diff --git a/egs/tedlium/s5_r3/local/chain/run_tdnn_1c.sh b/egs/tedlium/s5_r3/local/chain/run_tdnn_1c.sh
new file mode 100755
index 00000000000..5b531fd663d
--- /dev/null
+++ b/egs/tedlium/s5_r3/local/chain/run_tdnn_1c.sh
@@ -0,0 +1,249 @@
+#!/bin/bash
+
+# 1g is as 1f but moving to a factorized TDNN (TDNN-F) model, re-tuning it, and
+#  switching to unconstrained egs (the last of which gives around 0.1%
+#  improvement).  (Note: I don't believe the Tedlium TDNN models were,
+#  previously, very well-tuned).
+
+# local/chain/compare_wer_general.sh exp/chain_cleaned/tdnn1f_sp_bi exp/chain_cleaned/tdnn1g_sp
+# System                tdnn1f_sp_bi tdnn1g_sp
+# WER on dev(orig)            8.9       7.9
+# WER on dev(rescored)        8.1       7.3
+# WER on test(orig)           9.1       8.0
+# WER on test(rescored)       8.6       7.6
+# Final train prob        -0.1026   -0.0637
+# Final valid prob        -0.1031   -0.0750
+# Final train prob (xent)   -1.4370   -0.9792
+# Final valid prob (xent)   -1.4670   -0.9951
+# Num-params                 6994800   9431072
+
+# steps/info/chain_dir_info.pl exp/chain_cleaned/tdnn1g_sp
+# exp/chain_cleaned/tdnn1g_sp: num-iters=108 nj=3..12 num-params=9.4M dim=40+100->3600 combine=-0.060->-0.060 (over 2) xent:train/valid[71,107,final]=(-1.30,-0.985,-0.979/-1.29,-1.00,-0.995) logprob:train/valid[71,107,final]=(-0.098,-0.065,-0.064/-0.100,-0.075,-0.075)
+
+## how you run this (note: this assumes that the run_tdnn.sh soft link points here;
+## otherwise call it directly in its location).
+# by default, with cleanup:
+# local/chain/run_tdnn.sh
+
+# without cleanup:
+# local/chain/run_tdnn.sh  --train-set train --gmm tri3 --nnet3-affix "" &
+
+set -e -o pipefail
+
+# First the options that are passed through to run_ivector_common.sh
+# (some of which are also used in this script directly).
+stage=0
+nj=15
+decode_nj=15
+xent_regularize=0.1
+dropout_schedule='0,0@0.20,0.5@0.50,0'
+
+train_set=train_cleaned
+gmm=tri3_cleaned  # the gmm for the target data
+num_threads_ubm=1
+nnet3_affix=_cleaned  # cleanup affix for nnet3 and chain dirs, e.g. _cleaned
+
+# The rest are configs specific to this script.  Most of the parameters
+# are just hardcoded at this level, in the commands below.
+train_stage=-10
+tree_affix=  # affix for tree directory, e.g. "a" or "b", in case we change the configuration.
+tdnn_affix=1g  #affix for TDNN directory, e.g. "a" or "b", in case we change the configuration.
+common_egs_dir=  # you can set this to use previously dumped egs.
+remove_egs=true
+
+# 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
+
+#local/nnet3/run_ivector_common.sh --stage $stage \
+#                                  --nj $nj \
+#                                  --train-set $train_set \
+#                                  --gmm $gmm \
+#                                  --num-threads-ubm $num_threads_ubm \
+#                                  --nnet3-affix "$nnet3_affix"
+
+
+gmm_dir=exp/$gmm
+ali_dir=exp/${gmm}_ali_${train_set}_sp
+tree_dir=exp/chain${nnet3_affix}/tree_bi${tree_affix}
+lat_dir=exp/chain${nnet3_affix}/${gmm}_${train_set}_sp_lats
+dir=exp/chain${nnet3_affix}/tdnn${tdnn_affix}_sp_bi
+train_data_dir=data/${train_set}_sp_hires
+lores_train_data_dir=data/${train_set}_sp
+train_ivector_dir=exp/nnet3${nnet3_affix}/ivectors_${train_set}_sp_hires
+
+
