run_tdnn.sh 5.12 KB
#!/bin/bash

# this is a script to train the nnet3 TDNN acoustic model


stage=1
affix=
train_stage=-10
reporting_email=
common_egs_dir=
remove_egs=true
egs_stage=0

. ./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.  Otherwise, call this script with --use-gpu false
EOF
fi

# do the common parts of the script.
local/nnet3/run_ivector_common.sh --stage $stage || exit 1;

ali_dir=exp/tri5a_rvb_ali
dir=exp/nnet3/tdnn
dir=$dir${affix:+_$affix}

if [ $stage -le 7 ]; then
  echo "$0: creating neural net configs using the xconfig parser";
  num_targets=$(tree-info $ali_dir/tree | grep num-pdfs | awk '{print $2}')

  mkdir -p $dir/configs
  cat <<EOF > $dir/configs/network.xconfig
  input dim=100 name=ivector
  input dim=40 name=input

  # please note that it is important to have input layer with the name=input
  # as the layer immediately preceding the fixed-affine-layer to enable
  # the use of short notation for the descriptor
  fixed-affine-layer name=lda input=Append(-2,-1,0,1,2,ReplaceIndex(ivector, t, 0)) affine-transform-file=$dir/configs/lda.mat

  # the first splicing is moved before the lda layer, so no splicing here
  relu-renorm-layer name=tdnn1 dim=1248
  relu-renorm-layer name=tdnn2 dim=1248 input=Append(-1,2)
  relu-renorm-layer name=tdnn3 dim=1248 input=Append(-3,3)
  relu-renorm-layer name=tdnn4 dim=1248 input=Append(-3,3)
  relu-renorm-layer name=tdnn5 dim=1248 input=Append(-7,2)
  relu-renorm-layer name=tdnn6 dim=1248
  output-layer name=output dim=$num_targets max-change=1.5
EOF
  steps/nnet3/xconfig_to_configs.py --xconfig-file $dir/configs/network.xconfig --config-dir $dir/configs/
fi

if [ $stage -le 8 ]; then
  if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
    utils/create_split_dir.pl \
     /export/b0{3,4,5,6}/$USER/kaldi-data/egs/aspire-$(date +'%m_%d_%H_%M')/s5/$dir/egs/storage $dir/egs/storage
  fi

  steps/nnet3/train_dnn.py --stage=$train_stage \
    --cmd="$decode_cmd" \
    --feat.online-ivector-dir exp/nnet3/ivectors_train \
    --feat.cmvn-opts="--norm-means=false --norm-vars=false" \
    --trainer.num-epochs 3 \
    --trainer.optimization.num-jobs-initial 4 \
    --trainer.optimization.num-jobs-final 22 \
    --trainer.optimization.initial-effective-lrate 0.0017 \
    --trainer.optimization.final-effective-lrate 0.00017 \
    --egs.dir "$common_egs_dir" \
    --egs.stage "$egs_stage" \
    --cleanup.remove-egs $remove_egs \
    --cleanup.preserve-model-interval 50 \
    --feat-dir=data/train_rvb_hires \
    --ali-dir $ali_dir \
    --lang data/lang \
    --reporting.email="$reporting_email" \
    --dir=$dir  || exit 1;

fi


#ASpIRE decodes
if [ $stage -le 9 ]; then
  local/nnet3/prep_test_aspire.sh --stage 1 --decode-num-jobs 30  --affix "v7" \
   --window 10 --overlap 5 \
   --sub-speaker-frames 6000 --max-count 75 --ivector-scale 0.75  \
   --pass2-decode-opts "--min-active 1000" \
   dev_aspire data/lang exp/tri5a/graph_pp $dir
fi


exit 0;

# final result
# %WER 31.0 | 2120 27217 | 74.8 16.1 9.1 5.9 31.0 77.9 | -0.707 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iterfinal_pp_fg/score_14/penalty_0.0/ctm.filt.filt.sys

# intermediate results
#%WER 34.2 | 2120 27212 | 71.6 18.3 10.2 5.8 34.2 80.2 | -0.613 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter100_pp_fg/score_14/penalty_0.0/ctm.filt.filt.sys

#%WER 32.8 | 2120 27212 | 73.2 17.3 9.4 6.0 32.8 79.3 | -0.657 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter200_pp_fg/score_13/penalty_0.0/ctm.filt.filt.sys

#%WER 32.3 | 2120 27215 | 73.7 17.1 9.2 6.0 32.3 79.7 | -0.676 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter300_pp_fg/score_13/penalty_0.0/ctm.filt.filt.sys

#%WER 31.7 | 2120 27215 | 74.3 16.8 8.9 6.0 31.7 78.9 | -0.690 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter400_pp_fg/score_13/penalty_0.0/ctm.filt.filt.sys

#%WER 31.6 | 2120 27216 | 74.5 16.6 8.8 6.1 31.6 79.7 | -0.723 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter500_pp_fg/score_13/penalty_0.0/ctm.filt.filt.sys

#%WER 31.3 | 2120 27216 | 74.9 16.6 8.5 6.2 31.3 78.4 | -0.737 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter600_pp_fg/score_12/penalty_0.0/ctm.filt.filt.sys

#%WER 31.2 | 2120 27216 | 74.7 16.2 9.1 5.9 31.2 79.0 | -0.708 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter700_pp_fg/score_14/penalty_0.0/ctm.filt.filt.sys

#%WER 31.1 | 2120 27219 | 74.7 16.4 8.9 5.9 31.1 78.4 | -0.732 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter800_pp_fg/score_12/penalty_0.25/ctm.filt.filt.sys

#%WER 31.1 | 2120 27220 | 74.9 16.3 8.8 6.0 31.1 78.1 | -0.719 | exp/nnet3/tdnn/decode_dev_aspire_whole_uniformsegmented_win10_over5_v6_200jobs_iter1000_pp_fg/score_13/penalty_0.0/ctm.filt.filt.sys

exit 0;