trunk: various minor script extensions (supporting max-count option in iVector extraction script)

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@4870 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8

egs/wsj/s5/steps/online/nnet2/extract_ivectors_online.sh

@@ -34,6 +34,13 @@ posterior_scale=0.1 # Scale on the acoustic posteriors, intended to account for
                     # with the neural nets trained with these iVectors.
 compress=true       # If true, compress the iVectors stored on disk (it's lossy
                     # compression, as used for feature matrices).
+max_count=0         # The use of this option (e.g. --max-count 100) can make
+                    # iVectors more consistent for different lengths of
+                    # utterance, by scaling up the prior term when the
+                    # data-count exceeds this value.  The data-count is after
+                    # posterior-scaling, so assuming the posterior-scale is 0.1,
+                    # --max-count 100 starts having effect after 1000 frames, or
+                    # 10 seconds of data.
 
 # End configuration section.
 
@@ -95,6 +102,7 @@ echo "--num-gselect=$num_gselect"  >>$ieconf
 echo "--min-post=$min_post" >>$ieconf
 echo "--posterior-scale=$posterior_scale" >>$ieconf
 echo "--max-remembered-frames=1000" >>$ieconf # the default
+echo "--max-count=$max_count" >>$ieconf
 
 
 

egs/wsj/s5/steps/online/nnet2/prepare_online_decoding.sh

@@ -23,6 +23,12 @@ posterior_scale=0.1 # Scale on the acoustic posteriors, intended to account for
 min_post=0.025 # Minimum posterior to use (posteriors below this are pruned out)
                # caution: you should use the same value in the online-estimation
                # code.
+max_count=100   # This max-count of 100 can make iVectors more consistent for
+                # different lengths of utterance, by scaling up the prior term
+                # when the data-count exceeds this value.  The data-count is
+                # after posterior-scaling, so assuming the posterior-scale is
+                # 0.1, --max-count 100 starts having effect after 1000 frames,
+                # or 10 seconds of data.
 iter=final
 # End configuration.
 
@@ -138,6 +144,7 @@ if [ ! -z "$iedir" ]; then
   echo "--min-post=$min_post" >>$ieconf
   echo "--posterior-scale=$posterior_scale" >>$ieconf # this is currently the default in the scripts.
   echo "--max-remembered-frames=1000" >>$ieconf # the default
+  echo "--max-count=$max_count" >>$ieconf
 fi
 
 if $add_pitch; then

egs/wsj/s5/steps/select_feats.sh

@@ -49,6 +49,12 @@ rm $data/feats.scp 2>/dev/null
 # use "name" as part of name of the archive.
 name=`basename $data`
 
+for j in $(seq $nj); do 
+  # the next command does nothing unless $mfccdir/storage/ exists, see
+  # utils/create_data_link.pl for more info.
+  utils/create_data_link.pl $ark_dir/pasted_$name.$j.ark
+done
+
 $cmd JOB=1:$nj $logdir/append.JOB.log \
    select-feats "$selector" scp:$data_in/split$nj/JOB/feats.scp ark:- \| \
    copy-feats --compress=$compress ark:- \