+for f in $gmm_dir/final.mdl $train_data_dir/feats.scp $train_ivector_dir/ivector_online.scp \
+    $lores_train_data_dir/feats.scp $ali_dir/ali.1.gz $gmm_dir/final.mdl; do
+  [ ! -f $f ] && echo "$0: expected file $f to exist" && exit 1
+done
+
+if [ $stage -le 14 ]; then
+  echo "$0: creating lang directory with one state per phone."
+  # 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.]
+  if [ -d data/lang_chain ]; then
+    if [ data/lang_chain/L.fst -nt data/lang/L.fst ]; then
+      echo "$0: data/lang_chain already exists, not overwriting it; continuing"
+    else
+      echo "$0: data/lang_chain already exists and seems to be older than data/lang..."
+      echo " ... not sure what to do.  Exiting."
+      exit 1;
+    fi
+  else
+    cp -r data/lang data/lang_chain
+    silphonelist=$(cat data/lang_chain/phones/silence.csl) || exit 1;
+    nonsilphonelist=$(cat data/lang_chain/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 >data/lang_chain/topo
+  fi
+fi
+
+if [ $stage -le 15 ]; then
+  # Get the alignments as lattices (gives the chain training more freedom).
+  # use the same num-jobs as the alignments
+  steps/align_fmllr_lats.sh --nj 100 --cmd "$train_cmd" ${lores_train_data_dir} \
+    data/lang $gmm_dir $lat_dir
+  rm $lat_dir/fsts.*.gz # save space
+fi
+
+if [ $stage -le 16 ]; then
+  # Build a tree using our new topology.  We know we have alignments for the
+  # speed-perturbed data (local/nnet3/run_ivector_common.sh made them), so use
+  # those.
+  if [ -f $tree_dir/final.mdl ]; then
+    echo "$0: $tree_dir/final.mdl already exists, refusing to overwrite it."
+    exit 1;
+  fi
+  steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 \
+      --context-opts "--context-width=2 --central-position=1" \
+      --cmd "$train_cmd" 4000 ${lores_train_data_dir} data/lang_chain $ali_dir $tree_dir
+fi
+
+if [ $stage -le 17 ]; then
+  mkdir -p $dir
+
+  echo "$0: creating neural net configs using the xconfig parser";
+
+  num_targets=$(tree-info $tree_dir/tree |grep num-pdfs|awk '{print $2}')
+  learning_rate_factor=$(echo "print 0.5/$xent_regularize" | python)
+  affine_opts="l2-regularize=0.008 dropout-proportion=0.0 dropout-per-dim-continuous=true"
+  tdnnf_opts="l2-regularize=0.008 dropout-proportion=0.0 bypass-scale=0.66"
+  linear_opts="l2-regularize=0.008 orthonormal-constraint=-1.0"
+  prefinal_opts="l2-regularize=0.008"
+  output_opts="l2-regularize=0.002"
+
+  mkdir -p $dir/configs
+  cat <<EOF > $dir/configs/network.xconfig
+  input dim=100 name=ivector
+  input dim=40 name=input
+
+  # please note that it is important to have input layer with the name=input
+  # as the layer immediately preceding the fixed-affine-layer to enable
+  # the use of short notation for the descriptor
+  fixed-affine-layer name=lda input=Append(-1,0,1,ReplaceIndex(ivector, t, 0)) affine-transform-file=$dir/configs/lda.mat
+
+  # the first splicing is moved before the lda layer, so no splicing here
+  relu-batchnorm-dropout-layer name=tdnn1 $affine_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=256 $linear_opts
+
+  prefinal-layer name=prefinal-chain input=prefinal-l $prefinal_opts big-dim=1024 small-dim=256
+  output-layer name=output include-log-softmax=false dim=$num_targets $output_opts
+
+  prefinal-layer name=prefinal-xent input=prefinal-l $prefinal_opts big-dim=1024 small-dim=256
+  output-layer name=output-xent dim=$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 18 ]; then
