/training/local/chain/run_tdnn.sh

#!/bin/bash

# Set -e here so that we catch if any executable fails immediately
set -euo pipefail

# (some of which are also used in this script directly).
stage=-1
decode_nj=10
train_set=train
gmm=tri3
nnet3_affix=
suffix=

# The rest are configs specific to this script.  Most of the parameters
# are just hardcoded at this level, in the commands below.
affix=   # affix for the TDNN directory name
tree_affix=
train_stage=-10
get_egs_stage=-10
decode_iter=

# training options
# training chunk-options
chunk_width=140,100,160
common_egs_dir=
xent_regularize=0.1
dropout_schedule='0,0@0.20,0.5@0.50,0'

# training options
srand=0
remove_egs=true

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

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

# Problem: We have removed the "train_" prefix of our training set in
# the alignment directory names! Bad!
gmm_dir=exp/$gmm
ali_dir=exp/${gmm}_ali
tree_dir=exp/chain${suffix}/tree${tree_affix:+_$tree_affix}
lang=data/lang_chain${suffix}
lat_dir=exp/chain${suffix}/${gmm}_${train_set}_lats
dir=exp/chain${suffix}/tdnn${affix}
train_data_dir=data/${train_set}

for f in $gmm_dir/final.mdl $train_data_dir/feats.scp $ali_dir/ali.1.gz; do
  [ ! -f $f ] && echo "$0: expected file $f to exist" && exit 1
done

if [ $stage -le 9 ]; then
    local/chain/run_ivector_common.sh \
                          --train-set ${train_set} \
                          --gmm ${gmm} \
                          --suffix "${suffix}"
fi

if [ $stage -le 10 ]; then
  echo "$0: creating lang directory $lang with chain-type topology"
  rm -rf $lang
  cp -r data/lang $lang
  silphonelist=$(cat $lang/phones/silence.csl) 
  nonsilphonelist=$(cat $lang/phones/nonsilence.csl) 
  steps/nnet3/chain/gen_topo.py $nonsilphonelist $silphonelist >$lang/topo
fi

if [ $stage -le 11 ]; then
    steps/align_fmllr_lats.sh --nj 20 --cmd "$train_cmd" ${train_data_dir} \
         data/lang $gmm_dir $lat_dir
fi

if [ $stage -le 12 ]; then
  steps/nnet3/chain/build_tree.sh \
    --frame-subsampling-factor 3 \
    --context-opts "--context-width=2 --central-position=1" \
    --cmd "$train_cmd" 2500 ${train_data_dir} \
    $lang $ali_dir $tree_dir
fi

if [ $stage -le 13 ]; then
  echo "$0: creating neural net configs using the xconfig parser";

  num_targets=$(tree-info $tree_dir/tree | grep num-pdfs | awk '{print $2}')
  learning_rate_factor=$(echo "print (0.5/$xent_regularize)" | python)

  affine_opts="l2-regularize=0.008 dropout-proportion=0.0 dropout-per-dim=true dropout-per-dim-continuous=true"
  tdnnf_opts="l2-regularize=0.008 dropout-proportion=0.0 bypass-scale=0.75"
  linear_opts="l2-regularize=0.008 orthonormal-constraint=-1.0"
  prefinal_opts="l2-regularize=0.008"
  output_opts="l2-regularize=0.002"

  mkdir -p $dir/configs
  cat <<EOF > $dir/configs/network.xconfig

  input dim=40 name=ivector
  input dim=40 name=input

  idct-layer name=idct input=input dim=40 cepstral-lifter=22 affine-transform-file=$dir/configs/idct.mat
  batchnorm-component name=batchnorm0 input=idct
  spec-augment-layer name=spec-augment freq-max-proportion=0.5 time-zeroed-proportion=0.2 time-mask-max-frames=20
  delta-layer name=delta input=spec-augment

  no-op-component name=input2 input=Append(delta, ReplaceIndex(ivector, t, 0))

  # the first splicing is moved before the lda layer, so no splicing here
  relu-batchnorm-dropout-layer name=tdnn1 $affine_opts dim=512 input=input2
  tdnnf-layer name=tdnnf2 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=1
  tdnnf-layer name=tdnnf3 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=1
  tdnnf-layer name=tdnnf4 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=1
  tdnnf-layer name=tdnnf5 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=0
  tdnnf-layer name=tdnnf6 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=3
  tdnnf-layer name=tdnnf7 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=3
  tdnnf-layer name=tdnnf8 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=3
  tdnnf-layer name=tdnnf9 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=3
  tdnnf-layer name=tdnnf10 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=3
  tdnnf-layer name=tdnnf11 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=3
  tdnnf-layer name=tdnnf12 $tdnnf_opts dim=512 bottleneck-dim=96 time-stride=3
  linear-component name=prefinal-l dim=192 $linear_opts

  ## adding the layers for chain branch
  prefinal-layer name=prefinal-chain input=prefinal-l $prefinal_opts small-dim=192 big-dim=512
  output-layer name=output include-log-softmax=false dim=$num_targets $output_opts

  # adding the layers for xent branch
  prefinal-layer name=prefinal-xent input=prefinal-l $prefinal_opts small-dim=192 big-dim=512
  output-layer name=output-xent dim=$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 14 ]; then
  steps/nnet3/chain/train.py --stage $train_stage \
    --cmd "$cuda_cmd" \
    --feat.online-ivector-dir exp/chain${suffix}/ivectors_${train_set} \
    --feat.cmvn-opts "--norm-means=false --norm-vars=false" \
    --chain.xent-regularize $xent_regularize \
    --chain.leaky-hmm-coefficient 0.1 \
    --chain.l2-regularize 0.0 \
    --chain.apply-deriv-weights false \
    --chain.lm-opts="--num-extra-lm-states=2000" \
    --egs.cmd "$get_egs_cmd" \
    --egs.dir "$common_egs_dir" \
    --egs.stage $get_egs_stage \
    --egs.opts "--frames-overlap-per-eg 0 --constrained false" \
    --egs.chunk-width $chunk_width \
    --trainer.dropout-schedule $dropout_schedule \
    --trainer.add-option="--optimization.memory-compression-level=2" \
    --trainer.num-chunk-per-minibatch 64 \
    --trainer.frames-per-iter 2500000 \
    --trainer.num-epochs 20 \
    --trainer.optimization.num-jobs-initial 1 \
    --trainer.optimization.num-jobs-final 1 \
    --trainer.optimization.initial-effective-lrate 0.001 \
    --trainer.optimization.final-effective-lrate 0.0001 \
    --trainer.max-param-change 2.0 \
    --cleanup.remove-egs $remove_egs \
    --feat-dir $train_data_dir \
    --tree-dir $tree_dir \
    --lat-dir $lat_dir \
    --dir $dir  || exit 1;
fi