Removing little-used feature: time-reversed, and forward-backward, decoding.

5dfa20aa · Daniel Povey · f0fab215 · 5dfa20aa · f0fab215 · 5dfa20aa
Commit 5dfa20aa authored 8 years ago by Daniel Povey
--- a/egs/aspire/s5/local/score.sh
+++ b/egs/aspire/s5/local/score.sh
@@ -5,7 +5,6 @@
 cmd=run.pl
 min_lmwt=5
 max_lmwt=17
-reverse=false
 #end configuration section.

 [ -f ./path.sh ] && . ./path.sh
@@ -16,7 +15,6 @@ if [ $# -ne 3 ]; then
  echo "    --cmd (run.pl|queue.pl...)      # specify how to run the sub-processes."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -26,7 +24,7 @@ dir=$3

 model=$dir/../final.mdl # assume model one level up from decoding dir.

-hubscr=$KALDI_ROOT/tools/sctk/bin/hubscr.pl 
+hubscr=$KALDI_ROOT/tools/sctk/bin/hubscr.pl
 [ ! -f $hubscr ] && echo "Cannot find scoring program at $hubscr" && exit 1;
 hubdir=`dirname $hubscr`

@@ -39,14 +37,14 @@ name=`basename $data`; # e.g. eval2000
 mkdir -p $dir/scoring/log

 function filter_text {
-  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; } 
+  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; }
   while(<STDIN>) { @A  = split(" ", $_); $id = shift @A; print "$id ";
     foreach $a (@A) { if (!defined $bad{$a}) { print "$a "; }} print "\n"; }' \
   '[noise]' '[laughter]' '[vocalized-noise]' '<unk>' '%hesitation'
 }

 #function filter_text {
-#  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; } 
+#  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; }
 #   while(<STDIN>) { @A  = split(" ", $_); $id = shift @A; print "$id ";
 #     foreach $a (@A) { if (!defined $bad{$a}) { print "$a "; }} print "\n"; }' \
 #   '[NOISE]' '[LAUGHTER]' '[VOCALIZED-NOISE]' '<UNK>' '%HESITATION'
@@ -59,11 +57,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
 for lmwt in `seq $min_lmwt $max_lmwt`; do
  utils/int2sym.pl -f 2- $lang/words.txt <$dir/scoring/$lmwt.tra | \
   filter_text > $dir/scoring/$lmwt.txt || exit 1;
-  if $reverse; then
-    mv $dir/scoring/$lmwt.txt $dir/scoring/$lmwt.txt.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.txt.orig >$dir/scoring/$lmwt.txt
-  fi
 done