+  if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
+    utils/create_split_dir.pl \
+     /export/b0{5,6,7,8}/$USER/kaldi-data/egs/ami-$(date +'%m_%d_%H_%M')/s5/$dir/egs/storage $dir/egs/storage
+  fi
+
+ steps/nnet3/chain/train.py --stage $train_stage \
+    --cmd "$decode_cmd" \
+    --feat.online-ivector-dir $train_ivector_dir \
+    --feat.cmvn-opts "--norm-means=false --norm-vars=false" \
+    --chain.xent-regularize $xent_regularize \
+    --chain.leaky-hmm-coefficient 0.1 \
+    --chain.l2-regularize 0.0 \
+    --chain.apply-deriv-weights false \
+    --chain.lm-opts="--num-extra-lm-states=2000" \
+    --trainer.dropout-schedule $dropout_schedule \
+    --trainer.add-option="--optimization.memory-compression-level=2" \
+    --egs.dir "$common_egs_dir" \
+    --egs.opts "--frames-overlap-per-eg 0 --constrained false" \
+    --egs.chunk-width 150,110,100 \
+    --trainer.num-chunk-per-minibatch 64 \
+    --trainer.frames-per-iter 5000000 \
+    --trainer.num-epochs 6 \
+    --trainer.optimization.num-jobs-initial 3 \
+    --trainer.optimization.num-jobs-final 12 \
+    --trainer.optimization.initial-effective-lrate 0.00025 \
+    --trainer.optimization.final-effective-lrate 0.000025 \
+    --trainer.max-param-change 2.0 \
+    --cleanup.remove-egs $remove_egs \
+    --feat-dir $train_data_dir \
+    --tree-dir $tree_dir \
+    --lat-dir $lat_dir \
+    --dir $dir
+fi
+
+
+
+if [ $stage -le 19 ]; then
+  # Note: it might appear that this data/lang_chain 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
+fi
+
+if [ $stage -le 20 ]; then
+  rm $dir/.error 2>/dev/null || true
+  for dset in dev test; do
+      (
+      steps/nnet3/decode.sh --num-threads 4 --nj $decode_nj --cmd "$decode_cmd" \
+          --acwt 1.0 --post-decode-acwt 10.0 \
+          --online-ivector-dir exp/nnet3${nnet3_affix}/ivectors_${dset}_hires \
+          --scoring-opts "--min-lmwt 5 " \
+         $dir/graph data/${dset}_hires $dir/decode_${dset} || exit 1;
+      steps/lmrescore_const_arpa.sh --cmd "$decode_cmd" data/lang data/lang_rescore \
+        data/${dset}_hires ${dir}/decode_${dset} ${dir}/decode_${dset}_rescore || exit 1
+    ) || touch $dir/.error &
+  done
+  wait
+  if [ -f $dir/.error ]; then
+    echo "$0: something went wrong in decoding"
+    exit 1
+  fi
+fi
+exit 0
diff --git a/egs/wsj/s5/steps/cleanup/internal/align_ctm_ref.py b/egs/wsj/s5/steps/cleanup/internal/align_ctm_ref.py
index 848ca61ebe4..d3e012da13c 100755
--- a/egs/wsj/s5/steps/cleanup/internal/align_ctm_ref.py
+++ b/egs/wsj/s5/steps/cleanup/internal/align_ctm_ref.py
@@ -127,7 +127,7 @@ def read_text(text_file):
                 "Did not get enough columns; line {0} in {1}"
                 "".format(line, text_file.name))
         elif len(parts) == 1:
-            logger.warn("Empty transcript for utterance %s in %s", 
+            logger.warn("Empty transcript for utterance %s in %s",
                         parts[0], text_file.name)
             yield parts[0], []
         else:
diff --git a/egs/wsj/s5/steps/cleanup/internal/get_ctm_edits.py b/egs/wsj/s5/steps/cleanup/internal/get_ctm_edits.py
index a19c5344572..3032a4b434a 100755
--- a/egs/wsj/s5/steps/cleanup/internal/get_ctm_edits.py
+++ b/egs/wsj/s5/steps/cleanup/internal/get_ctm_edits.py
@@ -1,4 +1,4 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
 
