From 1f57c99713739a9d11dd314082cc7180443d7110 Mon Sep 17 00:00:00 2001
From: Ewald Enzinger <ewald.enzinger@entn.at>
Date: Fri, 30 Nov 2018 13:09:07 -0800
Subject: [PATCH] add HUB4 spanish tdnn-f and cnn-tdnn script

---
 .../s5/local/chain/compare_wer.sh             | 135 ++++++++
 .../s5/local/chain/run_cnn_tdnn.sh            |   1 +
 egs/hub4_spanish/s5/local/chain/run_tdnn.sh   |   2 +-
 .../s5/local/chain/tuning/run_cnn_tdnn_1a.sh  | 287 ++++++++++++++++++
 .../s5/local/chain/tuning/run_tdnn_1a.sh      |  12 +-
 .../s5/local/chain/tuning/run_tdnn_1b.sh      | 273 +++++++++++++++++
 6 files changed, 708 insertions(+), 2 deletions(-)
 create mode 100755 egs/hub4_spanish/s5/local/chain/compare_wer.sh
 create mode 120000 egs/hub4_spanish/s5/local/chain/run_cnn_tdnn.sh
 create mode 100755 egs/hub4_spanish/s5/local/chain/tuning/run_cnn_tdnn_1a.sh
 create mode 100755 egs/hub4_spanish/s5/local/chain/tuning/run_tdnn_1b.sh

diff --git a/egs/hub4_spanish/s5/local/chain/compare_wer.sh b/egs/hub4_spanish/s5/local/chain/compare_wer.sh
new file mode 100755
index 00000000000..0194b86ac69
--- /dev/null
+++ b/egs/hub4_spanish/s5/local/chain/compare_wer.sh
@@ -0,0 +1,135 @@
+#!/bin/bash
+
+# this script is used for comparing decoding results between systems.
+# e.g. local/chain/compare_wer.sh exp/chain/tdnn_{c,d}_sp
+# For use with discriminatively trained systems you specify the epochs after a colon:
+# for instance,
+# local/chain/compare_wer.sh exp/chain/tdnn_c_sp exp/chain/tdnn_c_sp_smbr:{1,2,3}
+
+
+if [ $# == 0 ]; then
+  echo "Usage: $0: [--looped] [--online] <dir1> [<dir2> ... ]"
+  echo "e.g.: $0 exp/chain/tdnn_{b,c}_sp"
+  echo "or (with epoch numbers for discriminative training):"
+  echo "$0 exp/chain/tdnn_b_sp_disc:{1,2,3}"
+  exit 1
+fi
+
+echo "# $0 $*"
+
+include_looped=false
+if [ "$1" == "--looped" ]; then
+  include_looped=true
+  shift
+fi
+include_online=false
+if [ "$1" == "--online" ]; then
+  include_online=true
+  shift
+fi
+
+
+used_epochs=false
+
+# this function set_names is used to separate the epoch-related parts of the name
+# [for discriminative training] and the regular parts of the name.
+# If called with a colon-free directory name, like:
+#  set_names exp/chain/tdnn_lstm1e_sp_bi_smbr
+# it will set dir=exp/chain/tdnn_lstm1e_sp_bi_smbr and epoch_infix=""
+# If called with something like:
+#  set_names exp/chain/tdnn_d_sp_smbr:3
+# it will set dir=exp/chain/tdnn_d_sp_smbr and epoch_infix="_epoch3"
+
+
+set_names() {
+  if [ $# != 1 ]; then
+    echo "compare_wer_general.sh: internal error"
+    exit 1  # exit the program
+  fi
+  dirname=$(echo $1 | cut -d: -f1)
+  epoch=$(echo $1 | cut -s -d: -f2)
+  if [ -z $epoch ]; then
+    epoch_infix=""
+  else
+    used_epochs=true
+    epoch_infix=_epoch${epoch}
+  fi
+}
+
+
+
+echo -n "# System               "
+for x in $*; do   printf "% 10s" " $(basename $x)";   done
+echo
+
+strings=("#WER test ")
+
+for n in 0; do
+   echo -n "${strings[$n]}"
+   for x in $*; do
+     set_names $x  # sets $dirname and $epoch_infix
+    decode_names=(test)
+
+     wer=$(cat $dirname/decode_${decode_names[$n]}/wer_* | utils/best_wer.sh | awk '{print $2}')
+     printf "% 10s" $wer
+   done
+   echo
+   if $include_looped; then
+     echo -n "#             [looped:]    "
+     for x in $*; do
+       set_names $x  # sets $dirname and $epoch_infix
+       wer=$(cat $dirname/decode_looped_${decode_names[$n]}/wer_* | utils/best_wer.sh | awk '{print $2}')
+       printf "% 10s" $wer
+     done
+     echo
+   fi
+   if $include_online; then
+     echo -n "#             [online:]    "
+     for x in $*; do
+       set_names $x  # sets $dirname and $epoch_infix
+       wer=$(cat ${dirname}_online/decode_${decode_names[$n]}/wer_* | utils/best_wer.sh | awk '{print $2}')
+       printf "% 10s" $wer
+     done
+     echo
+   fi
+done
+
+
+if $used_epochs; then
+  exit 0;  # the diagnostics aren't comparable between regular and discriminatively trained systems.