 filter_text <$data/text >$dir/scoring/text.filt

--- a/egs/aurora4/s5/local/run_fwdbwd.sh
+++ b/egs/aurora4/s5/local/run_fwdbwd.sh
-#prepare reverse lexicon and language model for backwards decoding
-utils/prepare_lang.sh --reverse true data/local/dict "<SPOKEN_NOISE>" data/local/lang_tmp.reverse data/lang.reverse || exit 1;
-utils/reverse_lm.sh data/local/nist_lm/lm_bg_5k.arpa.gz data/lang.reverse data/lang_test_bg_5k.reverse || exit 1;
-utils/reverse_lm_test.sh data/lang_test_bg_5k data/lang_test_bg_5k.reverse || exit 1;
-
-# normal forward decoding
-utils/mkgraph.sh data/lang_test_bg_5k exp/tri2a exp/tri2a/graph_bg5k
-steps/decode_fwdbwd.sh --beam 10.0 --latbeam 4.0 --nj 8 --cmd "$decode_cmd" \
-  exp/tri2a/graph_bg5k data/test_eval92 exp/tri2a/decode_eval92_bg5k_10 || exit 1;
-
-# backward decoding
-utils/mkgraph.sh --reverse data/lang_test_bg_5k.reverse exp/tri2a exp/tri2a/graph_bg5k_r
-steps/decode_fwdbwd.sh --beam 10.0 --latbeam 4.0 --reverse true --nj 8 --cmd "$decode_cmd" \
-  exp/tri2a/graph_bg5k_r data/test_eval92 exp/tri2a/decode_eval92_bg5k_reverse10 || exit 1;
-
-# pingpong decoding
-steps/decode_fwdbwd.sh --beam 10.0 --max-beam 20.0 --reverse true --nj 8 --cmd "$decode_cmd" \
-  --first_pass exp/tri2a/decode_eval92_bg5k_10 exp/tri2a/graph_bg5k_r data/test_eval92 exp/tri2a/decode_eval92_bg5k_pingpong10 || exit 1;
-steps/decode_fwdbwd.sh --beam 10.0 --max-beam 20.0 --nj 8 --cmd "$decode_cmd" \
-  --first_pass exp/tri2a/decode_eval92_bg5k_reverse10 exp/tri2a/graph_bg5k data/test_eval92 exp/tri2a/decode_eval92_bg5k_pongping10 || exit 1;
-
-# same for bigger language models (on machine with 8GB RAM, you can run the whole decoding in 3-4 min without SGE)
-utils/prepare_lang.sh --reverse true data/local/dict_larger "<SPOKEN_NOISE>" data/local/lang_larger.reverse data/lang_bd.reverse || exit;
-utils/reverse_lm.sh --lexicon data/local/dict_larger/lexicon.txt data/local/local_lm/3gram-mincount/lm_pr6.0.gz data/lang_bd.reverse data/lang_test_bd_tgpr.reverse || exit 1;
-utils/reverse_lm_test.sh data/lang_test_bd_tgpr data/lang_test_bd_tgpr.reverse || exit 1;
-
-utils/mkgraph.sh data/lang_test_bd_tgpr exp/tri2a exp/tri2a/graph_bd_tgpr
-steps/decode_fwdbwd.sh --beam 10.0 --latbeam 4.0 --nj 4 --cmd run.pl \
-  exp/tri2a/graph_bd_tgpr data/test_eval92 exp/tri2a/decode_eval92_bdtgpr4_10 || exit 1;
-
-utils/mkgraph.sh --reverse data/lang_test_bd_tgpr.reverse exp/tri2a exp/tri2a/graph_bd_tgpr_r
-steps/decode_fwdbwd.sh --beam 10.0 --latbeam 4.0 --reverse true --nj 4 --cmd run.pl \
-  exp/tri2a/graph_bd_tgpr_r data/test_eval92 exp/tri2a/decode_eval92_bdtgpr4_reverse10 || exit 1;
-
-steps/decode_fwdbwd.sh --beam 10.0 --max-beam 20.0 --reverse true --nj 4 --cmd run.pl \
-  --first_pass exp/tri2a/decode_eval92_bdtgpr4_10 exp/tri2a/graph_bd_tgpr_r data/test_eval92 \
-  exp/tri2a/decode_eval92_bdtgpr4_pingpong10 || exit 1;
-
-steps/decode_fwdbwd.sh --beam 10.0 --max-beam 20.0 --nj 4 --cmd run.pl \
-  --first_pass exp/tri2a/decode_eval92_bdtgpr4_reverse10 exp/tri2a/graph_bd_tgpr data/test_eval92 \
-  exp/tri2a/decode_eval92_bdtgpr4_pongping10 || exit 1;
--- a/egs/aurora4/s5/local/score.sh
+++ b/egs/aurora4/s5/local/score.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0
 min_lmwt=7
 max_lmwt=17
@@ -25,7 +24,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -49,14 +47,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --word-symbol-table=$symtab \
    ark:- ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/babel/s5/local/score_map.sh
+++ b/egs/babel/s5/local/score_map.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 min_lmwt=7
 max_lmwt=17
 #end configuration section.
@@ -24,7 +23,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -46,14 +44,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --lm-scale=LMWT --word-symbol-table=$symtab \
    "ark:gunzip -c $dir/lat.*.gz|" ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/babel/s5b/local/score_map.sh
+++ b/egs/babel/s5b/local/score_map.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 min_lmwt=7
 max_lmwt=17
 #end configuration section.
@@ -24,7 +23,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -46,14 +44,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --lm-scale=LMWT --word-symbol-table=$symtab \
    "ark:gunzip -c $dir/lat.*.gz|" ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/babel/s5c/local/score_map.sh
+++ b/egs/babel/s5c/local/score_map.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 min_lmwt=7
 max_lmwt=17
 #end configuration section.
@@ -24,7 +23,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -46,14 +44,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --lm-scale=LMWT --word-symbol-table=$symtab \
    "ark:gunzip -c $dir/lat.*.gz|" ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/chime2/s5/local/score.sh
+++ b/egs/chime2/s5/local/score.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0
 min_lmwt=5
 max_lmwt=17
@@ -25,7 +24,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -49,14 +47,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --word-symbol-table=$symtab \
    ark:- ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/chime3/s5/local/score.sh
+++ b/egs/chime3/s5/local/score.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0
 min_lmwt=5
 max_lmwt=17
@@ -25,7 +24,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -49,14 +47,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --word-symbol-table=$symtab \
    ark:- ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/chime4/s5_1ch/local/score.sh
+++ b/egs/chime4/s5_1ch/local/score.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0
 min_lmwt=5
 max_lmwt=17
@@ -25,7 +24,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -49,14 +47,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --word-symbol-table=$symtab \
    ark:- ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/csj/s5/local/score_csj.sh
+++ b/egs/csj/s5/local/score_csj.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 min_lmwt=5
 max_lmwt=17
-reverse=false
 #end configuration section.

