run_tdnn_aug.sh

#!/bin/bash

# This recipe does multi-style training of TDNN model

# local/chain/compare_wer_general.sh --rt03 tdnn7q_sp tdnn1a_aug
# System                tdnn7q_sp tdnn1a_aug
# WER on train_dev(tg)      11.91     12.06
# WER on train_dev(fg)      10.99     10.92
# WER on eval2000(tg)        14.3      14.4
# WER on eval2000(fg)        12.8      12.9
# WER on rt03(tg)            17.2      17.1
# WER on rt03(fg)            15.1      14.8
# Final train prob         -0.062    -0.087
# Final valid prob         -0.074    -0.105
# Final train prob (xent)        -0.933    -1.164
# Final valid prob (xent)       -0.9027   -1.2246
# Num-parameters               18693376  18483664

set -e

# configs for 'chain'
stage=0
train_stage=-10
get_egs_stage=-10
num_epochs=12

# Augmentation options
aug_list="noise" # Original train dir is referred to as `clean`
num_reverb_copies=1
use_ivectors=false

affix=1a
suffix="_aug"
if [ -e data/rt03 ]; then maybe_rt03=rt03; else maybe_rt03= ; fi

decode_iter=
decode_nj=50

# training options
frames_per_eg=150,110,100
remove_egs=false
common_egs_dir=
xent_regularize=0.1
dropout_schedule='0,0@0.20,0.5@0.50,0'

test_online_decoding=false  # if true, it will run the last decoding stage.

# End configuration section.
echo "$0 $@"  # Print the command line for logging

. ./cmd.sh
. ./path.sh
. ./utils/parse_options.sh

dir=exp/chain/tdnn${affix}${suffix}

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

clean_set=train_nodup
clean_ali=tri4_ali_nodup
train_set=$clean_set$suffix # Will be prepared by the script local/nnet3/prepare_multistyle_data.sh
ali_dir=$clean_ali$suffix
treedir=exp/chain/tri5_7d_tree$suffix
lang=data/lang_chain_2y

#if [ $stage -le 0 ]; then
#  steps/align_fmllr.sh --nj 30 --cmd "$train_cmd" \
#    data/$clean_set data/lang exp/tri4 exp/$clean_ali
#fi

# First creates augmented data and then extracts features for it data
# The script also creates alignments for aug data by copying clean alignments
if [ $stage -le 0 ]; then
  local/nnet3/multi_condition/run_aug_common.sh --stage $stage \
    --aug-list "$aug_list" --num-reverb-copies $num_reverb_copies \
    --use-ivectors "$use_ivectors" \
    --train-set $clean_set --clean-ali $clean_ali || exit 1;
fi

if [ $stage -le 11 ]; then
  # Get the alignments as lattices (gives the LF-MMI training more freedom).
  # use the same num-jobs as the alignments
  prefixes=""
  include_original=false
  for n in $aug_list; do
    if [ "$n" == "reverb" ]; then
      for i in `seq 1 $num_reverb_copies`; do
        prefixes="$prefixes "reverb$i
      done
    elif [ "$n" != "clean" ]; then
      prefixes="$prefixes "$n
    else
      # The original train directory will not have any prefix
      # include_original flag will take care of copying the original lattices
      include_original=true
    fi
  done
  nj=$(cat exp/tri4_ali_nodup$suffix/num_jobs) || exit 1;
  steps/align_fmllr_lats.sh --nj $nj --cmd "$train_cmd" data/${clean_set} \
    data/lang exp/tri4 exp/tri4_lats_nodup${suffix}_clean
  rm exp/tri4_lats_nodup${suffix}_clean/fsts.*.gz # save space
  steps/copy_lat_dir.sh --nj $nj --cmd "$train_cmd" \
    --include-original "$include_original" --prefixes "$prefixes" \
    data/${train_set} exp/tri4_lats_nodup${suffix}_clean exp/tri4_lats_nodup${suffix} || exit 1;

