[egs] Update RM chain recipe with more recent configuration (kaldi-as…

…r#3237)

egs/rm/s5/RESULTS

@@ -233,10 +233,13 @@ for x in exp/nnet2_online_wsj/nnet_ms_a_smbr_0.00005/1/decode_*; do grep WER $x/
 # current best chain result with TDNN (check local/chain/run_tdnn_5g.sh)
 %WER 2.86 [ 358 / 12533, 46 ins, 61 del, 251 sub ] exp/chain/tdnn_5g/decode/wer_5_0.0
 %WER 2.71 [ 340 / 12533, 58 ins, 59 del, 223 sub ] exp/chain/tdnn_5n/decode/wer_4_0.0
+# This is a modified version of run_tdnn_5n.sh. It uses
+# a new configs convention for chain model after kaldi 5.2.
+%WER 1.52 [ 191 / 12533, 14 ins, 39 del, 138 sub ] exp/chain/tdnn_5o/decode/wer_4_0.5
 
 ### WSJ->RM Transfer learning using chain model ###
 %WER 1.68 [ 210 / 12533, 25 ins, 33 del, 152 sub ] exp/chain/tdnn_wsj_rm_1a/decode/wer_2_0.0
-
+ 
 ### nnet1 results ###
 
 # dnn4b, MFCC,LDA,fMLLR feaures, (Karel - 30.7.2015)

egs/rm/s5/local/chain/run_tdnn.sh

@@ -0,0 +1 @@
+tuning/run_tdnn_5o.sh

egs/rm/s5/local/chain/tuning/run_tdnn_5o.sh

@@ -0,0 +1,179 @@
+#!/bin/bash
+
+# this script is a modified version of run_tdnn_5n.sh. It uses
+# a new configs convention for chain model after kaldi 5.2.
+
+
+
+set -e
+
+# configs for 'chain'
+stage=0
+train_stage=-10
+get_egs_stage=-10
+xent_regularize=0.1
+dir=exp/chain/tdnn_5o
+
+# training options
+num_epochs=12
+initial_effective_lrate=0.005
+final_effective_lrate=0.0005
+max_param_change=2.0
+final_layer_normalize_target=0.5
+num_jobs_initial=2
+num_jobs_final=4
+minibatch_size=128
+frames_per_eg=150
+remove_egs=false
+#common_egs_dir=exp/chain/tdnn_5g/egs/
+common_egs_dir=
+# 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
+# nnet2 setup, and you can skip them by setting "--stage 4" if you have already
+# run those things.
+
+ali_dir=exp/tri3b_ali
+treedir=exp/chain/tri4_5o_tree
+lang=data/lang_chain_5o
+
+local/online/run_nnet2_common.sh --stage $stage || exit 1;
+
+if [ $stage -le 4 ]; then
+  # Get the alignments as lattices (gives the chain training more freedom).
+  # use the same num-jobs as the alignments
+  nj=$(cat exp/tri3b_ali/num_jobs) || exit 1;
+  steps/align_fmllr_lats.sh --nj $nj --cmd "$train_cmd" data/train \
+    data/lang exp/tri3b exp/tri3b_lats
+  rm exp/tri3b_lats/fsts.*.gz # save space
+fi
+
+if [ $stage -le 5 ]; 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 [ $stage -le 6 ]; then
+  # Build a tree using our new topology.
+  steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 \
+    --cmd "$train_cmd" 1200 data/train $lang $ali_dir $treedir
+fi
+
+if [ $stage -le 7 ]; then
+  mkdir -p $dir
+  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)
+  tdnn_opts="l2-regularize=0.01 dropout-proportion=0.0 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=0.005"
+
+  mkdir -p $dir/configs
+  cat <<EOF > $dir/configs/network.xconfig
+  input dim=50 name=ivector
+  input dim=13 name=input
+  # please note that it is important to have input layer with the name=input
+  # as the layer immediately preceding the fixed-affine-layer to enable
+  # the use of short notation for the descriptor
+  fixed-affine-layer name=lda input=Append(-1,0,1,ReplaceIndex(ivector, t, 0)) affine-transform-file=$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=1024
+  tdnnf-layer name=tdnnf2 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=1
