Kaldi / Kaldi first steps

Blame view

Scripts/steps/train_smbr.sh 6.15 KB
edit raw normal view history
ec85f8892   bigot benjamin   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
 
  #!/bin/bash
  # Copyright 2012  Johns Hopkins University (Author: Daniel Povey).  Apache 2.0.
  
  # sMBR training 
  # 4 iterations (by default) of Extended Baum-Welch update.
  #
  # For the numerator we have a fixed alignment rather than a lattice--
  # this actually follows from the way lattices are defined in Kaldi, which
  # is to have a single path for each word (output-symbol) sequence.
  
  # Begin configuration section.
  cmd=run.pl
  num_iters=4
  cancel=true # if true, cancel num and den counts on each frame.
  tau=400
  weight_tau=10
  acwt=0.1
  stage=0
  smooth_to_mode=true
  # 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 [ $# -ne 5 ]; then
    echo "Usage: steps/train_smbr.sh <data> <lang> <ali> <denlats> <exp>"
    echo " e.g.: steps/train_smbr.sh data/train_si84 data/lang exp/tri2b_ali_si84 exp/tri2b_denlats_si84 exp/tri2b_smbr"
    echo "Main options (for others, see top of script file)"
    echo "  --cancel (true|false)                            # cancel stats (true by default)"
    echo "  --cmd (utils/run.pl|utils/queue.pl <queue opts>) # how to run jobs."
    echo "  --config <config-file>                           # config containing options"
    echo "  --stage <stage>                                  # stage to do partial re-run from."
    echo "  --tau                                            # tau for i-smooth to last iter (default 200)"
    
    exit 1;
  fi
  
  data=$1
  lang=$2
  alidir=$3
  denlatdir=$4
  dir=$5
  mkdir -p $dir/log
  
  for f in $data/feats.scp $alidir/{tree,final.mdl,ali.1.gz} $denlatdir/lat.1.gz; do
    [ ! -f $f ] && echo "$0: no such file $f" && exit 1;
  done
  nj=`cat $alidir/num_jobs` || exit 1;
  [ "$nj" -ne "`cat $denlatdir/num_jobs`" ] && \
    echo "$alidir and $denlatdir have different num-jobs" && exit 1;
  
  sdata=$data/split$nj
  splice_opts=`cat $alidir/splice_opts 2>/dev/null`
  mkdir -p $dir/log
  cp $alidir/splice_opts $dir 2>/dev/null
  [[ -d $sdata && $data/feats.scp -ot $sdata ]] || split_data.sh $data $nj || exit 1;
  echo $nj > $dir/num_jobs
  
  cp $alidir/{final.mdl,tree} $dir
  
  silphonelist=`cat $lang/phones/silence.csl` || exit 1;
  
  # Set up features
  
  if [ -f $alidir/final.mat ]; then feat_type=lda; else feat_type=delta; fi
  echo "$0: feature type is $feat_type"
  
  case $feat_type in
    delta) feats="ark,s,cs:apply-cmvn --norm-vars=false --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | add-deltas ark:- ark:- |";;
    lda) feats="ark,s,cs:apply-cmvn --norm-vars=false --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | splice-feats $splice_opts ark:- ark:- | transform-feats $alidir/final.mat ark:- ark:- |"
      cp $alidir/final.mat $dir    
      ;;
    *) echo "Invalid feature type $feat_type" && exit 1;
  esac
  
  [ -f $alidir/trans.1 ] && echo Using transforms from $alidir && \
    feats="$feats transform-feats --utt2spk=ark:$sdata/JOB/utt2spk ark,s,cs:$alidir/trans.JOB ark:- ark:- |"
  
  lats="ark:gunzip -c $denlatdir/lat.JOB.gz|"
  