+fi
+
+
+echo -n "# Final train prob     "
+for x in $*; do
+  prob=$(grep Overall $x/log/compute_prob_train.final.log | grep -v xent | awk '{printf("%.4f", $8)}')
+  printf "% 10s" $prob
+done
+echo
+
+echo -n "# Final valid prob     "
+for x in $*; do
+  prob=$(grep Overall $x/log/compute_prob_valid.final.log | grep -v xent | awk '{printf("%.4f", $8)}')
+  printf "% 10s" $prob
+done
+echo
+
+echo -n "# Final train prob (xent)"
+for x in $*; do
+  prob=$(grep Overall $x/log/compute_prob_train.final.log | grep -w xent | awk '{printf("%.4f", $8)}')
+  printf "% 10s" $prob
+done
+echo
+
+echo -n "# Final valid prob (xent)"
+for x in $*; do
+  prob=$(grep Overall $x/log/compute_prob_valid.final.log | grep -w xent | awk '{printf("%.4f", $8)}')
+  printf "% 10s" $prob
+done
+echo
+
+echo -n "# Num-params              "
+for x in $*; do
+  printf "% 10s" $(grep num-parameters $x/log/progress.1.log | awk '{print $2}')
+done
+echo
diff --git a/egs/hub4_spanish/s5/local/chain/run_cnn_tdnn.sh b/egs/hub4_spanish/s5/local/chain/run_cnn_tdnn.sh
new file mode 120000
index 00000000000..ab83f3c43e8
--- /dev/null
+++ b/egs/hub4_spanish/s5/local/chain/run_cnn_tdnn.sh
@@ -0,0 +1 @@
+tuning/run_cnn_tdnn_1a.sh
\ No newline at end of file
diff --git a/egs/hub4_spanish/s5/local/chain/run_tdnn.sh b/egs/hub4_spanish/s5/local/chain/run_tdnn.sh
index 211957092f9..61f8f499182 120000
--- a/egs/hub4_spanish/s5/local/chain/run_tdnn.sh
+++ b/egs/hub4_spanish/s5/local/chain/run_tdnn.sh
@@ -1 +1 @@
-./tuning/run_tdnn_1a.sh
\ No newline at end of file
+tuning/run_tdnn_1b.sh
\ No newline at end of file
diff --git a/egs/hub4_spanish/s5/local/chain/tuning/run_cnn_tdnn_1a.sh b/egs/hub4_spanish/s5/local/chain/tuning/run_cnn_tdnn_1a.sh
new file mode 100755
index 00000000000..81915fec5a6
--- /dev/null
+++ b/egs/hub4_spanish/s5/local/chain/tuning/run_cnn_tdnn_1a.sh
@@ -0,0 +1,287 @@
+#!/bin/bash
+
+## This is taken from mini_librispeech.
