible.sh $lang/phones.txt $srcdir/phones.txt || exit 1;
cp $lang/phones.txt $dir || exit 1;
oov=`cat $lang/oov.int` || exit 1;
mkdir -p $dir
cp -RH $lang $dir/
# Compute grammar FST which corresponds to unigram decoding graph.
new_lang="$dir/"$(basename "$lang")
echo "Making unigram grammar FST in $new_lang"
cat $data/text | utils/sym2int.pl --map-oov $oov -f 2- $lang/words.txt | \
awk '{for(n=2;n<=NF;n++){ printf("%s ", $n); } printf("\n"); }' | \
utils/make_unigram_grammar.pl | fstcompile | fstarcsort --sort_type=ilabel > $new_lang/G.fst \
|| exit 1;
# mkgraph.sh expects a whole directory "lang", so put everything in one directory...
# it gets L_disambig.fst and G.fst (among other things) from $dir/lang, and
# final.mdl from $srcdir; the output HCLG.fst goes in $dir/graph.
echo "Compiling decoding graph in $dir/dengraph"
if [ -s $dir/dengraph/HCLG.fst ] && [ $dir/dengraph/HCLG.fst -nt $srcdir/final.mdl ]; then
echo "Graph $dir/dengraph/HCLG.fst already exists: skipping graph creation."
else
utils/mkgraph.sh $new_lang $srcdir $dir/dengraph || exit 1;
fi
if [ -f $srcdir/final.mat ]; then feat_type=lda; else feat_type=delta; fi
echo "$0: feature type is $feat_type"
case $feat_type in
delta) feats="ark,s,cs:apply-cmvn $cmvn_opts --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | add-deltas $delta_opts ark:- ark:- |";;
lda) feats="ark,s,cs:apply-cmvn $cmvn_opts --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | splice-feats $splice_opts ark:- ark:- | transform-feats $srcdir/final.mat ark:- ark:- |"
cp $srcdir/final.mat $dir
;;
*) echo "Invalid feature type $feat_type" && exit 1;
esac
if [ ! -z "$transform_dir" ]; then # add transforms to features...
echo "$0: using fMLLR transforms from $transform_dir"
[ ! -f $transform_dir/trans.1 ] && echo "Expected $transform_dir/trans.1 to exist." && exit 1;
[ "`cat $transform_dir/num_jobs`" -ne "$nj" ] \
&& echo "$0: mismatch in number of jobs with $transform_dir" && exit 1;
[ -f $srcdir/final.mat ] && ! cmp $transform_dir/final.mat $srcdir/final.mat && \
echo "$0: LDA transforms differ between $srcdir and $transform_dir"
feats="$feats transform-feats --utt2spk=ark:$sdata/JOB/utt2spk ark:$transform_dir/trans.JOB ark:- ark:- |"
else
if [ -f $srcdir/final.alimdl ]; then
echo "$0: you seem to have a SAT system but you did not supply the --transform-dir option.";
exit 1;
fi
fi
# if this job is interrupted by the user, we want any background jobs to be
# killed too.
cleanup() {
local pids=$(jobs -pr)
[ -n "$pids" ] && kill $pids
}
trap "cleanup" INT QUIT TERM EXIT
if [ $sub_split -eq 1 ]; then
$cmd --num-threads $num_threads JOB=1:$nj $dir/log/decode_den.JOB.log \
gmm-latgen-faster$thread_string --beam=$beam --lattice-beam=$lattice_beam --acoustic-scale=$acwt \
--max-mem=$max_mem --max-active=$max_active --word-symbol-table=$lang/words.txt $srcdir/final.mdl \
$dir/dengraph/HCLG.fst "$feats" "ark:|gzip -c >$dir/lat.JOB.gz" || exit 1;
else
# each job from 1 to $nj is split into multiple pieces (sub-split), and we aim
# to have at most two jobs running at each time. The idea is that if we have stragglers
# from one job, we can be processing another one at the same time.
rm $dir/.error 2>/dev/null
prev_pid=
for n in `seq $[nj+1]`; do
if [ $n -gt $nj ]; then
this_pid=
elif [ -f $dir/.done.$n ] && [ $dir/.done.$n -nt $srcdir/final.mdl ]; then
echo "Not processing subset $n as already done (delete $dir/.done.$n if not)";
this_pid=
else
sdata2=$data/split$nj/$n/split${sub_split}utt;
split_data.sh --per-utt $sdata/$n $sub_split || exit 1;
mkdir -p $dir/log/$n
mkdir -p $dir/part
feats_subset=`echo $feats | sed "s/trans.JOB/trans.$n/g" | sed s:JOB/:$n/split${sub_split}utt/JOB/:g`
$cmd --num-threads $num_threads JOB=1:$sub_split $dir/log/$n/decode_den.JOB.log \
gmm-latgen-faster$thread_string --beam=$beam --lattice-beam=$lattice_beam --acoustic-scale=$acwt \
--max-mem=$max_mem --max-active=$max_active --word-symbol-table=$lang/words.txt $srcdir/final.mdl \
$dir/dengraph/HCLG.fst "$feats_subset" "ark:|gzip -c >$dir/lat.$n.JOB.gz" || touch $dir/.error &
this_pid=$!
fi
if [ ! -z "$prev_pid" ]; then # Wait for the previous job; merge the previous set of lattices.
wait $prev_pid
[ -f $dir/.error ] && echo "$0: error generating denominator lattices" && exit 1;
rm $dir/.merge_error 2>/dev/null
echo Merging archives for data subset $prev_n
for k in `seq $sub_split`; do
gunzip -c $dir/lat.$prev_n.$k.gz || touch $dir/.merge_error;
done | gzip -c > $dir/lat.$prev_n.gz || touch $dir/.merge_error;
[ -f $dir/.merge_error ] && echo "$0: Merging lattices for subset $prev_n failed (or maybe some other error)" && exit 1;
rm $dir/lat.$prev_n.*.gz
touch $dir/.done.$prev_n
fi
prev_n=$n
prev_pid=$this_pid
done
fi
echo "$0: done generating denominator lattices."