 # Copyright 2016   Vimal Manohar
 #           2016   Johns Hopkins University (author: Daniel Povey)
@@ -116,17 +116,17 @@
 def OpenFiles():
     global ctm_edits_out, edits_in, ctm_in, symbol_table, oov_word
     try:
-        ctm_edits_out = open(args.ctm_edits_out, 'w')
+        ctm_edits_out = open(args.ctm_edits_out, 'w', encoding='utf-8')
     except:
         sys.exit("get_ctm_edits.py: error opening ctm-edits file {0} for output".format(
                 args.ctm_edits_out))
     try:
-        edits_in = open(args.edits_in)
+        edits_in = open(args.edits_in, encoding='utf-8')
     except:
         sys.exit("get_ctm_edits.py: error opening edits file {0} for input".format(
                 args.edits_in))
     try:
-        ctm_in = open(args.ctm_in)
+        ctm_in = open(args.ctm_in, encoding='utf-8')
     except:
         sys.exit("get_ctm_edits.py: error opening ctm file {0} for input".format(
                 args.ctm_in))
@@ -138,7 +138,7 @@ def OpenFiles():
             print("get_ctm_edits.py: error: if you set the the --symbol-table option "
                   "you must also set the --oov option", file = sys.stderr)
         try:
-            f = open(args.symbol_table, 'r')
+            f = open(args.symbol_table, 'r', encoding='utf-8')
             for line in f.readlines():
                 [ word, integer ] = line.split()
                 if int(integer) == args.oov:
diff --git a/egs/wsj/s5/steps/cleanup/internal/get_non_scored_words.py b/egs/wsj/s5/steps/cleanup/internal/get_non_scored_words.py
index aa71fa47d84..69e0242eafb 100755
--- a/egs/wsj/s5/steps/cleanup/internal/get_non_scored_words.py
+++ b/egs/wsj/s5/steps/cleanup/internal/get_non_scored_words.py
@@ -1,4 +1,4 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
 