+  tdnnf-layer name=tdnnf3 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=1
+  tdnnf-layer name=tdnnf4 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=1
+  tdnnf-layer name=tdnnf5 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=0
+  tdnnf-layer name=tdnnf6 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=3
+  tdnnf-layer name=tdnnf7 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=3
+  tdnnf-layer name=tdnnf8 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=3
+  tdnnf-layer name=tdnnf9 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=3
+  tdnnf-layer name=tdnnf10 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=3
+  tdnnf-layer name=tdnnf11 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=3
+  tdnnf-layer name=tdnnf12 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=3
+  tdnnf-layer name=tdnnf13 $tdnnf_opts dim=1024 bottleneck-dim=128 time-stride=3
+  linear-component name=prefinal-l dim=192 $linear_opts
+  prefinal-layer name=prefinal-chain input=prefinal-l $prefinal_opts big-dim=1024 small-dim=192
+  output-layer name=output include-log-softmax=false dim=$num_targets $output_opts
+  prefinal-layer name=prefinal-xent input=prefinal-l $prefinal_opts big-dim=1024 small-dim=192
+  output-layer name=output-xent dim=$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 8 ]; then
+ steps/nnet3/chain/train.py --stage $train_stage \
+    --cmd "$decode_cmd" \
+    --feat.online-ivector-dir exp/nnet2_online/ivectors \
+    --feat.cmvn-opts "--norm-means=false --norm-vars=false" \
+    --chain.xent-regularize 0.1 \
+    --chain.leaky-hmm-coefficient 0.1 \
+    --chain.l2-regularize 0.00005 \
+    --chain.apply-deriv-weights false \
+    --chain.lm-opts="--num-extra-lm-states=200" \
+    --egs.dir "$common_egs_dir" \
+    --egs.opts "--frames-overlap-per-eg 0" \
+    --egs.chunk-width $frames_per_eg \
+    --trainer.num-chunk-per-minibatch $minibatch_size \
+    --trainer.frames-per-iter 1000000 \
+    --trainer.num-epochs $num_epochs \
+    --trainer.optimization.num-jobs-initial $num_jobs_initial \
+    --trainer.optimization.num-jobs-final $num_jobs_final \
+    --trainer.optimization.initial-effective-lrate $initial_effective_lrate \
+    --trainer.optimization.final-effective-lrate $final_effective_lrate \
+    --trainer.max-param-change $max_param_change \
+    --cleanup.remove-egs $remove_egs \
+    --feat-dir data/train \
+    --tree-dir $treedir \
+    --lat-dir exp/tri3b_lats \
+    --dir $dir
+fi
+
+if [ $stage -le 9 ]; then
+  steps/online/nnet2/extract_ivectors_online.sh --cmd "$train_cmd" --nj 4 \
+    data/test exp/nnet2_online/extractor exp/nnet2_online/ivectors_test || exit 1;
+fi
+
+if [ $stage -le 10 ]; 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 data/lang $dir $dir/graph
+  steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+    --scoring-opts "--min-lmwt 1" \
+    --nj 20 --cmd "$decode_cmd" \
+    --online-ivector-dir exp/nnet2_online/ivectors_test \
+    $dir/graph data/test $dir/decode || exit 1;
+fi
+
+if [ $stage -le 11 ]; then
+  utils/mkgraph.sh --self-loop-scale 1.0 data/lang_ug $dir $dir/graph_ug
+  steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+    --nj 20 --cmd "$decode_cmd" \
+    --online-ivector-dir exp/nnet2_online/ivectors_test \
+    $dir/graph_ug data/test $dir/decode_ug || exit 1;
+fi
+wait;
+exit 0;