Blame view

egs/wsj/s5/steps/train_mono.sh 5.8 KB
8dcb6dfcb   Yannick Estève   first commit
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 2012  Johns Hopkins University (Author: Daniel Povey)
  #           2019  Xiaohui Zhang
  # Apache 2.0
  
  
  # To be run from ..
  # Flat start and monophone training, with delta-delta features.
  # This script applies cepstral mean normalization (per speaker).
  
  # Begin configuration section.
  nj=4
  cmd=run.pl
  scale_opts="--transition-scale=1.0 --acoustic-scale=0.1 --self-loop-scale=0.1"
  num_iters=40    # Number of iterations of training
  max_iter_inc=30 # Last iter to increase #Gauss on.
  initial_beam=6 # beam used in the first iteration (set smaller to speed up initialization)
  regular_beam=10 # beam used after the first iteration
  retry_beam=40
  totgauss=1000 # Target #Gaussians.
  careful=false
  boost_silence=1.0 # Factor by which to boost silence likelihoods in alignment
  realign_iters="1 2 3 4 5 6 7 8 9 10 12 14 16 18 20 23 26 29 32 35 38";
  config= # name of config file.
  stage=-4
  power=0.25 # exponent to determine number of gaussians from occurrence counts
  norm_vars=false # deprecated, prefer --cmvn-opts "--norm-vars=false"
  cmvn_opts=  # can be used to add extra options to cmvn.
  delta_opts= # can be used to add extra options to add-deltas
  # End configuration section.
  
  echo "$0 $@"  # Print the command line for logging
  
  if [ -f path.sh ]; then . ./path.sh; fi
  . parse_options.sh || exit 1;
  
  if [ $# != 3 ]; then
    echo "Usage: steps/train_mono.sh [options] <data-dir> <lang-dir> <exp-dir>"
    echo " e.g.: steps/train_mono.sh data/train.1k data/lang exp/mono"
    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."
    exit 1;
  fi
  
  data=$1
  lang=$2
  dir=$3
  
  oov_sym=`cat $lang/oov.int` || exit 1;
  
  mkdir -p $dir/log
  echo $nj > $dir/num_jobs
  sdata=$data/split$nj;
  [[ -d $sdata && $data/feats.scp -ot $sdata ]] || split_data.sh $data $nj || exit 1;
  
  cp $lang/phones.txt $dir || exit 1;
  
  $norm_vars && cmvn_opts="--norm-vars=true $cmvn_opts"
  echo $cmvn_opts  > $dir/cmvn_opts # keep track of options to CMVN.
  [ ! -z $delta_opts ] && echo $delta_opts > $dir/delta_opts # keep track of options to 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:- |"
  example_feats="`echo $feats | sed s/JOB/1/g`";
  
  echo "$0: Initializing monophone system."
  
  [ ! -f $lang/phones/sets.int ] && exit 1;
  shared_phones_opt="--shared-phones=$lang/phones/sets.int"
  
  if [ $stage -le -3 ]; then
    # Note: JOB=1 just uses the 1st part of the features-- we only need a subset anyway.
    if ! feat_dim=`feat-to-dim "$example_feats" - 2>/dev/null` || [ -z $feat_dim ]; then
      feat-to-dim "$example_feats" -
      echo "error getting feature dimension"
      exit 1;
    fi
    $cmd JOB=1 $dir/log/init.log \
      gmm-init-mono $shared_phones_opt "--train-feats=$feats subset-feats --n=10 ark:- ark:-|" $lang/topo $feat_dim \
      $dir/0.mdl $dir/tree || exit 1;
  fi
  
  numgauss=`gmm-info --print-args=false $dir/0.mdl | grep gaussians | awk '{print $NF}'`
  incgauss=$[($totgauss-$numgauss)/$max_iter_inc] # per-iter increment for #Gauss
  
  if [ $stage -le -2 ]; then
    echo "$0: Compiling training graphs"
    $cmd JOB=1:$nj $dir/log/compile_graphs.JOB.log \
      compile-train-graphs --read-disambig-syms=$lang/phones/disambig.int $dir/tree $dir/0.mdl  $lang/L.fst \
      "ark:sym2int.pl --map-oov $oov_sym -f 2- $lang/words.txt < $sdata/JOB/text|" \
      "ark:|gzip -c >$dir/fsts.JOB.gz" || exit 1;
  fi
  
  if [ $stage -le -1 ]; then
    echo "$0: Aligning data equally (pass 0)"
    $cmd JOB=1:$nj $dir/log/align.0.JOB.log \
      align-equal-compiled "ark:gunzip -c $dir/fsts.JOB.gz|" "$feats" ark,t:-  \| \
      gmm-acc-stats-ali --binary=true $dir/0.mdl "$feats" ark:- \
      $dir/0.JOB.acc || exit 1;
  fi
  
  # In the following steps, the --min-gaussian-occupancy=3 option is important, otherwise
  # we fail to est "rare" phones and later on, they never align properly.
  
  if [ $stage -le 0 ]; then
    gmm-est --min-gaussian-occupancy=3  --mix-up=$numgauss --power=$power \
      $dir/0.mdl "gmm-sum-accs - $dir/0.*.acc|" $dir/1.mdl 2> $dir/log/update.0.log || exit 1;
    rm $dir/0.*.acc
  fi
  
  beam=$initial_beam # will change to regular_beam below after 1st pass
  # note: using slightly wider beams for WSJ vs. RM.
  x=1
  while [ $x -lt $num_iters ]; do
    echo "$0: Pass $x"
    if [ $stage -le $x ]; then
      if echo $realign_iters | grep -w $x >/dev/null; then
        echo "$0: Aligning data"
        mdl="gmm-boost-silence --boost=$boost_silence `cat $lang/phones/optional_silence.csl` $dir/$x.mdl - |"
        $cmd JOB=1:$nj $dir/log/align.$x.JOB.log \
          gmm-align-compiled $scale_opts --beam=$beam --retry-beam=$retry_beam --careful=$careful "$mdl" \
          "ark:gunzip -c $dir/fsts.JOB.gz|" "$feats" "ark,t:|gzip -c >$dir/ali.JOB.gz" \
          || exit 1;
      fi
      $cmd JOB=1:$nj $dir/log/acc.$x.JOB.log \
        gmm-acc-stats-ali  $dir/$x.mdl "$feats" "ark:gunzip -c $dir/ali.JOB.gz|" \
        $dir/$x.JOB.acc || exit 1;
  
      $cmd $dir/log/update.$x.log \
        gmm-est --write-occs=$dir/$[$x+1].occs --mix-up=$numgauss --power=$power $dir/$x.mdl \
        "gmm-sum-accs - $dir/$x.*.acc|" $dir/$[$x+1].mdl || exit 1;
      rm $dir/$x.mdl $dir/$x.*.acc $dir/$x.occs 2>/dev/null
    fi
    if [ $x -le $max_iter_inc ]; then
       numgauss=$[$numgauss+$incgauss];
    fi
    beam=$regular_beam
    x=$[$x+1]
  done
  
  ( cd $dir; rm final.{mdl,occs} 2>/dev/null; ln -s $x.mdl final.mdl; ln -s $x.occs final.occs )
  
  
  steps/diagnostic/analyze_alignments.sh --cmd "$cmd" $lang $dir
  utils/summarize_warnings.pl $dir/log
  
  steps/info/gmm_dir_info.pl $dir
  
  echo "$0: Done training monophone system in $dir"
  
  exit 0
  
  # example of showing the alignments:
  # show-alignments data/lang/phones.txt $dir/30.mdl "ark:gunzip -c $dir/ali.0.gz|" | head -4