 # Copyright 2016   Vimal Manohar
 #           2016   Johns Hopkins University (author: Daniel Povey)
@@ -90,7 +90,7 @@ def read_lang(lang_dir):
         raise
 
     try:
-        for line in open(lang_dir + '/words.txt').readlines():
+        for line in open(lang_dir + '/words.txt', encoding='utf-8').readlines():
             [ word, integer ] = line.split()
             if int(integer) in silence_word_ints:
                 non_scored_words.add(word)
diff --git a/egs/wsj/s5/steps/cleanup/internal/get_pron_stats.py b/egs/wsj/s5/steps/cleanup/internal/get_pron_stats.py
index a33ba85d9fa..3ea217b6589 100755
--- a/egs/wsj/s5/steps/cleanup/internal/get_pron_stats.py
+++ b/egs/wsj/s5/steps/cleanup/internal/get_pron_stats.py
@@ -75,14 +75,14 @@ def ReadEntries(file_handle):
 # Each entry in the list represents the pronounciation candidate(s) of a word.
 # For each non-<eps> word, the entry is a list: [utt_id, word, set(pronunciation_candidates)]. e.g:
 # [911Mothers_2010W-0010916-0012901-1, other, set('AH DH ER', 'AH DH ER K AH N')]
-# For each <eps>, we split the phones it aligns to into two parts: "nonsil_left", 
+# For each <eps>, we split the phones it aligns to into two parts: "nonsil_left",
 # which includes phones before the first silphone, and "nonsil_right", which includes
-# phones after the last silphone. For example, for <eps> : 'V SIL B AH SIL', 
+# phones after the last silphone. For example, for <eps> : 'V SIL B AH SIL',
 # nonsil_left is 'V' and nonsil_right is empty ''. After processing an <eps> entry
 # in ctm_prons, we put it in "info" as an entry:  [utt_id, word, nonsil_right]
 # only if it's nonsil_right segment is not empty, which may be used when processing
 # the next word.
-# 
+#
 # Normally, one non-<eps> word is only aligned to one pronounciation candidate. However
 # when there is a preceding/following <eps>, like in the following example, we
 # assume the phones aligned to <eps> should be statistically distributed
@@ -90,7 +90,7 @@ def ReadEntries(file_handle):
 # Thus we append the "nonsil_left" segment of these phones to the pronounciation
 # of the preceding word, if the last phone of this pronounciation is not a silence phone,
 # Similarly we can add a pron candidate to the following word.
-# 
+#
 # For example, for the following part of a ctm_prons file:
 # 911Mothers_2010W-0010916-0012901-1 other AH DH ER
 # 911Mothers_2010W-0010916-0012901-1 <eps> K AH N SIL B
@@ -99,11 +99,11 @@ def ReadEntries(file_handle):
 # 911Mothers_2010W-0010916-0012901-1 when W EH N
 # 911Mothers_2010W-0010916-0012901-1 people P IY P AH L
 # 911Mothers_2010W-0010916-0012901-1 <eps> SIL
-# 911Mothers_2010W-0010916-0012901-1 heard HH ER 
+# 911Mothers_2010W-0010916-0012901-1 heard HH ER
 # 911Mothers_2010W-0010916-0012901-1 <eps> D
 # 911Mothers_2010W-0010916-0012901-1 that SIL DH AH T
 # 911Mothers_2010W-0010916-0012901-1 my M AY
-# 
+#
 # The corresponding segment in the "info" list is:
 # [911Mothers_2010W-0010916-0012901-1, other, set('AH DH ER', 'AH DH ER K AH N')]
 # [911Mothers_2010W-0010916-0012901-1, <eps>, 'B'
@@ -113,7 +113,7 @@ def ReadEntries(file_handle):
 # [911Mothers_2010W-0010916-0012901-1, <eps>, 'D']
 # [911Mothers_2010W-0010916-0012901-1, that, set('SIL DH AH T')]
 # [911Mothers_2010W-0010916-0012901-1, my, set('M AY')]
-# 
+#
 # Then we accumulate pronouciation stats from "info". Basically, for each occurence
 # of a word, each pronounciation candidate gets equal soft counts. e.g. In the above
 # example, each pron candidate of "because" gets a count of 1/4. The stats is stored
@@ -139,20 +139,20 @@ def GetStatsFromCtmProns(silphones, optional_silence, non_scored_words, ctm_pron
         # So we apply the same merging method in these cases.
         if word == '<eps>' or (word in non_scored_words and word != '<unk>' and word != '<UNK>'):
             nonsil_left = []
-            nonsil_right = [] 
+            nonsil_right = []
             for phone in phones:
                 if phone in silphones:
                     break
                 nonsil_left.append(phone)
-            
+
             for phone in reversed(phones):
                 if phone in silphones:
                     break
                 nonsil_right.insert(0, phone)
-            
+
             # info[-1][0] is the utt_id of the last entry
-            if len(nonsil_left) > 0 and len(info) > 0 and utt == info[-1][0]: 
-                # pron_ext is a set of extended pron candidates. 
+            if len(nonsil_left) > 0 and len(info) > 0 and utt == info[-1][0]:
+                # pron_ext is a set of extended pron candidates.
                 pron_ext = set()
                 # info[-1][2] is the set of pron candidates of the last entry.
                 for pron in info[-1][2]:
@@ -211,7 +211,7 @@ def GetStatsFromCtmProns(silphones, optional_silence, non_scored_words, ctm_pron
                 stats[(word, phones)] = stats.get((word, phones), 0) + count
     return stats
 