+
+# local/chain/compare_wer.sh --online exp/chain/tdnn1a_sp exp/chain/cnn_tdnn1a_sp
+# System                 tdnn1a_sp cnn_tdnn1a_sp
+#WER test                    14.19         13.47
+#             [online:]      14.26         13.57
+# Final train prob         -0.0707       -0.0911
+# Final valid prob         -0.1225       -0.1145
+# Final train prob (xent)  -1.1117       -1.3038
+# Final valid prob (xent)  -1.3199       -1.3374
+# Num-params               6945216       4471200
+
+# steps/info/chain_dir_info.pl exp/chain/cnn_tdnn1a_sp
+# exp/chain/cnn_tdnn1a_sp: num-iters=102 nj=2..5 num-params=4.5M dim=40+100->2272 combine=-0.101->-0.097 (over 5) xent:train/valid[67,101,final]=(-1.46,-1.31,-1.30/-1.47,-1.34,-1.34) logprob:train/valid[67,101,final]=(-0.112,-0.097,-0.091/-0.129,-0.121,-0.114)
+
+# Set -e here so that we catch if any executable fails immediately
+set -euo pipefail
+
+# First the options that are passed through to run_ivector_common.sh
+# (some of which are also used in this script directly).
+stage=0
+decode_nj=10
+train_set=train
+test_sets=eval
+gmm=tri5
+nnet3_affix=
+
+# The rest are configs specific to this script.  Most of the parameters
+# are just hardcoded at this level, in the commands below.
+affix=1a   # 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
+dropout_schedule='0,0@0.20,0.3@0.50,0'
+# we don't need extra left/right context for TDNN systems.
+chunk_left_context=0
+chunk_right_context=0
+common_egs_dir=
+xent_regularize=0.1
+
+# training options
+srand=0
+remove_egs=true
+reporting_email=
+
+#decode options
+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 $gmm \
+                                  --nnet3-affix "$nnet3_affix" || exit 1;
+
+# 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_${train_set}_sp
+tree_dir=exp/chain${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}/cnn_tdnn${affix}_sp
+train_data_dir=data/${train_set}_sp_hires
+lores_train_data_dir=data/${train_set}_sp
+train_ivector_dir=exp/nnet3${nnet3_affix}/ivectors_${train_set}_sp_hires
+
+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
+
+if [ $stage -le 10 ]; then
+  echo "$0: creating lang directory $lang with chain-type topology"
+  # 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.]
+  if [ -d $lang ]; then
+    if [ $lang/L.fst -nt data/lang/L.fst ]; then
+      echo "$0: $lang already exists, not overwriting it; continuing"
+    else
+      echo "$0: $lang already exists and seems to be older than data/lang..."
+      echo " ... not sure what to do.  Exiting."
+      exit 1;
+    fi
+  else
+    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
+fi
+
+if [ $stage -le 11 ]; then
+  # Get the alignments as lattices (gives the chain training more freedom).
+  # use the same num-jobs as the alignments
+  steps/align_fmllr_lats.sh --nj 75 --cmd "$train_cmd" ${lores_train_data_dir} \
+    data/lang $gmm_dir $lat_dir
+  rm $lat_dir/fsts.*.gz # save space
+fi
+
+if [ $stage -le 12 ]; then
+  # Build a tree using our new topology.  We know we have alignments for the
+  # speed-perturbed data (local/nnet3/run_ivector_common.sh made them), so use
+  # those.  The num-leaves is always somewhat less than the num-leaves from
+  # the GMM baseline.
+   if [ -f $tree_dir/final.mdl ]; then
+     echo "$0: $tree_dir/final.mdl already exists, refusing to overwrite it."
+     exit 1;
+  fi
+  steps/nnet3/chain/build_tree.sh \
+    --frame-subsampling-factor 3 \
+    --context-opts "--context-width=2 --central-position=1" \
+    --cmd "$train_cmd" 3500 ${lores_train_data_dir} \
+    $lang $ali_dir $tree_dir
+fi
+
+
+if [ $stage -le 13 ]; then
+  mkdir -p $dir
+  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)
+
+  cnn_opts="l2-regularize=0.03"
+  ivector_affine_opts="l2-regularize=0.03"
+  tdnn_opts="l2-regularize=0.03 dropout-proportion=0.0 dropout-per-dim-continuous=true"
+  tdnnf_first_opts="l2-regularize=0.03 dropout-proportion=0.0 bypass-scale=0.0"
+  tdnnf_opts="l2-regularize=0.03 dropout-proportion=0.0 bypass-scale=0.66"
+  linear_opts="l2-regularize=0.03 orthonormal-constraint=-1.0"
+  prefinal_opts="l2-regularize=0.03"
+  output_opts="l2-regularize=0.015"
+
+  mkdir -p $dir/configs
+  cat <<EOF > $dir/configs/network.xconfig
+  input dim=100 name=ivector
+  input dim=40 name=input
+
+  # this takes the MFCCs and generates filterbank coefficients.  The MFCCs
+  # are more compressible so we prefer to dump the MFCCs to disk rather
+  # than filterbanks.
