File size: 6,137 Bytes
61ff961
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
#!/bin/bash
#
# Recipe for the Dsing baseline
# Based mostly on the Librispeech recipe
#
# Copyright 2019  Gerardo Roa
#                 University of Sheffield
# Apache 2.0

# Begin configuration section

nj=40
stage=0
dsing=1   #  Set:  1  for DSing1
          #        3  for DSing3
          #        30 for DSing30

# For TDNN-F only
decode_nj=1

# End configuration section
. ./utils/parse_options.sh

. ./path.sh
. ./cmd.sh

set -e # exit on error

# Sing! 300x30x2 corpus path
# please change the path accordingly
sing_corpus=/fastdata/acp13gr/DAMP300x30x20/sing_300x30x2


echo "Using steps and utils from WSJ recipe"
[[ ! -L "wav" ]] && ln -s $sing_corpus wav
[[ ! -L "steps" ]] && ln -s $KALDI_ROOT/egs/wsj/s5/steps
[[ ! -L "utils" ]] && ln -s $KALDI_ROOT/egs/wsj/s5/utils


trainset=train${dsing}
devset="dev"
testset="test"

# This script also needs the phonetisaurus g2p, srilm, sox
./local/check_tools.sh || exit 1


echo; echo "===== Starting at  $(date +"%D_%T") ====="; echo


if [ $stage -le 1 ]; then
    mkdir -p data/local/dict
    cp conf/corpus.txt  data/local/corpus.txt  # Corpus.txt for language model

    for datadir in $devset $testset $trainset; do
        python local/prepare_data.py data/ wav/ conf/${datadir}.json $datadir
    done

    # Selecting the top 25000 words by frequency
    # Is expected that the final size will be larger as
    # it use all the words with the same frequency avoiding an arbitrary cut-off
    local/prepare_dict.sh --words 26000

    utils/prepare_lang.sh data/local/dict "<UNK>" data/local/lang data/lang

    local/train_lms_srilm.sh \
        --train-text data/local/corpus.txt \
        --dev_text data/dev/text \
        --oov-symbol "<UNK>" --words-file data/lang/words.txt \
        data/ data/srilm

    # Compiles G for DSing Preconstructed trigram LM
    utils/format_lm.sh  data/lang data/srilm/best_3gram.gz data/local/dict/lexicon.txt data/lang_3G
    utils/format_lm.sh  data/lang data/srilm/best_4gram.gz data/local/dict/lexicon.txt data/lang_4G
fi

# Features Extraction
if [[ $stage -le 2 ]]; then

  echo
  echo "============================="
  echo "---- MFCC FEATURES EXTRACTION ----"
  echo "=====  $(date +"%D_%T") ====="

  for datadir in $trainset $devset $testset; do
    echo; echo "---- $datadir"
    utils/fix_data_dir.sh data/$datadir
    steps/make_mfcc.sh --cmd "$train_cmd" --nj $nj data/${datadir} exp/make_mfcc/${datadir} mfcc
    steps/compute_cmvn_stats.sh data/${datadir}
    utils/fix_data_dir.sh data/$datadir
  done
fi

if [[ $stage -le 3 ]]; then

    echo
    echo "============================="
    echo "-------- Train GMM ----------"
    echo

    echo
    echo "Mono"
    echo "=====  $(date +"%D_%T") ====="

    # Monophone
    steps/train_mono.sh --nj $nj --cmd "$train_cmd"  \
          data/${trainset} data/lang exp/mono

    steps/align_si.sh --nj $nj --cmd "$train_cmd" \
      data/${trainset} data/lang exp/mono exp/mono_ali
    
    echo
    echo "Tri 1 - delta-based triphones"
    echo "=====  $(date +"%D_%T") ====="
    
    # Tri1      


    steps/train_deltas.sh  --cmd "$train_cmd" 2000 15000 \
      data/${trainset} data/lang exp/mono_ali exp/tri1

    steps/align_si.sh --nj $nj --cmd "$train_cmd"  \
     data/${trainset} data/lang exp/tri1 exp/tri1_ali
      
    echo
    echo "Tri 2 - LDA-MLLT triphones"
    echo "=====  $(date +"%D_%T") ====="
    # Tri2  


    steps/train_lda_mllt.sh --cmd "$train_cmd" 2500 20000 \
      data/${trainset} data/lang exp/tri1_ali exp/tri2b

    steps/align_si.sh --nj $nj --cmd "$train_cmd"  \
      data/${trainset} data/lang exp/tri2b exp/tri2b_ali
      
    echo
    echo "Tri 3 - SAT triphones"
    echo "=====  $(date +"%D_%T") ====="
   
    # Tri3 SAT


    steps/train_sat.sh --cmd "$train_cmd" 3000 25000 \
      data/${trainset} data/lang exp/tri2b_ali exp/tri3b

    utils/mkgraph.sh data/lang_3G exp/tri3b exp/tri3b/graph
   
    echo
    echo "------ End Train GMM --------"
    echo "=====  $(date +"%D_%T") ====="
fi


if [[ $stage -le 4 ]]; then
    echo
    echo "============================="
    echo "------- Decode TRI3B --------"
    echo "=====  $(date +"%D_%T") ====="
    echo


    
    echo; echo "--------decode ${devset}"; echo
    steps/decode_fmllr.sh --config conf/decode.config --nj $nj --cmd "$decode_cmd" \
      --scoring-opts "--min-lmwt 10 --max-lmwt 20" --num-threads 4  \
      exp/tri3b/graph data/${devset} exp/tri3b/decode_${devset}


    # Scoring test model with the best
    lmwt=$(cat exp/tri3b/decode_${devset}/scoring_kaldi/wer_details/lmwt)
    wip=$(cat exp/tri3b/decode_${devset}/scoring_kaldi/wer_details/wip)
    echo; echo "--------decode ${testset}"
    echo "Using [lmwt=$lmwt, wip=$wip] to score"; echo
    steps/decode_fmllr.sh --config conf/decode.config --nj $nj --cmd "$decode_cmd" \
      --scoring-opts "--min_lmwt $lmwt --max_lmwt $lmwt --word_ins_penalty $wip" --num-threads 4  \
      exp/tri3b/graph data/${testset} exp/tri3b/decode_${testset}


fi

# Produce clean data
if [[ $stage -le 5 ]]; then
    echo
    echo "============================="
    echo "------- Cleanup Tri3b -------"
    echo "=====  $(date +"%D_%T") ====="
    echo

    steps/cleanup/clean_and_segment_data.sh --nj $nj --cmd "$train_cmd"  \
    --segmentation-opts "--min-segment-length 0.3 --min-new-segment-length 0.6" \
      data/${trainset} data/lang exp/tri3b exp/tri3b_cleaned \
      data/${trainset}_cleaned
fi

if [[ $stage -le 6 ]]; then
    echo
    echo "=================="
    echo "----- TDNN-F -----"
    echo "=====  $(date +"%D_%T") ====="
    echo

    local/chain/run_tdnn_1d.sh --nj $nj --decode_nj $decode_nj \
    --train_set ${trainset}_cleaned --test_sets "$devset $testset" \
    --gmm tri3b_cleaned --nnet3-affix _${trainset}_cleaned
fi


if [[ $stage -le 7 ]]; then
    echo
    echo "============================="
    echo "------- FINAL SCORES --------"
    echo "=====  $(date +"%D_%T") ====="
    echo

    for x in `find exp/* -name "best_wer"`; do
        cat $x | grep -v ".si"
    done
fi

echo
echo "=====  $(date +"%D_%T") ====="
echo "===== PROCESS ENDED ====="
echo

exit 1