Iam

egs/iam/v2/local/chain/compare_wer.sh

@@ -50,6 +50,36 @@ for x in $*; do
 done
 echo
 
+echo -n "# WER val "
+for x in $*; do
+ wer=$(cat $x/decode_val/scoring_kaldi/best_wer | awk '{print $2}')
+ printf "% 10s" $wer
+done
+echo
+
+echo -n "# WER (rescored) val "
+for x in $*; do
+ wer="--"
+ [ -d $x/decode_val_rescored ] && wer=$(cat $x/decode_test_rescored/scoring_kaldi/best_wer | awk '{print $2}')
+ printf "% 10s" $wer
+done
+echo
+
+echo -n "# CER val "
+for x in $*; do
+ cer=$(cat $x/decode_val/scoring_kaldi/best_cer | awk '{print $2}')
+ printf "% 10s" $cer
+done
+echo
+
+echo -n "# CER (rescored) val "
+for x in $*; do
+ cer="--"
+ [ -d $x/decode_val_rescored ] && cer=$(cat $x/decode_test_rescored/scoring_kaldi/best_cer | awk '{print $2}')
+ printf "% 10s" $cer
+done
+echo
+
 if $used_epochs; then
  exit 0; # the diagnostics aren't comparable between regular and discriminatively trained systems.
 fi

egs/iam/v2/local/chain/tuning/run_cnn_e2eali_1a.sh

@@ -22,6 +22,7 @@ stage=0
 
 nj=30
 train_set=train
+decode_val=true
 nnet3_affix= # affix for exp dirs, e.g. it was _cleaned in tedlium.
 affix=_1a #affix for TDNN+LSTM directory e.g. "1a" or "1b", in case we change the configuration.
 e2echain_model_dir=exp/chain/e2e_cnn_1a
@@ -243,3 +244,21 @@ if [ $stage -le 7 ]; then
  --nj $nj --cmd "$cmd" \
  $dir/graph data/test $dir/decode_test || exit 1;
 fi
+
+if [ $stage -le 8 ] && $decode_val; then
+ frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+ steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+ --extra-left-context $chunk_left_context \
+ --extra-right-context $chunk_right_context \
+ --extra-left-context-initial 0 \
+ --extra-right-context-final 0 \
+ --frames-per-chunk $frames_per_chunk \
+ --nj $nj --cmd "$cmd" \
+ $dir/graph data/val $dir/decode_val || exit 1;
+
+ steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
+ data/val $dir/decode_val{,_rescored} || exit 1
+fi
+
+echo "Done. Date: $(date). Results:"
+local/chain/compare_wer.sh $dir

egs/iam/v2/local/chain/tuning/run_cnn_e2eali_1b.sh

@@ -23,6 +23,7 @@ stage=0
 
 nj=30
 train_set=train
+decode_val=true
 nnet3_affix= # affix for exp dirs, e.g. it was _cleaned in tedlium.
 affix=_1b #affix for TDNN+LSTM directory e.g. "1a" or "1b", in case we change the configuration.
 e2echain_model_dir=exp/chain/e2e_cnn_1a
@@ -249,3 +250,21 @@ if [ $stage -le 7 ]; then
  steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
  data/test $dir/decode_test{,_rescored} || exit 1
 fi
+
+if [ $stage -le 8 ] && $decode_val; then
+ frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+ steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+ --extra-left-context $chunk_left_context \
+ --extra-right-context $chunk_right_context \
+ --extra-left-context-initial 0 \
+ --extra-right-context-final 0 \
+ --frames-per-chunk $frames_per_chunk \
+ --nj $nj --cmd "$cmd" \
+ $dir/graph data/val $dir/decode_val || exit 1;
+
+ steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
+ data/val $dir/decode_val{,_rescored} || exit 1
+fi
+
+echo "Done. Date: $(date). Results:"
+local/chain/compare_wer.sh $dir

egs/iam/v2/local/chain/tuning/run_cnn_e2eali_1c.sh

@@ -25,6 +25,7 @@ stage=0
 
 nj=30
 train_set=train
+decode_val=true
 nnet3_affix= # affix for exp dirs, e.g. it was _cleaned in tedlium.
 affix=_1c #affix for TDNN+LSTM directory e.g. "1a" or "1b", in case we change the configuration.
 e2echain_model_dir=exp/chain/e2e_cnn_1a
@@ -251,3 +252,21 @@ if [ $stage -le 7 ]; then
  steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
  data/test $dir/decode_test{,_rescored} || exit 1
 fi
+
+if [ $stage -le 8 ] && $decode_val; then
+ frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+ steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+ --extra-left-context $chunk_left_context \
+ --extra-right-context $chunk_right_context \
+ --extra-left-context-initial 0 \
+ --extra-right-context-final 0 \
+ --frames-per-chunk $frames_per_chunk \
+ --nj $nj --cmd "$cmd" \
+ $dir/graph data/val $dir/decode_val || exit 1;
+
+ steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
+ data/val $dir/decode_val{,_rescored} || exit 1
+fi
+
+echo "Done. Date: $(date). Results:"
+local/chain/compare_wer.sh $dir

