#!/bin/bash
set -e

## Adapted from swbd for librispeech by David van Leeuwen

# 7n is a kind of factorized TDNN, with skip connections

# steps/info/chain_dir_info.pl exp/chain_cleaned/tdnn_1c_sp
# exp/chain_cleaned/tdnn_1c_sp: num-iters=1307 nj=3..16 num-params=20.1M dim=40+100->6024 combine=-0.051->-0.050 (over 23) xent:train/valid[869,1306,final]=(-0.808,-0.767,-0.771/-0.828,-0.780,-0.787) logprob:train/valid[869,1306,final]=(-0.051,-0.049,-0.047/-0.059,-0.056,-0.056)

# local/chain/compare_wer.sh exp/chain_cleaned/tdnn_1b_sp exp/chain_cleaned/tdnn_1c_sp
# System                      tdnn_1b_sp tdnn_1c_sp
# WER on dev(fglarge)              3.77      3.35
# WER on dev(tglarge)              3.90      3.49
# WER on dev(tgmed)                4.89      4.30
# WER on dev(tgsmall)              5.47      4.78
# WER on dev_other(fglarge)       10.05      8.76
# WER on dev_other(tglarge)       10.80      9.26
# WER on dev_other(tgmed)         13.07     11.21
# WER on dev_other(tgsmall)       14.46     12.47
# WER on test(fglarge)             4.20      3.87
# WER on test(tglarge)             4.28      4.08
# WER on test(tgmed)               5.31      4.80
# WER on test(tgsmall)             5.97      5.25
# WER on test_other(fglarge)      10.44      8.95
# WER on test_other(tglarge)      11.05      9.41
# WER on test_other(tgmed)        13.36     11.52
# WER on test_other(tgsmall)      14.90     12.66
# Final train prob              -0.0670   -0.0475
# Final valid prob              -0.0704   -0.0555
# Final train prob (xent)       -1.0502   -0.7708
# Final valid prob (xent)       -1.0441   -0.7874

# configs for 'chain'
stage=0
decode_nj=50
train_set=train_960_cleaned
gmm=tri6b_cleaned
nnet3_affix=_cleaned

# The rest are configs specific to this script.  Most of the parameters
# are just hardcoded at this level, in the commands below.
affix=1c
tree_affix=
train_stage=-10
get_egs_stage=-10
decode_iter=

# TDNN options
frames_per_eg=150,110,100
remove_egs=true
common_egs_dir=
xent_regularize=0.1

test_online_decoding=true  # 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

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

# The iVector-extraction and feature-dumping parts are the same as the standard
# nnet3 setup, and you can skip them by setting "--stage 11" if you have already
# run those things.

local/nnet3/run_ivector_common.sh --stage $stage \
                                  --train-set $train_set \
                                  --gmm $gmm \
                                  --num-threads-ubm 6 --num-processes 3 \
                                  --nnet3-affix "$nnet3_affix" || exit 1;

gmm_dir=exp/$gmm
ali_dir=exp/${gmm}_ali_${train_set}_sp
tree_dir=exp/chain${nnet3_affix}/tree_sp${tree_affix:+_$tree_affix}
lang=data/lang_chain
lat_dir=exp/chain${nnet3_affix}/${gmm}_${train_set}_sp_lats
dir=exp/chain${nnet3_affix}/tdnn${affix:+_$affix}_sp
train_data_dir=data/${train_set}_sp_hires
lores_train_data_dir=data/${train_set}_sp
train_ivector_dir=exp/nnet3${nnet3_affix}/ivectors_${train_set}_sp_hires

# if we are using the speed-perturbed data we need to generate
# alignments for it.

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

# Please take this as a reference on how to specify all the options of
# local/chain/run_chain_common.sh
local/chain/run_chain_common.sh --stage $stage \
                                --gmm-dir $gmm_dir \
                                --ali-dir $ali_dir \
                                --lores-train-data-dir ${lores_train_data_dir} \
                                --lang $lang \
                                --lat-dir $lat_dir \
				--num-leaves 7000 \
                                --tree-dir $tree_dir || exit 1;

if [ $stage -le 14 ]; 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)
  opts="l2-regularize=0.002"
  linear_opts="orthonormal-constraint=1.0"
  output_opts="l2-regularize=0.0005 bottleneck-dim=256"

  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(-1,0,1,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-batchnorm-layer name=tdnn1 $opts dim=1280
  linear-component name=tdnn2l dim=256 $linear_opts input=Append(-1,0)
  relu-batchnorm-layer name=tdnn2 $opts input=Append(0,1) dim=1280
  linear-component name=tdnn3l dim=256 $linear_opts
  relu-batchnorm-layer name=tdnn3 $opts dim=1280
  linear-component name=tdnn4l dim=256 $linear_opts input=Append(-1,0)
  relu-batchnorm-layer name=tdnn4 $opts input=Append(0,1) dim=1280
  linear-component name=tdnn5l dim=256 $linear_opts
  relu-batchnorm-layer name=tdnn5 $opts dim=1280 input=Append(tdnn5l, tdnn3l)
  linear-component name=tdnn6l dim=256 $linear_opts input=Append(-3,0)
  relu-batchnorm-layer name=tdnn6 $opts input=Append(0,3) dim=1280
  linear-component name=tdnn7l dim=256 $linear_opts input=Append(-3,0)
  relu-batchnorm-layer name=tdnn7 $opts input=Append(0,3,tdnn6l,tdnn4l,tdnn2l) dim=1280
  linear-component name=tdnn8l dim=256 $linear_opts input=Append(-3,0)
  relu-batchnorm-layer name=tdnn8 $opts input=Append(0,3) dim=1280
  linear-component name=tdnn9l dim=256 $linear_opts input=Append(-3,0)
  relu-batchnorm-layer name=tdnn9 $opts input=Append(0,3,tdnn8l,tdnn6l,tdnn4l) dim=1280
  linear-component name=tdnn10l dim=256 $linear_opts input=Append(-3,0)
  relu-batchnorm-layer name=tdnn10 $opts input=Append(0,3) dim=1280
  linear-component name=tdnn11l dim=256 $linear_opts input=Append(-3,0)
  relu-batchnorm-layer name=tdnn11 $opts input=Append(0,3,tdnn10l,tdnn8l,tdnn6l) dim=1280
  linear-component name=prefinal-l dim=256 $linear_opts