+  idct-layer name=idct input=input dim=40 cepstral-lifter=22 affine-transform-file=$dir/configs/idct.mat
+
+  linear-component name=ivector-linear $ivector_affine_opts dim=200 input=ReplaceIndex(ivector, t, 0)
+  batchnorm-component name=ivector-batchnorm target-rms=0.025
+
+  batchnorm-component name=idct-batchnorm input=idct
+  combine-feature-maps-layer name=combine_inputs input=Append(idct-batchnorm, ivector-batchnorm) num-filters1=1 num-filters2=5 height=40
+
+  conv-relu-batchnorm-layer name=cnn1 $cnn_opts height-in=40 height-out=40 time-offsets=-1,0,1 height-offsets=-1,0,1 num-filters-out=48 learning-rate-factor=0.333 max-change=0.25
+  conv-relu-batchnorm-layer name=cnn2 $cnn_opts height-in=40 height-out=40 time-offsets=-1,0,1 height-offsets=-1,0,1 num-filters-out=48
+  conv-relu-batchnorm-layer name=cnn3 $cnn_opts height-in=40 height-out=20 height-subsample-out=2 time-offsets=-1,0,1 height-offsets=-1,0,1 num-filters-out=64
+  conv-relu-batchnorm-layer name=cnn4 $cnn_opts height-in=20 height-out=20 time-offsets=-1,0,1 height-offsets=-1,0,1 num-filters-out=64
+  conv-relu-batchnorm-layer name=cnn5 $cnn_opts height-in=20 height-out=10 height-subsample-out=2 time-offsets=-1,0,1 height-offsets=-1,0,1 num-filters-out=64
+  conv-relu-batchnorm-layer name=cnn6 $cnn_opts height-in=10 height-out=5 height-subsample-out=2 time-offsets=-1,0,1 height-offsets=-1,0,1 num-filters-out=128
+
+  # the first TDNN-F layer has no bypass (since dims don't match), and a larger bottleneck so the
+  # information bottleneck doesn't become a problem.  (we use time-stride=0 so no splicing, to
+  # limit the num-parameters).
+  tdnnf-layer name=tdnnf7 $tdnnf_first_opts dim=768 bottleneck-dim=192 time-stride=0
+  tdnnf-layer name=tdnnf8 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf9 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf10 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf11 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf12 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf13 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf14 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf15 $tdnnf_opts dim=768 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 $prefinal_opts small-dim=192 big-dim=768
+  output-layer name=output include-log-softmax=false dim=$num_targets $output_opts
+
+  # adding the layers for xent branch
+  prefinal-layer name=prefinal-xent input=prefinal-l $prefinal_opts small-dim=192 big-dim=768
+  output-layer name=output-xent dim=$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
+  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/hub4_spanish-$(date +'%m_%d_%H_%M')/s5/$dir/egs/storage $dir/egs/storage
+  fi
+
+  steps/nnet3/chain/train.py --stage=$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" \
+    --trainer.dropout-schedule $dropout_schedule \
+    --trainer.add-option="--optimization.memory-compression-level=2" \
+    --trainer.srand=$srand \
+    --trainer.max-param-change=2.0 \
+    --trainer.num-epochs=10 \
+    --trainer.frames-per-iter=3000000 \
+    --trainer.optimization.num-jobs-initial=2 \
+    --trainer.optimization.num-jobs-final=5 \
+    --trainer.optimization.initial-effective-lrate=0.001 \
+    --trainer.optimization.final-effective-lrate=0.0001 \
+    --trainer.num-chunk-per-minibatch=256,128,64 \
+    --trainer.optimization.momentum=0.0 \
+    --egs.chunk-width=$chunk_width \
+    --egs.chunk-left-context=$chunk_left_context \
+    --egs.chunk-right-context=$chunk_right_context \
+    --egs.chunk-left-context-initial=0 \
+    --egs.chunk-right-context-final=0 \
+    --egs.dir="$common_egs_dir" \
+    --egs.opts="--frames-overlap-per-eg 0" \
+    --cleanup.remove-egs=$remove_egs \
+    --use-gpu=true \
+    --reporting.email="$reporting_email" \
+    --feat-dir=$train_data_dir \
+    --tree-dir=$tree_dir \
+    --lat-dir=$lat_dir \
+    --dir=$dir  || exit 1;
+fi
+
+if [ $stage -le 15 ]; then
+  # Note: it's not important to give mkgraph.sh the lang directory with the
+  # matched topology (since it gets the topology file from the model).