  cur_mdl=$alidir/final.mdl
  x=0
  while [ $x -lt $num_iters ]; do
    echo "Iteration $x of sMBR training"
    # Note: the num and den states are accumulated at the same time, so we
    # can cancel them per frame.
    if [ $stage -le $x ]; then
      $cmd JOB=1:$nj $dir/log/acc.$x.JOB.log \
        gmm-rescore-lattice $cur_mdl "$lats" "$feats" ark:- \| \
        lattice-to-smbr-post --acoustic-scale=$acwt $cur_mdl \
          "ark,s,cs:gunzip -c $alidir/ali.JOB.gz | ali-to-post ark:- ark:- |" ark:- ark:- \| \
        gmm-acc-stats2 $cur_mdl "$feats" ark,s,cs:- \
          $dir/num_acc.$x.JOB.acc $dir/den_acc.$x.JOB.acc || exit 1;
  
      n=`echo $dir/{num,den}_acc.$x.*.acc | wc -w`;
      [ "$n" -ne $[$nj*2] ] && \
        echo "Wrong number of sMBR accumulators $n versus 2*$nj" && exit 1;
      $cmd $dir/log/den_acc_sum.$x.log \
        gmm-sum-accs $dir/den_acc.$x.acc $dir/den_acc.$x.*.acc || exit 1;
      rm $dir/den_acc.$x.*.acc
      $cmd $dir/log/num_acc_sum.$x.log \
        gmm-sum-accs $dir/num_acc.$x.acc $dir/num_acc.$x.*.acc || exit 1;
      rm $dir/num_acc.$x.*.acc
  
    # note: this tau value is for smoothing towards model parameters, not
    # as in the Boosted MMI paper, not towards the ML stats as in the earlier
    # work on discriminative training (e.g. my thesis).  
    # You could use gmm-ismooth-stats to smooth to the ML stats, if you had
    # them available [here they're not available if cancel=true].
      if ! $smooth_to_model; then
        echo "Iteration $x of sMBR: computing ml (smoothing) stats"
        $cmd JOB=1:$nj $dir/log/acc_ml.$x.JOB.log \
          gmm-acc-stats $cur_mdl "$feats" \
            "ark,s,cs:gunzip -c $alidir/ali.JOB.gz | ali-to-post ark:- ark:- |" \
            $dir/ml.$x.JOB.acc || exit 1;
        $cmd $dir/log/acc_ml_sum.$x.log \
          gmm-sum-accs $dir/ml.$x.acc $dir/ml.$x.*.acc || exit 1;
        rm $dir/ml.$x.*.acc
        num_stats="gmm-ismooth-stats --tau=$tau $dir/ml.$x.acc $dir/num_acc.$x.acc -|"
      else 
        num_stats="gmm-ismooth-stats --smooth-from-model=true --tau=$tau $cur_mdl $dir/num_acc.$x.acc -|"
      fi  
      
      $cmd $dir/log/update.$x.log \
        gmm-est-gaussians-ebw $cur_mdl "$num_stats" $dir/den_acc.$x.acc - \| \
        gmm-est-weights-ebw - $dir/num_acc.$x.acc $dir/den_acc.$x.acc $dir/$[$x+1].mdl || exit 1;
      rm $dir/{den,num}_acc.$x.acc
    fi
    cur_mdl=$dir/$[$x+1].mdl
  
    # Some diagnostics: the objective function progress and auxiliary-function
    # improvement.
  
   tail -n 50 $dir/log/acc.$x.*.log | perl -e 'while(<STDIN>) { if(m/lattice-to-smbr-post.+Overall average frame-accuracy is (\S+) over (\S+) frames/) { $tot_objf += $1*$2; $tot_frames += $2; }} $tot_objf /= $tot_frames; print "$tot_objf $tot_frames
  "; ' > $dir/tmpf
    objf=`cat $dir/tmpf | awk '{print $1}'`;
    nf=`cat $dir/tmpf | awk '{print $2}'`;
    rm $dir/tmpf
    impr=`grep -w Overall $dir/log/update.$x.log | awk '{x += $10*$12;} END{print x;}'`
    impr=`perl -e "print ($impr*$acwt/$nf);"` # We multiply by acwt, and divide by $nf which is the "real" number of frames.
    # This gives us a projected objective function improvement.
    echo "Iteration $x: objf was $objf, sMBR auxf change was $impr" | tee $dir/objf.$x.log
    x=$[$x+1]
  done
  
  echo "sMBR training finished"
  
  rm $dir/final.mdl 2>/dev/null
  ln -s $x.mdl $dir/final.mdl
  
  exit 0;