 [ -f ./path.sh ] && . ./path.sh
@@ -20,7 +19,6 @@ if [ $# -ne 3 ]; then
  echo "    --cmd (run.pl|queue.pl...)      # specify how to run the sub-processes."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -30,7 +28,7 @@ dir=$3

 model=$dir/../final.mdl # assume model one level up from decoding dir.

-hubscr=$KALDI_ROOT/tools/sctk/bin/hubscr.pl 
+hubscr=$KALDI_ROOT/tools/sctk/bin/hubscr.pl
 #hubscr=$KALDI_ROOT/tools/sctk-2.4.0/bin/hubscr.pl
 [ ! -f $hubscr ] && echo "Cannot find scoring program at $hubscr" && exit 1;
 hubdir=`dirname $hubscr`
@@ -43,19 +41,19 @@ name=`basename $data`; # e.g. eval1

 mkdir -p $dir/scoring/log
 mkdir -p $dir/label
-mkdir -p $dir/label/log 
+mkdir -p $dir/label/log
 mkdir -p $dir/label/wer

 function filter_text_mor {
-  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; } 
+  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; }
   while(<STDIN>) { @A  = split(" ", $_); $id = shift @A; print "$id ";
     foreach $a (@A) { if (!defined $bad{$a}) { print "$a "; }} print "\n"; }' \
   '<UNK>'
 }

 function filter_text {
-  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; }                                                                                       
-   while(<STDIN>) { @A  = split(" ", $_); $id = shift @A; print "$id ";                                                                                                     
+  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; }
+   while(<STDIN>) { @A  = split(" ", $_); $id = shift @A; print "$id ";
     foreach $a (@A) { if (!defined $bad{$a}){ @W=split(/\+/,$a); $word=$W[0]; { print "$word "; }}} print "\n"; }' \
   '<UNK>'
 }
@@ -68,14 +66,9 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
 for lmwt in `seq $min_lmwt $max_lmwt`; do
  utils/int2sym.pl -f 2- $lang/words.txt <$dir/scoring/$lmwt.tra | \
   filter_text > $dir/scoring/$lmwt.txt || exit 1;
-  
+
  utils/int2sym.pl -f 2- $lang/words.txt <$dir/scoring/$lmwt.tra | \
   filter_text_mor > $dir/label/${lmwt}-trans.text || exit 1;
-  if $reverse; then
-    mv $dir/scoring/$lmwt.txt $dir/scoring/$lmwt.txt.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.txt.orig >$dir/scoring/$lmwt.txt
-  fi
 done

 filter_text <$data/text >$dir/scoring/text.filt

--- a/egs/fisher_english/s5/local/score.sh
+++ b/egs/fisher_english/s5/local/score.sh
@@ -5,7 +5,6 @@
 cmd=run.pl
 min_lmwt=5
 max_lmwt=17
-reverse=false
 #end configuration section.