+  utils/mkgraph.sh \
+    --self-loop-scale 1.0 data/langp_test \
+    $tree_dir $dir/graph || exit 1;
+fi
+
+if [ $stage -le 16 ]; then
+  frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+  nspk=$(wc -l <data/eval/spk2utt)
+  steps/nnet3/decode.sh \
+    --acwt 1.0 --post-decode-acwt 10.0 \
+    --frames-per-chunk $frames_per_chunk \
+    --nj $nspk --cmd "$decode_cmd"  --num-threads 4 \
+    --online-ivector-dir exp/nnet3${nnet3_affix}/ivectors_eval_hires \
+    $dir/graph data/eval_hires $dir/decode_test || exit 1
+fi
+
+if $test_online_decoding && [ $stage -le 17 ]; 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} ${dir}_online
+
+    nspk=$(wc -l <data/eval/spk2utt)
+    steps/online/nnet3/decode.sh \
+      --acwt 1.0 --post-decode-acwt 10.0 \
+      --nj $nspk --cmd "$decode_cmd" \
+      $dir/graph  data/eval ${dir}_online/decode_test  || exit 1
+fi
+
+
+exit 0;
diff --git a/egs/hub4_spanish/s5/local/chain/tuning/run_tdnn_1a.sh b/egs/hub4_spanish/s5/local/chain/tuning/run_tdnn_1a.sh
index 470833cec70..23a55f93023 100755
--- a/egs/hub4_spanish/s5/local/chain/tuning/run_tdnn_1a.sh
+++ b/egs/hub4_spanish/s5/local/chain/tuning/run_tdnn_1a.sh
@@ -179,7 +179,7 @@ fi
 if [ $stage -le 14 ]; 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/mini_librispeech-$(date +'%m_%d_%H_%M')/s5/$dir/egs/storage $dir/egs/storage
+     /export/b0{3,4,5,6}/$USER/kaldi-data/egs/hub4_spanish-$(date +'%m_%d_%H_%M')/s5/$dir/egs/storage $dir/egs/storage
   fi
 
   steps/nnet3/chain/train.py --stage=$train_stage \
@@ -227,6 +227,16 @@ if [ $stage -le 15 ]; then
     $tree_dir $dir/graph || exit 1;
 fi
 
+if [ $stage -le 16 ]; then
+  frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+  nspk=$(wc -l <data/eval/spk2utt)
+  steps/nnet3/decode.sh \
+    --acwt 1.0 --post-decode-acwt 10.0 \
+    --frames-per-chunk $frames_per_chunk \
+    --nj $nspk --cmd "$decode_cmd"  --num-threads 4 \
+    --online-ivector-dir exp/nnet3${nnet3_affix}/ivectors_eval_hires \
+    $dir/graph data/eval_hires $dir/decode_test || exit 1
+fi
 
 if [ $stage -le 17 ]; then
   # note: if the features change (e.g. you add pitch features), you will have to
diff --git a/egs/hub4_spanish/s5/local/chain/tuning/run_tdnn_1b.sh b/egs/hub4_spanish/s5/local/chain/tuning/run_tdnn_1b.sh
new file mode 100755
index 00000000000..724bb1e0794
--- /dev/null
+++ b/egs/hub4_spanish/s5/local/chain/tuning/run_tdnn_1b.sh
@@ -0,0 +1,273 @@
+#!/bin/bash
+
+## This is taken from mini_librispeech
+
+# local/chain/compare_wer.sh exp/chain/tdnn1a_sp exp/chain/tdnn1b_sp
+# System                 tdnn1a_sp tdnn1b_sp
+#WER test                    14.19     13.89
+#             [online:]      14.26     14.02
+# Final train prob         -0.0707   -0.0941
+# Final valid prob         -0.1225   -0.1165
+# Final train prob (xent)  -1.1117   -1.3456
+# Final valid prob (xent)  -1.3199   -1.3938
