Blame view

egs/sprakbanken/s5/local/nnet3/run_tdnn.sh 3.37 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
  #!/bin/bash
  
  #    This is the standard "tdnn" system, built in nnet3
  
  # by default:
  # local/nnet3/run_tdnn.sh
  
  set -e -o pipefail -u
  
  # First the options that are passed through to run_ivector_common.sh
  # (some of which are also used in this script directly).
  stage=0
  nj=30
  decode_nj=7
  min_seg_len=1.55
  train_set=train
  gmm=tri3b  # this is the source gmm-dir for the data-type of interest; it
                    # should have alignments for the specified training data.
  num_threads_ubm=32
  nnet3_affix=  # cleanup affix for exp dirs, e.g. _cleaned
  tdnn_affix=  #affix for TDNN directory e.g. "a" or "b", in case we change the configuration.
  
  # Options which are not passed through to run_ivector_common.sh
  train_stage=-10
  splice_indexes="-2,-1,0,1,2 -1,2 -3,3 -7,2 -3,3 0 0"
  remove_egs=true
  relu_dim=750
  num_epochs=3
  
  . ./cmd.sh
  . ./path.sh
  . ./utils/parse_options.sh
  
  if ! cuda-compiled; then
    cat <<EOF && exit 1
  This script is intended to be used with GPUs but you have not compiled Kaldi with CUDA
  If you want to use GPUs (and have them), go to src/, and configure and make on a machine
  where "nvcc" is installed.
  EOF
  fi
  
  local/nnet3/run_ivector_common.sh --stage $stage \
                                    --nj $nj \
                                    --min-seg-len $min_seg_len \
                                    --train-set $train_set \
                                    --gmm $gmm \
                                    --num-threads-ubm $num_threads_ubm \
                                    --nnet3-affix "$nnet3_affix"
  
  
  # Some of these destinations are created by run_ivector_common.sh
  
  gmm_dir=exp/${gmm}
  graph_dir=$gmm_dir/graph_tg
  ali_dir=exp/${gmm}_ali_${train_set}_sp_comb
  dir=exp/nnet3${nnet3_affix}/tdnn${tdnn_affix}_sp
  train_data_dir=data/${train_set}_sp_hires_comb
  train_ivector_dir=exp/nnet3${nnet3_affix}/ivectors_${train_set}_sp_hires_comb
  
  
  for f in $train_data_dir/feats.scp $train_ivector_dir/ivector_online.scp \
       $graph_dir/HCLG.fst $ali_dir/ali.1.gz $gmm_dir/final.mdl; do
    [ ! -f $f ] && echo "$0: expected file $f to exist" && exit 1
  done
  
  
  if [ $stage -le 12 ]; then
    if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
      utils/create_split_dir.pl \
       /export/b0{3,5,6}/$USER/kaldi-data/egs/sprakbanken-$(date +'%m_%d_%H_%M')/s5/$dir/egs/storage $dir/egs/storage
    fi
  
    steps/nnet3/tdnn/train.sh --stage $train_stage \
      --num-epochs $num_epochs --num-jobs-initial 2 --num-jobs-final 12 \
      --splice-indexes "$splice_indexes" \
      --feat-type raw \
      --online-ivector-dir ${train_ivector_dir} \
      --cmvn-opts "--norm-means=false --norm-vars=false" \
      --initial-effective-lrate 0.0015 --final-effective-lrate 0.00015 \
      --cmd "$decode_cmd" \
      --relu-dim "$relu_dim" \
      --remove-egs "$remove_egs" \
      $train_data_dir data/lang $ali_dir $dir
  fi
  
  if [ $stage -le 13 ]; then
    rm $dir/.error || true 2>/dev/null
     (
      steps/nnet3/decode.sh --nj 7 --cmd "$decode_cmd"  --num-threads 4 \
          --online-ivector-dir exp/nnet3${nnet3_affix}/ivectors_test_hires \
          ${graph_dir} data/test_hires ${dir}/decode_test || exit 1
      steps/nnet3/decode.sh --nj 12 --cmd "$decode_cmd"  --num-threads 4 \
          --online-ivector-dir exp/nnet3${nnet3_affix}/ivectors_dev_hires \
        ${graph_dir} data/dev_hires ${dir}/decode_dev || exit 1
     ) || touch $dir/.error &
  
    wait
    [ -f $dir/.error ] && echo "$0: there was a problem while decoding" && exit 1
  fi
  
  
  exit 0;