-def WriteStats(stats, file_handle):            
+def WriteStats(stats, file_handle):
     for word_pron, count in stats.items():
         print('{0} {1} {2}'.format(count, word_pron[0], word_pron[1]), file=file_handle)
     file_handle.close()
@@ -222,7 +222,7 @@ def Main():
     non_scored_words = ReadEntries(args.non_scored_words_file_handle)
     optional_silence = ReadEntries(args.optional_silence_file_handle)
     stats = GetStatsFromCtmProns(silphones, optional_silence.pop(), non_scored_words, args.ctm_prons_file_handle)
-    WriteStats(stats, args.stats_file_handle)            
+    WriteStats(stats, args.stats_file_handle)
 
 if __name__ == "__main__":
     Main()
diff --git a/egs/wsj/s5/steps/cleanup/internal/make_one_biased_lm.py b/egs/wsj/s5/steps/cleanup/internal/make_one_biased_lm.py
index e41a67705e9..68055729fd9 100755
--- a/egs/wsj/s5/steps/cleanup/internal/make_one_biased_lm.py
+++ b/egs/wsj/s5/steps/cleanup/internal/make_one_biased_lm.py
@@ -1,4 +1,4 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
 
 # Copyright 2016  Johns Hopkins University (Author: Daniel Povey)
 # Apache 2.0.
@@ -142,16 +142,18 @@ def CompletelyDiscountLowCountStates(self, min_count):
         hist_to_total_count = self.GetHistToTotalCount()
         for n in reversed(list(range(2, self.ngram_order))):
             this_order_counts = self.counts[n]
+            to_delete = []
             for hist in this_order_counts.keys():
                 if hist_to_total_count[hist] < min_count:
                     # we need to completely back off this count.
                     word_to_count = this_order_counts[hist]
-                    del this_order_counts[hist] # delete the key from the dict.
+                    # mark this key for deleting
+                    to_delete.append(hist)
                     backoff_hist = hist[1:]  # this will be a tuple not a list.
                     for word, count in word_to_count.items():
                         self.AddCount(backoff_hist, word, count)
-
-
+            for hist in to_delete:
+                del this_order_counts[hist]
 
     # This backs off the counts according to Kneser-Ney (unmodified,
     # with interpolation).
@@ -200,7 +202,7 @@ def AddTopWords(self, top_words_file):
         word_to_count = self.counts[0][empty_history]
         total = sum(word_to_count.values())
         try:
-            f = open(top_words_file)
+            f = open(top_words_file, mode='r', encoding='utf-8')
         except:
             sys.exit("make_one_biased_lm.py: error opening top-words file: "
                      "--top-words=" + top_words_file)
diff --git a/egs/wsj/s5/steps/cleanup/internal/modify_ctm_edits.py b/egs/wsj/s5/steps/cleanup/internal/modify_ctm_edits.py
index d6f0d0f6b23..af63ca27d2b 100755
--- a/egs/wsj/s5/steps/cleanup/internal/modify_ctm_edits.py
+++ b/egs/wsj/s5/steps/cleanup/internal/modify_ctm_edits.py
@@ -1,4 +1,4 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
 
 # Copyright 2016   Vimal Manohar
 #           2016   Johns Hopkins University (author: Daniel Povey)
@@ -105,7 +105,7 @@
 def ReadNonScoredWords(non_scored_words_file):
     global non_scored_words
     try:
-        f = open(non_scored_words_file)
+        f = open(non_scored_words_file, encoding='utf-8')
     except:
         sys.exit("modify_ctm_edits.py: error opening file: "
                  "--non-scored-words=" + non_scored_words_file)
@@ -317,12 +317,12 @@ def ProcessUtterance(split_lines_of_utt):
 
 def ProcessData():
     try:
-        f_in = open(args.ctm_edits_in)
+        f_in = open(args.ctm_edits_in, encoding='utf-8')
     except:
         sys.exit("modify_ctm_edits.py: error opening ctm-edits input "
                  "file {0}".format(args.ctm_edits_in))
     try:
-        f_out = open(args.ctm_edits_out, 'w')
+        f_out = open(args.ctm_edits_out, 'w', encoding='utf-8')
     except:
         sys.exit("modify_ctm_edits.py: error opening ctm-edits output "
                  "file {0}".format(args.ctm_edits_out))
diff --git a/egs/wsj/s5/steps/cleanup/internal/segment_ctm_edits.py b/egs/wsj/s5/steps/cleanup/internal/segment_ctm_edits.py
index 39d6cb6ed80..e571fefb84c 100755
--- a/egs/wsj/s5/steps/cleanup/internal/segment_ctm_edits.py
+++ b/egs/wsj/s5/steps/cleanup/internal/segment_ctm_edits.py
@@ -1,4 +1,5 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
+
 
 # Copyright 2016   Vimal Manohar
 #           2016   Johns Hopkins University (author: Daniel Povey)
@@ -894,7 +895,7 @@ def AccWordStatsForUtterance(split_lines_of_utt,
 
 def PrintWordStats(word_stats_out):
     try:
-        f = open(word_stats_out, 'w')
+        f = open(word_stats_out, 'w', encoding='utf-8')
     except:
         sys.exit("segment_ctm_edits.py: error opening word-stats file --word-stats-out={0} "
                  "for writing".format(word_stats_out))
@@ -924,23 +925,23 @@ def PrintWordStats(word_stats_out):
 