+# Num-params               6945216   5186240
+
+# steps/info/chain_dir_info.pl exp/chain/tdnn1b_sp
+# exp/chain/tdnn1b_sp: num-iters=102 nj=2..5 num-params=5.2M dim=40+100->2272 combine=-0.105->-0.100 (over 6) xent:train/valid[67,101,final]=(-1.54,-1.34,-1.35/-1.56,-1.39,-1.39) logprob:train/valid[67,101,final]=(-0.116,-0.099,-0.094/-0.135,-0.123,-0.116)
+
+# Set -e here so that we catch if any executable fails immediately
+set -euo pipefail
+
+# First the options that are passed through to run_ivector_common.sh
+# (some of which are also used in this script directly).
+stage=0
+decode_nj=10
+train_set=train
+test_sets=eval
+gmm=tri5
+nnet3_affix=
+
+# The rest are configs specific to this script.  Most of the parameters
+# are just hardcoded at this level, in the commands below.
+affix=1b   # 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
+dropout_schedule='0,0@0.20,0.3@0.50,0'
+# we don't need extra left/right context for TDNN systems.
+chunk_left_context=0
+chunk_right_context=0
+common_egs_dir=
+xent_regularize=0.1
+
+# training options
+srand=0
+remove_egs=true
+reporting_email=
+
+#decode options
+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 $gmm \
+                                  --nnet3-affix "$nnet3_affix" || exit 1;
+
+# 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_${train_set}_sp
+tree_dir=exp/chain${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${affix}_sp
+train_data_dir=data/${train_set}_sp_hires
+lores_train_data_dir=data/${train_set}_sp
+train_ivector_dir=exp/nnet3${nnet3_affix}/ivectors_${train_set}_sp_hires
+
+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
+
+if [ $stage -le 10 ]; then
+  echo "$0: creating lang directory $lang with chain-type topology"
+  # 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.]
+  if [ -d $lang ]; then
+    if [ $lang/L.fst -nt data/lang/L.fst ]; then
+      echo "$0: $lang already exists, not overwriting it; continuing"
+    else
+      echo "$0: $lang already exists and seems to be older than data/lang..."
+      echo " ... not sure what to do.  Exiting."
+      exit 1;
+    fi
+  else
+    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
+fi
+
+if [ $stage -le 11 ]; then
+  # Get the alignments as lattices (gives the chain training more freedom).
+  # use the same num-jobs as the alignments
+  steps/align_fmllr_lats.sh --nj 75 --cmd "$train_cmd" ${lores_train_data_dir} \
+    data/lang $gmm_dir $lat_dir
+  rm $lat_dir/fsts.*.gz # save space
+fi
+
+if [ $stage -le 12 ]; then
+  # Build a tree using our new topology.  We know we have alignments for the
+  # speed-perturbed data (local/nnet3/run_ivector_common.sh made them), so use
+  # those.  The num-leaves is always somewhat less than the num-leaves from
+  # the GMM baseline.
+   if [ -f $tree_dir/final.mdl ]; then
+     echo "$0: $tree_dir/final.mdl already exists, refusing to overwrite it."