 [ -f ./path.sh ] && . ./path.sh
@@ -17,7 +16,6 @@ if [ $# -ne 3 ]; then
  echo "    --cmd (run.pl|queue.pl...)      # specify how to run the sub-processes."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -37,7 +35,7 @@ mkdir -p $dir/scoring/log


 function filter_text {
-  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; } 
+  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; }
   while(<STDIN>) { @A  = split(" ", $_); $id = shift @A; print "$id ";
     foreach $a (@A) { if (!defined $bad{$a}) { print "$a "; }} print "\n"; }' \
   '[NOISE]' '[LAUGHTER]' '[VOCALIZED-NOISE]' '<UNK>' '%HESITATION'
@@ -50,11 +48,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
 for lmwt in `seq $min_lmwt $max_lmwt`; do
  utils/int2sym.pl -f 2- $lang/words.txt <$dir/scoring/$lmwt.tra | \
   filter_text > $dir/scoring/$lmwt.txt || exit 1;
-  if $reverse; then
-    mv $dir/scoring/$lmwt.txt $dir/scoring/$lmwt.txt.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.txt.orig >$dir/scoring/$lmwt.txt
-  fi
 done

 filter_text <$data/text >$dir/scoring/text.filt

--- a/egs/fisher_swbd/s5/local/score.sh
+++ b/egs/fisher_swbd/s5/local/score.sh
@@ -12,7 +12,6 @@ cmd=run.pl
 stage=0
 min_lmwt=5
 max_lmwt=17
-reverse=false
 word_ins_penalty=0.0,0.5,1.0
 #end configuration section.

@@ -26,7 +25,6 @@ if [ $# -ne 3 ]; then
  echo "    --stage (0|1|2)                 # start scoring script from part-way through."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi


--- a/egs/fisher_swbd/s5/local/score_basic.sh
+++ b/egs/fisher_swbd/s5/local/score_basic.sh
@@ -5,7 +5,6 @@
 cmd=run.pl
 min_lmwt=5
 max_lmwt=17
-reverse=false
 word_ins_penalty=0.0,0.5,1.0
 #end configuration section.

@@ -18,7 +17,6 @@ if [ $# -ne 3 ]; then
  echo "    --cmd (run.pl|queue.pl...)      # specify how to run the sub-processes."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -28,7 +26,7 @@ dir=$3

 model=$dir/../final.mdl # assume model one level up from decoding dir.

-hubscr=$KALDI_ROOT/tools/sctk/bin/hubscr.pl 
+hubscr=$KALDI_ROOT/tools/sctk/bin/hubscr.pl
 [ ! -f $hubscr ] && echo "Cannot find scoring program at $hubscr" && exit 1;
 hubdir=`dirname $hubscr`

@@ -42,7 +40,7 @@ mkdir -p $dir/scoring/log


 function filter_text {
-  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; } 
+  perl -e 'foreach $w (@ARGV) { $bad{$w} = 1; }
   while(<STDIN>) { @A  = split(" ", $_); $id = shift @A; print "$id ";
     foreach $a (@A) { if (!defined $bad{$a}) { print "$a "; }} print "\n"; }' \
   '[NOISE]' '[LAUGHTER]' '[VOCALIZED-NOISE]' '<UNK>' '%HESITATION'
@@ -58,11 +56,6 @@ for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do
  for lmwt in `seq $min_lmwt $max_lmwt`; do
    utils/int2sym.pl -f 2- $lang/words.txt <$dir/scoring/$lmwt.${wip}.tra | \
     filter_text > $dir/scoring/$lmwt.${wip}.txt || exit 1;
-    if $reverse; then
-      mv $dir/scoring/$lmwt.${wip}.txt $dir/scoring/$lmwt.${wip}.txt.orig
-      awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-         <$dir/scoring/$lmwt.${wip}.txt.orig >$dir/scoring/$lmwt.${wip}.txt
-    fi
  done
 done


--- a/egs/fisher_swbd/s5/local/score_sclite.sh
+++ b/egs/fisher_swbd/s5/local/score_sclite.sh
@@ -6,7 +6,6 @@ cmd=run.pl
 stage=0
 min_lmwt=5
 max_lmwt=17
-reverse=false
 word_ins_penalty=0.0,0.5,1.0
 #end configuration section.