  if [[ $aug_list =~ "clean" ]]; then
    ln -sf tri4_lats_nodup${suffix} exp/tri4_lats_nodup${suffix}_clean2
  else
    include_original=true
    steps/copy_lat_dir.sh --nj $nj --cmd "$train_cmd" \
    --include-original "$include_original" --prefixes "$prefixes" \
    data/${train_set}_clean exp/tri4_lats_nodup${suffix}_clean exp/tri4_lats_nodup${suffix}_clean2 || exit 1;
  fi
fi

if [ $stage -le 12 ]; then
  # Create a version of the lang/ directory that has one state per phone in the
  # topo file. [note, it really has two states.. the first one is only repeated
  # once, the second one has zero or more repeats.]
  rm -rf $lang
  cp -r data/lang $lang
  silphonelist=$(cat $lang/phones/silence.csl) || exit 1;
  nonsilphonelist=$(cat $lang/phones/nonsilence.csl) || exit 1;
  # Use our special topology... note that later on may have to tune this
  # topology.
  steps/nnet3/chain/gen_topo.py $nonsilphonelist $silphonelist >$lang/topo
fi

if [ $stage -le 13 ]; then
  # Build a tree using our new topology. This is the critically different
  # step compared with other recipes.
  steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 \
      --context-opts "--context-width=2 --central-position=1" \
      --cmd "$train_cmd" 7000 data/$train_set $lang exp/$ali_dir $treedir
fi

if [ $stage -le 14 ]; then
  echo "$0: creating neural net configs using the xconfig parser";

  num_targets=$(tree-info $treedir/tree |grep num-pdfs|awk '{print $2}')
  learning_rate_factor=$(echo "print 0.5/$xent_regularize" | python)
  affine_opts="l2-regularize=0.01 dropout-proportion=0.0 dropout-per-dim=true dropout-per-dim-continuous=true"
  tdnnf_opts="l2-regularize=0.01 dropout-proportion=0.0 bypass-scale=0.66"
  linear_opts="l2-regularize=0.01 orthonormal-constraint=-1.0"
  prefinal_opts="l2-regularize=0.01"
  output_opts="l2-regularize=0.002"

  mkdir -p $dir/configs

  cat <<EOF > $dir/configs/network.xconfig
  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(-1,0,1) affine-transform-file=$dir/configs/lda.mat

  # the first splicing is moved before the lda layer, so no splicing here
  relu-batchnorm-dropout-layer name=tdnn1 $affine_opts dim=1536
  tdnnf-layer name=tdnnf2 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=1
  tdnnf-layer name=tdnnf3 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=1
  tdnnf-layer name=tdnnf4 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=1
  tdnnf-layer name=tdnnf5 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=0
  tdnnf-layer name=tdnnf6 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf7 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf8 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf9 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf10 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf11 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf12 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf13 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf14 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  tdnnf-layer name=tdnnf15 $tdnnf_opts dim=1536 bottleneck-dim=160 time-stride=3
  linear-component name=prefinal-l dim=256 $linear_opts

  prefinal-layer name=prefinal-chain input=prefinal-l $prefinal_opts big-dim=1536 small-dim=256
  output-layer name=output include-log-softmax=false dim=$num_targets $output_opts

  prefinal-layer name=prefinal-xent input=prefinal-l $prefinal_opts big-dim=1536 small-dim=256
  output-layer name=output-xent dim=$num_targets learning-rate-factor=$learning_rate_factor $output_opts
EOF
  steps/nnet3/xconfig_to_configs.py --xconfig-file $dir/configs/network.xconfig --config-dir $dir/configs/
fi

if [ $stage -le 15 ]; then
  if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
    utils/create_split_dir.pl \
     /export/b0{5,6,7,8}/$USER/kaldi-data/egs/swbd-$(date +'%m_%d_%H_%M')/s5c/$dir/egs/storage $dir/egs/storage
  fi