+     exit 1;
+  fi
+  steps/nnet3/chain/build_tree.sh \
+    --frame-subsampling-factor 3 \
+    --context-opts "--context-width=2 --central-position=1" \
+    --cmd "$train_cmd" 3500 ${lores_train_data_dir} \
+    $lang $ali_dir $tree_dir
+fi
+
+
+if [ $stage -le 13 ]; then
+  mkdir -p $dir
+  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)
+
+  tdnn_opts="l2-regularize=0.03 dropout-proportion=0.0 dropout-per-dim-continuous=true"
+  tdnnf_opts="l2-regularize=0.03 dropout-proportion=0.0 bypass-scale=0.66"
+  linear_opts="l2-regularize=0.03 orthonormal-constraint=-1.0"
+  prefinal_opts="l2-regularize=0.03"
+  output_opts="l2-regularize=0.015"
+
+  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-dropout-layer name=tdnn1 $tdnn_opts dim=768
+  tdnnf-layer name=tdnnf2 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=1
+  tdnnf-layer name=tdnnf3 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=1
+  tdnnf-layer name=tdnnf4 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=1
+  tdnnf-layer name=tdnnf5 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=0
+  tdnnf-layer name=tdnnf6 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf7 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf8 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf9 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf10 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf11 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf12 $tdnnf_opts dim=768 bottleneck-dim=96 time-stride=3
+  tdnnf-layer name=tdnnf13 $tdnnf_opts dim=768 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 $prefinal_opts small-dim=192 big-dim=768
+  output-layer name=output include-log-softmax=false dim=$num_targets $output_opts
+
+  # adding the layers for xent branch
+  prefinal-layer name=prefinal-xent input=prefinal-l $prefinal_opts small-dim=192 big-dim=768
+  output-layer name=output-xent dim=$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
+  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/hub4_spanish-$(date +'%m_%d_%H_%M')/s5/$dir/egs/storage $dir/egs/storage
+  fi
+
+  steps/nnet3/chain/train.py --stage=$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" \
+    --trainer.dropout-schedule $dropout_schedule \
+    --trainer.add-option="--optimization.memory-compression-level=2" \
+    --trainer.srand=$srand \
+    --trainer.max-param-change=2.0 \
+    --trainer.num-epochs=10 \
+    --trainer.frames-per-iter=3000000 \
+    --trainer.optimization.num-jobs-initial=2 \
+    --trainer.optimization.num-jobs-final=5 \
+    --trainer.optimization.initial-effective-lrate=0.001 \
+    --trainer.optimization.final-effective-lrate=0.0001 \
+    --trainer.num-chunk-per-minibatch=256,128,64 \
+    --egs.cmd="run.pl --max-jobs-run 12" \
+    --egs.chunk-width=$chunk_width \
+    --egs.chunk-left-context=$chunk_left_context \
+    --egs.chunk-right-context=$chunk_right_context \
+    --egs.chunk-left-context-initial=0 \
+    --egs.chunk-right-context-final=0 \
+    --egs.dir="$common_egs_dir" \
+    --egs.opts="--frames-overlap-per-eg 0" \
+    --cleanup.remove-egs=$remove_egs \
+    --use-gpu=true \
+    --reporting.email="$reporting_email" \
+    --feat-dir=$train_data_dir \
+    --tree-dir=$tree_dir \
+    --lat-dir=$lat_dir \
+    --dir=$dir  || exit 1;
+fi
+
+if [ $stage -le 15 ]; then
+  # Note: it's not important to give mkgraph.sh the lang directory with the
+  # matched topology (since it gets the topology file from the model).
+  utils/mkgraph.sh \
+    --self-loop-scale 1.0 data/langp_test \
+    $tree_dir $dir/graph || exit 1;
+fi
+
+if [ $stage -le 16 ]; then
+  frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+  nspk=$(wc -l <data/eval/spk2utt)
+  steps/nnet3/decode.sh \
+    --acwt 1.0 --post-decode-acwt 10.0 \
+    --frames-per-chunk $frames_per_chunk \
+    --nj $nspk --cmd "$decode_cmd"  --num-threads 4 \
+    --online-ivector-dir exp/nnet3${nnet3_affix}/ivectors_eval_hires \
+    $dir/graph data/eval_hires $dir/decode_test || exit 1
+fi
+
+if [ $stage -le 17 ]; 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} ${dir}_online
+
+    nspk=$(wc -l <data/eval/spk2utt)
+    steps/online/nnet3/decode.sh \
+      --acwt 1.0 --post-decode-acwt 10.0 \
+      --nj $nspk --cmd "$decode_cmd" \
+      $dir/graph  data/eval ${dir}_online/decode_test  || exit 1
+fi
+
+
+exit 0;