 def ProcessData():
     try:
-        f_in = open(args.ctm_edits_in)
+        f_in = open(args.ctm_edits_in, encoding='utf-8')
     except:
         sys.exit("segment_ctm_edits.py: error opening ctm-edits input "
                  "file {0}".format(args.ctm_edits_in))
     try:
-        text_output_handle = open(args.text_out, 'w')
+        text_output_handle = open(args.text_out, 'w', encoding='utf-8')
     except:
         sys.exit("segment_ctm_edits.py: error opening text output "
                  "file {0}".format(args.text_out))
     try:
-        segments_output_handle = open(args.segments_out, 'w')
+        segments_output_handle = open(args.segments_out, 'w', encoding='utf-8')
     except:
         sys.exit("segment_ctm_edits.py: error opening segments output "
                  "file {0}".format(args.text_out))
     if args.ctm_edits_out != None:
         try:
-            ctm_edits_output_handle = open(args.ctm_edits_out, 'w')
+            ctm_edits_output_handle = open(args.ctm_edits_out, 'w', encoding='utf-8')
         except:
             sys.exit("segment_ctm_edits.py: error opening ctm-edits output "
                      "file {0}".format(args.ctm_edits_out))
@@ -994,7 +995,7 @@ def ProcessData():
 def ReadNonScoredWords(non_scored_words_file):
     global non_scored_words
     try:
-        f = open(non_scored_words_file)
+        f = open(non_scored_words_file, encoding='utf-8')
     except:
         sys.exit("segment_ctm_edits.py: error opening file: "
                  "--non-scored-words=" + non_scored_words_file)
@@ -1015,7 +1016,7 @@ def ReadNonScoredWords(non_scored_words_file):
 oov_symbol = None
 if args.oov_symbol_file != None:
     try:
-        with open(args.oov_symbol_file) as f:
+        with open(args.oov_symbol_file, encoding='utf-8') as f:
             line = f.readline()
             assert len(line.split()) == 1
             oov_symbol = line.split()[0]
diff --git a/egs/wsj/s5/steps/cleanup/make_biased_lms.py b/egs/wsj/s5/steps/cleanup/make_biased_lms.py
index ab508eedc9c..4b1fd320221 100755
--- a/egs/wsj/s5/steps/cleanup/make_biased_lms.py
+++ b/egs/wsj/s5/steps/cleanup/make_biased_lms.py
@@ -1,4 +1,4 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
 
 from __future__ import print_function
 import sys
@@ -55,7 +55,7 @@ def ProcessGroupOfLines(group_of_lines):
     try:
         command = "steps/cleanup/internal/make_one_biased_lm.py " + args.lm_opts
         p = subprocess.Popen(command, shell = True, stdin = subprocess.PIPE,
-                            stdout = sys.stdout, stderr = sys.stderr)
+                             stdout = sys.stdout, stderr = sys.stderr)
         for line in group_of_lines:
             a = line.split()
             if len(a) == 0:
@@ -63,13 +63,15 @@ def ProcessGroupOfLines(group_of_lines):
             utterance_id = a[0]
             # print <utt> <utt-group> to utterance-map file
             print(utterance_id, group_utterance_id, file = utterance_map_file)
-            rest_of_line = ' '.join(a[1:])  # get rid of utterance id.
-            print(rest_of_line, file=p.stdin)
+            rest_of_line = ' '.join(a[1:]) + '\n' # get rid of utterance id.
+            p.stdin.write(rest_of_line.encode('utf-8'))
         p.stdin.close()
         assert p.wait() == 0
-    except Exception as e:
-        sys.exit("make_biased_lms.py: error calling subprocess, command was: " +
-                 command + ", error was : " + str(e))
+    except Exception:
+        sys.stderr.write(
+            "make_biased_lms.py: error calling subprocess, command was: " +
+            command)
+        raise
     # Print a blank line; this terminates the FST in the Kaldi fst-archive
     # format.
     print("")
diff --git a/egs/wsj/s5/utils/perturb_data_dir_speed.sh b/egs/wsj/s5/utils/perturb_data_dir_speed.sh
index a50cdb04be4..99c9cbdb1f0 100755
--- a/egs/wsj/s5/utils/perturb_data_dir_speed.sh
+++ b/egs/wsj/s5/utils/perturb_data_dir_speed.sh
@@ -102,6 +102,9 @@ fi
 if [ -f $srcdir/spk2gender ]; then
   utils/apply_map.pl -f 1 $destdir/spk_map <$srcdir/spk2gender >$destdir/spk2gender
 fi
+if [ -f $srcdir/utt2lang ]; then
+  utils/apply_map.pl -f 1 $destdir/utt_map <$srcdir/utt2lang >$destdir/utt2lang
+fi
 
 #prepare speed-perturbed utt2dur
 if [ ! -f $srcdir/utt2dur ]; then