  relu-batchnorm-layer name=prefinal-chain input=prefinal-l $opts dim=1280
  output-layer name=output include-log-softmax=false dim=$num_targets $output_opts

  relu-batchnorm-layer name=prefinal-xent input=prefinal-l $opts dim=1280
  output-layer name=output-xent dim=$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
  steps/nnet3/chain/train.py --stage $train_stage \
    --cmd "$decode_cmd" \
    --feat.online-ivector-dir $train_ivector_dir \
    --feat.cmvn-opts "--norm-means=false --norm-vars=false" \
    --chain.xent-regularize $xent_regularize \
    --chain.leaky-hmm-coefficient 0.1 \
    --chain.l2-regularize 0.0 \
    --chain.apply-deriv-weights false \
    --chain.lm-opts="--num-extra-lm-states=2000" \
    --egs.dir "$common_egs_dir" \
    --egs.stage $get_egs_stage \
    --egs.opts "--frames-overlap-per-eg 0" \
    --egs.chunk-width $frames_per_eg \
    --trainer.num-chunk-per-minibatch 128 \
    --trainer.frames-per-iter 1500000 \
    --trainer.num-epochs 6 \
    --trainer.optimization.num-jobs-initial 3 \
    --trainer.optimization.num-jobs-final 16 \
    --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

graph_dir=$dir/graph_tgsmall
if [ $stage -le 16 ]; then
  # Note: it might appear that this $lang directory is mismatched, and it is as
  # far as the 'topo' is concerned, but this script doesn't read the 'topo' from
  # the lang directory.
  utils/mkgraph.sh --self-loop-scale 1.0 --remove-oov data/lang_test_tgsmall $dir $graph_dir
  # remove <UNK> from the graph, and convert back to const-FST.
  fstrmsymbols --apply-to-output=true --remove-arcs=true "echo 3|" $graph_dir/HCLG.fst - | \
    fstconvert --fst_type=const > $graph_dir/temp.fst
  mv $graph_dir/temp.fst $graph_dir/HCLG.fst
fi

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 test_clean test_other dev_clean dev_other; do
      (
      steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
          --nj $decode_nj --cmd "$decode_cmd" $iter_opts \
          --online-ivector-dir exp/nnet3${nnet3_affix}/ivectors_${decode_set}_hires \
          $graph_dir data/${decode_set}_hires $dir/decode_${decode_set}${decode_iter:+_$decode_iter}_tgsmall || exit 1
      steps/lmrescore.sh --cmd "$decode_cmd" --self-loop-scale 1.0 data/lang_test_{tgsmall,tgmed} \
          data/${decode_set}_hires $dir/decode_${decode_set}${decode_iter:+_$decode_iter}_{tgsmall,tgmed} || exit 1
      steps/lmrescore_const_arpa.sh \
          --cmd "$decode_cmd" data/lang_test_{tgsmall,tglarge} \
          data/${decode_set}_hires $dir/decode_${decode_set}${decode_iter:+_$decode_iter}_{tgsmall,tglarge} || exit 1
      steps/lmrescore_const_arpa.sh \
          --cmd "$decode_cmd" data/lang_test_{tgsmall,fglarge} \
          data/${decode_set}_hires $dir/decode_${decode_set}${decode_iter:+_$decode_iter}_{tgsmall,fglarge} || exit 1
      ) || touch $dir/.error &
  done
  wait
  if [ -f $dir/.error ]; then
    echo "$0: something went wrong in decoding"
    exit 1
  fi
fi

if $test_online_decoding && [ $stage -le 18 ]; then
  # note: if the features change (e.g. you add pitch features), you will have to
  # change the options of the following command line.
  steps/online/nnet3/prepare_online_decoding.sh \
       --mfcc-config conf/mfcc_hires.conf \
       $lang exp/nnet3${nnet3_affix}/extractor $dir ${dir}_online

  rm $dir/.error 2>/dev/null || true
  for data in test_clean test_other dev_clean dev_other; do
    (
      nspk=$(wc -l <data/${data}_hires/spk2utt)
      # note: we just give it "data/${data}" as it only uses the wav.scp, the
      # feature type does not matter.
      steps/online/nnet3/decode.sh \
          --acwt 1.0 --post-decode-acwt 10.0 \
          --nj $nspk --cmd "$decode_cmd" \
          $graph_dir data/${data} ${dir}_online/decode_${data}_tgsmall || exit 1

    ) || touch $dir/.error &
  done
  wait
  if [ -f $dir/.error ]; then
    echo "$0: something went wrong in decoding"
    exit 1
  fi
fi


exit 0;