modified run_common_lang.sh to used make_mfcc{,_pitch}.sh instead of …

…make_pitch.sh.

egs/babel_multilang/s5/local/nnet3/extract_ivector_lang.sh

@@ -20,7 +20,7 @@ ivector_suffix=_gb
 lang=$1
 global_extractor=$2
 
-if [ $stage -le 8 ]; then
+if [ $stage -le 7 ]; then
   # We extract iVectors on all the train_nodup data, which will be what we
   # train the system on.
   # having a larger number of speakers is helpful for generalization, and to

egs/babel_multilang/s5/local/nnet3/run_common_langs.sh

@@ -59,57 +59,39 @@ if [ "$speed_perturb" == "true" ]; then
 fi
 
 if [ $stage -le 3 ] && [ ! -f data/$lang/${train_set}_hires/.done ]; then
+  hires_config="--mfcc-config conf/mfcc_hires.conf"
+  feat_suffix=_hires_mfcc
   mfccdir=mfcc_hires/$lang
+  if $use_pitch; then
+    feat_suffix=${feat_suffix}_pitch
+    hires_config="$hires_configs --pitch-config $pitch_conf"
+    mfccdir=${mfccdir}_pitch/$lang
+  fi
+
   if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $mfccdir/storage ]; then
     date=$(date +'%m_%d_%H_%M')
     utils/create_split_dir.pl /export/b0{1,2,3,4}/$USER/kaldi-data/egs/$lang-$date/s5c/$mfccdir/storage $mfccdir/storage
   fi
 
+
   for dataset in $train_set ; do
-    utils/copy_data_dir.sh data/$lang/$dataset data/$lang/${dataset}_hires
+    data_dir=data/$lang/${dataset}${feat_suffix}
+    log_dir=exp/$lang/make${feat_suffix}/$dataset
 
-    # scale the waveforms, this is useful as we don't use CMVN
-    data_dir=data/$lang/${dataset}_hires
+    utils/copy_data_dir.sh data/$lang/$dataset ${data_dir} || exit 1;
 
-    utils/data/perturb_data_dir_volume.sh $data_dir || exit 1 ;
+    # scale the waveforms, this is useful as we don't use CMVN
+    utils/data/perturb_data_dir_volume.sh $data_dir || exit 1;
 
-    steps/make_mfcc.sh --nj 70 --mfcc-config conf/mfcc_hires.conf \
-      --cmd "$train_cmd" data/$lang/${dataset}_hires exp/$lang/make_hires/$dataset $mfccdir;
+    steps/make${feat_suffix}.sh --nj 70 $hires_config \
+      --cmd "$train_cmd" ${data_dir} $log_dir $mfccdir;
 
-    steps/compute_cmvn_stats.sh data/$lang/${dataset}_hires exp/$lang/make_hires/${dataset} $mfccdir;
+    steps/compute_cmvn_stats.sh ${data_dir} $log_dir $mfccdir;
 
     # Remove the small number of utterances that couldn't be extracted for some
     # reason (e.g. too short; no such file).
-    utils/fix_data_dir.sh data/$lang/${dataset}_hires;
+    utils/fix_data_dir.sh ${data_dir};
   done
-  touch data/$lang/${train_set}_hires/.done
+  touch ${data_dir}/.done
 fi
-
-if [ $stage -le 4 ]; then
-  if [[ "$use_pitch" == "true" ]]; then
-    pitchdir=pitch/$lang
-    train_set=${train_set}_hires
-    for dataset in $train_set; do
-      if $use_pitch; then
-        mkdir -p $pitchdir
-        if [ ! -f data/$lang/${dataset}_pitch/feats.scp ]; then
-          echo "$0: Generating pitch features for data/$lang as use_pitch=$use_pitch"
-          utils/copy_data_dir.sh data/$lang/$dataset data/$lang/${dataset}_pitch
-          steps/make_pitch.sh --nj 70 --pitch-config $pitch_conf \
-            --cmd "$train_cmd" data/$lang/${dataset}_pitch exp/$lang/make_pitch/${dataset} $pitchdir;
-        fi
-        feat_suffix=_pitch
-      fi
-
-      if [ ! -f data/$lang/${dataset}_mfcc${feat_suffix}/feats.scp ]; then
-        steps/append_feats.sh --nj 16 --cmd "$train_cmd" data/$lang/${dataset} \
-          data/$lang/${dataset}${feat_suffix} data/$lang/${dataset}_mfcc${feat_suffix} \
-          exp/$lang/append_mfcc${feat_suffix}/${dataset} mfcc${feat_suffix}/$lang
-
-        steps/compute_cmvn_stats.sh data/$lang/${dataset}_mfcc${feat_suffix} exp/$lang/make_cmvn_mfcc${feat_suffix}/${x} mfcc${feat_suffix}/$lang
-      fi
-    done
-  fi
-fi
-
 exit 0;

egs/babel_multilang/s5/local/nnet3/run_shared_ivector_extractor.sh

@@ -38,7 +38,7 @@ langconf=conf/$lda_mllt_lang/lang.conf
 [ ! -f $langconf ] && echo 'Language configuration does not exist! Use the configurations in conf/lang/* as a startup' && exit 1
 . $langconf || exit 1;
 