@@ -20,7 +19,6 @@ if [ $# -ne 3 ]; then
  echo "    --stage (0|1|2)                 # start scoring script from part-way through."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -30,7 +28,7 @@ dir=$3

 model=$dir/../final.mdl # assume model one level up from decoding dir.

-hubscr=$KALDI_ROOT/tools/sctk/bin/hubscr.pl 
+hubscr=$KALDI_ROOT/tools/sctk/bin/hubscr.pl
 [ ! -f $hubscr ] && echo "Cannot find scoring program at $hubscr" && exit 1;
 hubdir=`dirname $hubscr`

@@ -39,13 +37,6 @@ for f in $data/stm $data/glm $lang/words.txt $lang/phones/word_boundary.int \
  [ ! -f $f ] && echo "$0: expecting file $f to exist" && exit 1;
 done

-align_word=
-reorder_opt=
-if $reverse; then
-  align_word="lattice-reverse ark:- ark:- |"
-  reorder_opt="--reorder=false"
-fi
-
 if [ -f $dir/../frame_shift ]; then
  frame_shift_opt="--frame-shift=$(cat $dir/../frame_shift)"
  echo "$0: $dir/../frame_shift exists, using $frame_shift_opt"
@@ -66,7 +57,7 @@ if [ $stage -le 0 ]; then
      lattice-scale --lm-scale=LMWT "ark:gunzip -c $dir/lat.*.gz|" ark:- \| \
      lattice-add-penalty --word-ins-penalty=$wip ark:- ark:- \| \
      lattice-1best ark:- ark:- \| \
-      $align_word lattice-align-words $reorder_opts $lang/phones/word_boundary.int $model ark:- ark:- \| \
+      lattice-align-words $lang/phones/word_boundary.int $model ark:- ark:- \| \
      nbest-to-ctm $frame_shift_opt ark:- - \| \
      utils/int2sym.pl -f 5 $lang/words.txt  \| \
      utils/convert_ctm.pl $data/segments $data/reco2file_and_channel \
@@ -95,7 +86,7 @@ fi

 # Score the set...
 if [ $stage -le 2 ]; then
-  for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do  
+  for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do
    $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.${wip}.log \
      cp $data/stm $dir/score_LMWT_${wip}/ '&&' \
        $hubscr -p $hubdir -V -l english -h hub5 -g $data/glm -r $dir/score_LMWT_${wip}/stm $dir/score_LMWT_${wip}/${name}.ctm || exit 1;
@@ -106,7 +97,7 @@ fi
 case "$name" in eval2000* )
  # Score only the, swbd part...
  if [ $stage -le 3 ]; then
-    for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do 
+    for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do
      $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.swbd.LMWT.${wip}.log \
        grep -v '^en_' $data/stm '>' $dir/score_LMWT_${wip}/stm.swbd '&&' \
        grep -v '^en_' $dir/score_LMWT_${wip}/${name}.ctm '>' $dir/score_LMWT_${wip}/${name}.ctm.swbd '&&' \
@@ -115,7 +106,7 @@ case "$name" in eval2000* )
  fi
  # Score only the, callhome part...
  if [ $stage -le 3 ]; then
-    for  wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do  
+    for  wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do
      $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.callhm.LMWT.${wip}.log \
        grep -v '^sw_' $data/stm '>' $dir/score_LMWT_${wip}/stm.callhm '&&' \
        grep -v '^sw_' $dir/score_LMWT_${wip}/${name}.ctm '>' $dir/score_LMWT_${wip}/${name}.ctm.callhm '&&' \
@@ -125,7 +116,7 @@ case "$name" in eval2000* )
 ;;

 rt03* )
-    
+
  # Score only the swbd part...
  if [ $stage -le 3 ]; then
    for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do