egs/iam/v2/local/chain/tuning/run_cnn_e2eali_1d.sh

@@ -21,7 +21,7 @@ stage=0
 
 nj=30
 train_set=train
-test_dir=data/test
+decode_val=true
 nnet3_affix= # affix for exp dirs, e.g. it was _cleaned in tedlium.
 affix=_1d #affix for TDNN+LSTM directory e.g. "1a" or "1b", in case we change the configuration.
 e2echain_model_dir=exp/chain/e2e_cnn_1b
@@ -244,10 +244,25 @@ if [ $stage -le 7 ]; then
  --extra-right-context-final 0 \
  --frames-per-chunk $frames_per_chunk \
  --nj $nj --cmd "$cmd" \
- $dir/graph $test_dir $dir/decode_test || exit 1;
+ $dir/graph data/test $dir/decode_test || exit 1;
 
  steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
- $test_dir $dir/decode_test{,_rescored} || exit 1
+ data/test $dir/decode_test{,_rescored} || exit 1
+fi
+
+if [ $stage -le 8 ] && $decode_val; then
+ frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+ steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+ --extra-left-context $chunk_left_context \
+ --extra-right-context $chunk_right_context \
+ --extra-left-context-initial 0 \
+ --extra-right-context-final 0 \
+ --frames-per-chunk $frames_per_chunk \
+ --nj $nj --cmd "$cmd" \
+ $dir/graph data/val $dir/decode_val || exit 1;
+
+ steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
+ data/val $dir/decode_val{,_rescored} || exit 1
 fi
 
 echo "Done. Date: $(date). Results:"

egs/iam/v2/local/chain/tuning/run_e2e_cnn_1a.sh

@@ -25,6 +25,7 @@ stage=0
 train_stage=-10
 get_egs_stage=-10
 affix=1a
+nj=30
 
 # training options
 tdnn_dim=450
@@ -37,6 +38,7 @@ l2_regularize=0.00005
 frames_per_iter=1000000
 cmvn_opts="--norm-means=true --norm-vars=true"
 train_set=train
+decode_val=true
 lang_decode=data/lang
 lang_rescore=data/lang_rescore_6g
 
@@ -163,12 +165,22 @@ fi
 if [ $stage -le 5 ]; then
  frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
  steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
- --nj 30 --cmd "$cmd" \
+ --nj $nj --cmd "$cmd" \
  $dir/graph data/test $dir/decode_test || exit 1;
 
  steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
  data/test $dir/decode_test{,_rescored} || exit 1
 fi
 
+if [ $stage -le 6 ] && $decode_val; then
+ frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+ steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+ --nj $nj --cmd "$cmd" \
+ $dir/graph data/val $dir/decode_val || exit 1;
+
+ steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
+ data/val $dir/decode_val{,_rescored} || exit 1
+fi
+
 echo "Done. Date: $(date). Results:"
 local/chain/compare_wer.sh $dir

egs/iam/v2/local/chain/tuning/run_e2e_cnn_1b.sh

@@ -23,12 +23,14 @@ stage=0
 train_stage=-10
 get_egs_stage=-10
 affix=1b
+nj=30
 
 # training options
 tdnn_dim=450
 minibatch_size=150=100,64/300=50,32/600=25,16/1200=16,8
 common_egs_dir=
 train_set=train
+decode_val=true
 lang_decode=data/lang
 lang_rescore=data/lang_rescore_6g
 
@@ -149,12 +151,22 @@ fi
 if [ $stage -le 5 ]; then
  frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
  steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
- --nj 30 --cmd "$cmd" \
+ --nj $nj --cmd "$cmd" \
  $dir/graph data/test $dir/decode_test || exit 1;
 
  steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
  data/test $dir/decode_test{,_rescored} || exit 1
 fi
 
+if [ $stage -le 6 ] && $decode_val; then
+ frames_per_chunk=$(echo $chunk_width | cut -d, -f1)
+ steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+ --nj $nj --cmd "$cmd" \
+ $dir/graph data/val $dir/decode_val || exit 1;
+
+ steps/lmrescore_const_arpa.sh --cmd "$cmd" $lang_decode $lang_rescore \
+ data/val $dir/decode_val{,_rescored} || exit 1
+fi
+
 echo "Done. Date: $(date). Results:"
 local/chain/compare_wer.sh $dir