-if [ $stage -le 5 ]; then
+if [ $stage -le 4 ]; then
   # We need to build a small system just because we need the LDA+MLLT transform
   # to train the diag-UBM on top of.  We use --num-iters 13 because after we get
   # the transform (12th iter is the last), any further training is pointless.
@@ -51,13 +51,13 @@ if [ $stage -le 5 ]; then
     data/$lda_mllt_lang/lang exp/$lda_mllt_lang/tri5_ali${suffix} exp/$lda_mllt_lang/nnet3/tri3b
 fi
 
-if [ $stage -le 6 ]; then
+if [ $stage -le 5 ]; then
   # To train a diagonal UBM we don't need very much data, so use the smallest subset.
   steps/online/nnet2/train_diag_ubm.sh --cmd "$train_cmd" --nj 200 --num-frames 500000 \
     $multi_data_dir $numGaussUBM exp/$lda_mllt_lang/nnet3/tri3b $global_extractor_dir/diag_ubm
 fi
 
-if [ $stage -le 7 ]; then
+if [ $stage -le 6 ]; then
   # iVector extractors can be sensitive to the amount of data, but this one has a
   # fairly small dim (defaults to 100) so we don't use all of it, we use just the
   # 100k subset (just under half the data).

egs/babel_multilang/s5/local/nnet3/run_tdnn_multilingual.sh

@@ -155,7 +155,7 @@ else
 fi
 feat_dim=`feat-to-dim scp:${multi_data_dirs[0]}/feats.scp -`
 
-if [ $stage -le 9 ]; then
+if [ $stage -le 8 ]; then
   echo "$0: creating multilingual neural net configs using the xconfig parser";
   if [ -z $bnf_dim ]; then
     bnf_dim=1024
@@ -202,7 +202,7 @@ EOF
   nnet3-copy --edits="remove-output-nodes name=output-tmp" $dir/configs/ref.raw $dir/configs/ref.raw || exit 1;
 fi
 
-if [ $stage -le 10 ]; then
+if [ $stage -le 9 ]; then
   echo "$0: Generates separate egs dir per language for multilingual training."
   # sourcing the "vars" below sets
   #model_left_context=(something)
@@ -225,7 +225,7 @@ if [ -z $megs_dir ];then
   megs_dir=$dir/egs
 fi
 
-if [ $stage -le 11 ] && [ ! -z $megs_dir ]; then
+if [ $stage -le 10 ] && [ ! -z $megs_dir ]; then
   echo "$0: Generate multilingual egs dir using "
   echo "separate egs dirs for multilingual training."
   if [ ! -z "$lang2weight" ]; then
@@ -238,7 +238,7 @@ if [ $stage -le 11 ] && [ ! -z $megs_dir ]; then
     $num_langs ${common_egs_dir[@]} || exit 1;
 fi
 
-if [ $stage -le 12 ]; then
+if [ $stage -le 11 ]; then
   steps/nnet3/train_raw_dnn.py --stage=$train_stage \
     --cmd="$decode_cmd" \
     --feat.cmvn-opts="--norm-means=false --norm-vars=false" \
@@ -262,7 +262,7 @@ if [ $stage -le 12 ]; then
     --dir=$dir  || exit 1;
 fi
 
-if [ $stage -le 13 ]; then
+if [ $stage -le 12 ]; then
   for lang_index in `seq 0 $[$num_langs-1]`;do
     lang_dir=$dir/${lang_list[$lang_index]}
     mkdir -p  $lang_dir
@@ -282,7 +282,7 @@ if [ $stage -le 13 ]; then
 fi
 
 # decoding different languages
-if [ $stage -le 14 ]; then
+if [ $stage -le 13 ]; then
   num_decode_lang=${#decode_lang_list[@]}
   for lang_index in `seq 0 $[$num_decode_lang-1]`; do
     if [ ! -f $dir/${decode_lang_list[$lang_index]}/decode_dev10h.pem/.done ]; then