--- a/egs/gale_arabic/s5/local/score.sh
+++ b/egs/gale_arabic/s5/local/score.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0
 min_lmwt=7
 max_lmwt=17
@@ -25,7 +24,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -49,14 +47,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --word-symbol-table=$symtab \
    ark:- ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/librispeech/s5/local/score.sh
+++ b/egs/librispeech/s5/local/score.sh
@@ -9,7 +9,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0,0.5,1.0
 min_lmwt=7
 max_lmwt=17
@@ -27,7 +26,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -53,16 +51,6 @@ for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do
      ark:- ark,t:$dir/scoring/LMWT.$wip.tra || exit 1;
 done

-if $reverse; then
-  for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do
-    for lmwt in `seq $min_lmwt $max_lmwt`; do
-      mv $dir/scoring/$lmwt.$wip.tra $dir/scoring/$lmwt.$wip.tra.orig
-      awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-        <$dir/scoring/$lmwt.$wip.tra.orig >$dir/scoring/$lmwt.$wip.tra
-    done
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 for wip in $(echo $word_ins_penalty | sed 's/,/ /g'); do
  $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.$wip.log \

--- a/egs/reverb/s5/local/score.sh
+++ b/egs/reverb/s5/local/score.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0
 min_lmwt=7
 max_lmwt=17
@@ -25,7 +24,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -49,14 +47,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --word-symbol-table=$symtab \
    ark:- ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/sprakbanken/s5/local/run_fwdbwd.sh
+++ b/egs/sprakbanken/s5/local/run_fwdbwd.sh
-#prepare reverse lexicon and language model for backwards decoding
-utils/prepare_lang.sh --reverse true data/local/dict "<SPOKEN_NOISE>" data/local/lang_tmp.reverse data/lang.reverse || exit 1;
-utils/reverse_lm.sh data/local/nist_lm/lm_bg_5k.arpa.gz data/lang.reverse data/lang_test_bg_5k.reverse || exit 1;
-utils/reverse_lm_test.sh data/lang_test_bg_5k data/lang_test_bg_5k.reverse || exit 1;
-
-# normal forward decoding
-utils/mkgraph.sh data/lang_test_bg_5k exp/tri2a exp/tri2a/graph_bg5k
-steps/decode_fwdbwd.sh --beam 10.0 --latbeam 4.0 --nj 8 --cmd "$decode_cmd" \
-  exp/tri2a/graph_bg5k data/test_eval92 exp/tri2a/decode_eval92_bg5k_10 || exit 1;
-
-# backward decoding
-utils/mkgraph.sh --reverse data/lang_test_bg_5k.reverse exp/tri2a exp/tri2a/graph_bg5k_r
-steps/decode_fwdbwd.sh --beam 10.0 --latbeam 4.0 --reverse true --nj 8 --cmd "$decode_cmd" \
-  exp/tri2a/graph_bg5k_r data/test_eval92 exp/tri2a/decode_eval92_bg5k_reverse10 || exit 1;
-
-# pingpong decoding
-steps/decode_fwdbwd.sh --beam 10.0 --max-beam 20.0 --reverse true --nj 8 --cmd "$decode_cmd" \
-  --first_pass exp/tri2a/decode_eval92_bg5k_10 exp/tri2a/graph_bg5k_r data/test_eval92 exp/tri2a/decode_eval92_bg5k_pingpong10 || exit 1;
-steps/decode_fwdbwd.sh --beam 10.0 --max-beam 20.0 --nj 8 --cmd "$decode_cmd" \
-  --first_pass exp/tri2a/decode_eval92_bg5k_reverse10 exp/tri2a/graph_bg5k data/test_eval92 exp/tri2a/decode_eval92_bg5k_pongping10 || exit 1;
-
-# same for bigger language models (on machine with 8GB RAM, you can run the whole decoding in 3-4 min without SGE)
-utils/prepare_lang.sh --reverse true data/local/dict_larger "<SPOKEN_NOISE>" data/local/lang_larger.reverse data/lang_bd.reverse || exit;
-utils/reverse_lm.sh --lexicon data/local/dict_larger/lexicon.txt data/local/local_lm/3gram-mincount/lm_pr6.0.gz data/lang_bd.reverse data/lang_test_bd_tgpr.reverse || exit 1;
-utils/reverse_lm_test.sh data/lang_test_bd_tgpr data/lang_test_bd_tgpr.reverse || exit 1;
-
-utils/mkgraph.sh data/lang_test_bd_tgpr exp/tri2a exp/tri2a/graph_bd_tgpr
-steps/decode_fwdbwd.sh --beam 10.0 --latbeam 4.0 --nj 4 --cmd run.pl \
-  exp/tri2a/graph_bd_tgpr data/test_eval92 exp/tri2a/decode_eval92_bdtgpr4_10 || exit 1;
-
-utils/mkgraph.sh --reverse data/lang_test_bd_tgpr.reverse exp/tri2a exp/tri2a/graph_bd_tgpr_r
-steps/decode_fwdbwd.sh --beam 10.0 --latbeam 4.0 --reverse true --nj 4 --cmd run.pl \
-  exp/tri2a/graph_bd_tgpr_r data/test_eval92 exp/tri2a/decode_eval92_bdtgpr4_reverse10 || exit 1;
-
-steps/decode_fwdbwd.sh --beam 10.0 --max-beam 20.0 --reverse true --nj 4 --cmd run.pl \
-  --first_pass exp/tri2a/decode_eval92_bdtgpr4_10 exp/tri2a/graph_bd_tgpr_r data/test_eval92 \
-  exp/tri2a/decode_eval92_bdtgpr4_pingpong10 || exit 1;
-
-steps/decode_fwdbwd.sh --beam 10.0 --max-beam 20.0 --nj 4 --cmd run.pl \
-  --first_pass exp/tri2a/decode_eval92_bdtgpr4_reverse10 exp/tri2a/graph_bd_tgpr data/test_eval92 \
-  exp/tri2a/decode_eval92_bdtgpr4_pongping10 || exit 1;
--- a/egs/sprakbanken/s5/local/score.sh
+++ b/egs/sprakbanken/s5/local/score.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0
 min_lmwt=7
 max_lmwt=17
@@ -25,7 +24,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -49,14 +47,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --word-symbol-table=$symtab \
    ark:- ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \

