Blame view

egs/gp/s1/utils/mkgraph.sh 4.44 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
  #!/bin/bash
  # Copyright 2010-2011 Microsoft Corporation
  
  # Licensed under the Apache License, Version 2.0 (the "License");
  # you may not use this file except in compliance with the License.
  # You may obtain a copy of the License at
  #
  #  http://www.apache.org/licenses/LICENSE-2.0
  #
  # THIS CODE IS PROVIDED *AS IS* BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
  # KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED
  # WARRANTIES OR CONDITIONS OF TITLE, FITNESS FOR A PARTICULAR PURPOSE,
  # MERCHANTABLITY OR NON-INFRINGEMENT.
  # See the Apache 2 License for the specific language governing permissions and
  # limitations under the License.
  
  
  # This script creates a fully expanded decoding graph (HCLG) that represents
  # all the language-model, pronunciation dictionary (lexicon), context-dependency,
  # and HMM structure in our model.  The output is a Finite State Transducer
  # that has word-ids on the output, and pdf-ids on the input (these are indexes
  # that resolve to Gaussian Mixture Models).  
  # See
  #  http://kaldi-asr.org/doc/graph_recipe_test.html
  # (this is compiled from this repository using Doxygen,
  # the source for this part is in src/doc/graph_recipe_test.dox)
  
  
  N=3
  P=1
  clean=false
  
  for x in 1 2 3; do 
    if [ $1 == "--mono" ]; then
      N=1;
      P=0;
      shift;
    fi
    if [ $1 == "--clean" ]; then
      clean=true
      shift;
    fi
  
  done
  
  if [ $# != 3 ]; then
     echo "Usage: scripts/mkgraph.sh <test-lang-dir> <model-dir> <graphdir>"
     echo "e.g.: scripts/mkgraph.sh data/lang_test exp/tri1/ exp/tri1/graph"
     exit 1;
  fi
  
  if [ -f path.sh ]; then . ./path.sh; fi
  
  lang=$1
  tree=$2/tree
  model=$2/final.mdl
  dir=$3
  
  if $clean; then rm -r $lang/tmp; fi
  
  mkdir -p $dir
  
  tscale=1.0
  loopscale=0.1
  
  # If $lang/tmp/LG.fst does not exist or is older than its sources, make it...
  # (note: the [[ ]] brackets make the || type operators work (inside [ ], we
  # would have to use -o instead),  -f means file exists, and -ot means older than).
  
  required="$lang/L.fst $lang/G.fst $lang/phones_disambig.txt $lang/words.txt $lang/silphones.csl $model $tree"
  for f in $required; do
    [ ! -f $f ] && echo "mkgraph.sh: expected $f to exist" && exit 1;
  done
  
  mkdir -p $lang/tmp
  if [[ ! -f $lang/tmp/LG.fst || $lang/tmp/LG.fst -ot $lang/G.fst || \
        $lang/tmp/LG.fst -ot $lang/L_disambig.fst ]]; then
    fsttablecompose $lang/L_disambig.fst $lang/G.fst | fstdeterminizestar --use-log=true | \
      fstminimizeencoded  > $lang/tmp/LG.fst || exit 1;
    fstisstochastic $lang/tmp/LG.fst || echo "warning: LG not stochastic."
  fi
  
  if [ ! -f $lang/phones_disambig.txt ]; then
    echo "No such file $lang/phones_disambig.txt (supplied a training lang/ directory?)"
    exit 1;
  fi
  
  grep '#' $lang/phones_disambig.txt | awk '{print $2}' > $lang/tmp/disambig_phones.list
  
  
  clg=$lang/tmp/CLG_${N}_${P}.fst
  
  if [[ ! -f $clg || $clg -ot $lang/tmp/LG.fst ]]; then
    fstcomposecontext --context-size=$N --central-position=$P \
     --read-disambig-syms=$lang/tmp/disambig_phones.list \
     --write-disambig-syms=$lang/tmp/disambig_ilabels_${N}_${P}.list \
      $lang/tmp/ilabels_${N}_${P} < $lang/tmp/LG.fst >$clg
    fstisstochastic $clg  || echo "warning: CLG not stochastic."
  fi
  
  if [[ ! -f $dir/Ha.fst || $dir/Ha.fst -ot $model  \
      || $dir/Ha.fst -ot $lang/tmp/ilabels_${N}_${P} ]]; then
    make-h-transducer --disambig-syms-out=$dir/disambig_tid.list \
      --transition-scale=$tscale $lang/tmp/ilabels_${N}_${P} $tree $model \
       > $dir/Ha.fst  || exit 1;
  fi
  
  if [[ ! -f $dir/HCLGa.fst || $dir/HCLGa.fst -ot $dir/Ha.fst || \
        $dir/HCLGa.fst -ot $clg ]]; then
    fsttablecompose $dir/Ha.fst $clg | fstdeterminizestar --use-log=true \
      | fstrmsymbols $dir/disambig_tid.list | fstrmepslocal | \
       fstminimizeencoded > $dir/HCLGa.fst || exit 1;
    fstisstochastic $dir/HCLGa.fst || echo "HCLGa is not stochastic"
  fi
  
  if [[ ! -f $dir/HCLG.fst || $dir/HCLG.fst -ot $dir/HCLGa.fst ]]; then
    add-self-loops --self-loop-scale=$loopscale --reorder=true \
      $model < $dir/HCLGa.fst > $dir/HCLG.fst || exit 1;
  
    if [ $tscale == 1.0 -a $loopscale == 1.0 ]; then
      # No point doing this test if transition-scale not 1, as it is bound to fail. 
      fstisstochastic $dir/HCLG.fst || echo "Final HCLG is not stochastic."
    fi
  fi
  
  # keep a copy of the lexicon and a list of silence phones with HCLG...
  # this means we can decode without refrence to the $lang directory.
  cp $lang/words.txt $dir/
  cp $lang/silphones.csl $dir/
  
  # to make const fst:
  # fstconvert --fst_type=const $dir/HCLG.fst $dir/HCLG_c.fst
  
  echo "Finished making decoding graphs in $dir"