  steps/nnet3/chain/train.py --stage $train_stage \
    --cmd "$train_cmd" \
    --feat.cmvn-opts "--norm-means=false --norm-vars=false" \
    --chain.xent-regularize $xent_regularize \
    --chain.leaky-hmm-coefficient 0.1 \
    --chain.l2-regularize 0.0 \
    --chain.apply-deriv-weights false \
    --chain.lm-opts="--num-extra-lm-states=2000" \
    --trainer.dropout-schedule $dropout_schedule \
    --trainer.add-option="--optimization.memory-compression-level=2" \
    --egs.dir "$common_egs_dir" \
    --egs.stage $get_egs_stage \
    --egs.opts "--frames-overlap-per-eg 0 --constrained false" \
    --egs.chunk-width $frames_per_eg \
    --trainer.num-chunk-per-minibatch 64 \
    --trainer.frames-per-iter 1500000 \
    --trainer.num-epochs $num_epochs \
    --trainer.optimization.num-jobs-initial 3 \
    --trainer.optimization.num-jobs-final 16 \
    --trainer.optimization.initial-effective-lrate 0.00025 \
    --trainer.optimization.final-effective-lrate 0.000025 \
    --trainer.max-param-change 2.0 \
    --cleanup.remove-egs $remove_egs \
    --feat-dir data/${train_set}_hires \
    --tree-dir $treedir \
    --lat-dir exp/tri4_lats_nodup$suffix \
    --dir $dir  || exit 1;

fi

if [ $stage -le 16 ]; then
  utils/mkgraph.sh --self-loop-scale 1.0 data/lang_sw1_tg $dir $dir/graph_sw1_tg
fi


graph_dir=$dir/graph_sw1_tg
iter_opts=
if [ ! -z $decode_iter ]; then
  iter_opts=" --iter $decode_iter "
fi
if [ $stage -le 17 ]; then
  rm $dir/.error 2>/dev/null || true
  for decode_set in train_dev eval2000; do
      (
      steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
          --nj $decode_nj --cmd "$decode_cmd" $iter_opts \
          $graph_dir data/${decode_set}_hires \
          $dir/decode_${decode_set}${decode_iter:+_$decode_iter}_sw1_tg || exit 1;
      if $has_fisher; then
          steps/lmrescore_const_arpa.sh --cmd "$decode_cmd" \
            data/lang_sw1_{tg,fsh_fg} data/${decode_set}_hires \
            $dir/decode_${decode_set}${decode_iter:+_$decode_iter}_sw1_{tg,fsh_fg} || exit 1;
      fi
      ) || touch $dir/.error &
  done
  wait
  if [ -f $dir/.error ]; then
    echo "$0: something went wrong in decoding"
    exit 1
  fi
fi

if [ $stage -le 18 ]; then
  rm $dir/.error 2>/dev/null || true
  for x in eval2000; do
    for n in noise reverb music babble; do
      decode_set=${x}_${n}
      dataset=$decode_set
      steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
          --nj $decode_nj --cmd "$decode_cmd" $iter_opts \
          $graph_dir data/${decode_set}_hires \
          $dir/decode_${decode_set}${decode_iter:+_$decode_iter}_sw1_tg || exit 1;
      #if $has_fisher; then
      #    steps/lmrescore_const_arpa.sh --cmd "$decode_cmd" \
      #      data/lang_sw1_{tg,fsh_fg} data/${decode_set}_hires \
      #      $dir/decode_${decode_set}${decode_iter:+_$decode_iter}_sw1_{tg,fsh_fg} || exit 1;
      #fi
    done
  done
  if [ -f $dir/.error ]; then
    echo "$0: something went wrong in decoding"
    exit 1
  fi
fi

#if [ $stage -le 19 ]; then
#  rm $dir/.error 2>/dev/null || true
#  for decode_set in dev_aspire sdm1_eval ihm_eval; do
#    steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
#        --nj 10 --cmd "$decode_cmd" $iter_opts \
#        $graph_dir data/${decode_set}_hires \
#        $dir/decode_${decode_set} || exit 1;
#  done
#  if [ -f $dir/.error ]; then
#    echo "$0: something went wrong in decoding"
#    exit 1
#  fi
#fi

exit 0;