--- a/egs/swahili/s5/local/score.sh
+++ b/egs/swahili/s5/local/score.sh
@@ -8,7 +8,6 @@
 cmd=run.pl
 stage=0
 decode_mbr=true
-reverse=false
 word_ins_penalty=0.0
 min_lmwt=7
 max_lmwt=17
@@ -25,7 +24,6 @@ if [ $# -ne 3 ]; then
  echo "    --decode_mbr (true/false)       # maximum bayes risk decoding (confusion network)."
  echo "    --min_lmwt <int>                # minumum LM-weight for lattice rescoring "
  echo "    --max_lmwt <int>                # maximum LM-weight for lattice rescoring "
-  echo "    --reverse (true/false)          # score with time reversed features "
  exit 1;
 fi

@@ -49,14 +47,6 @@ $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/best_path.LMWT.log \
  lattice-best-path --word-symbol-table=$symtab \
    ark:- ark,t:$dir/scoring/LMWT.tra || exit 1;

-if $reverse; then
-  for lmwt in `seq $min_lmwt $max_lmwt`; do
-    mv $dir/scoring/$lmwt.tra $dir/scoring/$lmwt.tra.orig
-    awk '{ printf("%s ",$1); for(i=NF; i>1; i--){ printf("%s ",$i); } printf("\n"); }' \
-       <$dir/scoring/$lmwt.tra.orig >$dir/scoring/$lmwt.tra
-  done
-fi
-
 # Note: the double level of quoting for the sed command
 $cmd LMWT=$min_lmwt:$max_lmwt $dir/scoring/log/score.LMWT.log \
   cat $dir/scoring/LMWT.tra \| \