vimalmanohar · vimalmanohar · Jan 10, 2017 · Jan 19, 2017 · Jan 19, 2017 · Jan 19, 2017
diff --git a/.gitignore b/.gitignore
@@ -44,6 +44,9 @@ GRTAGS
 GPATH
 GSYMS
 
+# python compiled sources
+*.pyc
+
 # Make dependencies
 .depend.mk
 
@@ -57,8 +60,11 @@ GSYMS
 *.manifest
 
 # /src/
+/src/.short_version
+/src/base/version.h
 /src/doc/table/
 /src/doc/tools.dox
+/src/doc/*.html
 /src/htdocs/
 /src/html.tar.gz
 /src/kaldi.mk
@@ -82,6 +88,14 @@ GSYMS
 /tools/openfst-1.3.4/
 /tools/openfst-1.4.1.tar.gz
 /tools/openfst-1.4.1/
+/tools/openfst-1.5.4.tar.gz
+/tools/openfst-1.5.4/
+/tools/openfst-1.6.0.tar.gz
+/tools/openfst-1.6.0/
+/tools/openfst-1.6.1.tar.gz
+/tools/openfst-1.6.1/
+/tools/openfst-1.6.2.tar.gz
+/tools/openfst-1.6.2/
 /tools/pa_stable_v19_20111121.tgz
 /tools/portaudio/
 /tools/sctk-2.4.0-20091110-0958.tar.bz2
@@ -108,6 +122,14 @@ GSYMS
 /tools/pthreads
 /tools/pthreads*.zip
 /tools/sequitur
+/tools/sequitur-g2p
 /tools/srilm.tgz
+/tools/liblbfgs-1.10.tar.gz
+/tools/liblbfgs-1.10/
+/tools/openfst-1.5.0.tar.gz
+/tools/openfst-1.5.0/
+/tools/srilm-1.7.2-beta.tar.gz
+/tools/liblbfgs/
+/tools/sequitur-g2p/
 
 /kaldiwin_vs*
diff --git a/.travis.yml b/.travis.yml
@@ -10,24 +10,27 @@ addons:
   apt:
     sources:
       - ubuntu-toolchain-r-test
+      - llvm-toolchain-precise-3.8
     packages:
       - gdb
       - gcc-4.9
       - g++-4.9
       - gfortran-4.9
       - liblapack-dev
+      - clang-3.8
 
 branches:
   only:
     - master
+    - kaldi_52
 
 before_install:
   - cat /proc/sys/kernel/core_pattern
   - export XROOT=~/xroot
   - tools/extras/travis_install_bindeps.sh $XROOT
 
 script:
-  - CXX=g++-4.9
+  - CXX=clang++-3.8
     CFLAGS="-march=native"
     LDFLAGS="-llapack"
     INCDIRS="$XROOT/usr/include"

diff --git a/README.md b/README.md
@@ -40,25 +40,30 @@ Development pattern for contributors
 ------------------------------------
 
 1. [Create a personal fork](https://help.github.com/articles/fork-a-repo/)
-   of the [main Kaldi repository] (https://github.com/kaldi-asr/kaldi) in GitHub.
+   of the [main Kaldi repository](https://github.com/kaldi-asr/kaldi) in GitHub.
 2. Make your changes in a named branch different from `master`, e.g. you create
    a branch `my-awesome-feature`.
 3. [Generate a pull request](https://help.github.com/articles/creating-a-pull-request/)
    through the Web interface of GitHub.
-4. As a general rule, please follow [Google C++ Style Guide]
-   (https://google.github.io/styleguide/cppguide.html).
+4. As a general rule, please follow [Google C++ Style Guide](https://google.github.io/styleguide/cppguide.html).
    There are a [few exceptions in Kaldi](http://kaldi-asr.org/doc/style.html).
-   You can use the [Google's cpplint.py]
-   (https://raw.githubusercontent.com/google/styleguide/gh-pages/cpplint/cpplint.py)
+   You can use the [Google's cpplint.py](https://raw.githubusercontent.com/google/styleguide/gh-pages/cpplint/cpplint.py)
    to verify that your code is free of basic mistakes.
 
 Platform specific notes
 -----------------------
 
-PowerPC 64bits little-endian (ppc64le):
+### PowerPC 64bits little-endian (ppc64le)
+
 - Kaldi is expected to work out of the box in RHEL >= 7 and Ubuntu >= 16.04 with
   OpenBLAS, ATLAS, or CUDA.
-- CUDA drivers for ppc64le can be found at [https://developer.nvidia.com/cuda-downloads]
-  (https://developer.nvidia.com/cuda-downloads).
-- An [IBM Redbook] (https://www.redbooks.ibm.com/abstracts/redp5169.html) is
+- CUDA drivers for ppc64le can be found at [https://developer.nvidia.com/cuda-downloads](https://developer.nvidia.com/cuda-downloads).
+- An [IBM Redbook](https://www.redbooks.ibm.com/abstracts/redp5169.html) is
   available as a guide to install and configure CUDA.
+
+### Android
+
+- Kaldi supports cross compiling for Android using Android NDK, clang++ and
+  OpenBLAS.
+- See [this blog post](http://jcsilva.github.io/2017/03/18/compile-kaldi-android/)
+  for details.
diff --git a/egs/ami/s5/local/ami_ihm_data_prep.sh b/egs/ami/s5/local/ami_ihm_data_prep.sh
@@ -69,7 +69,7 @@ sed -e 's?.*/??' -e 's?.wav??' $dir/wav.flist | \
 awk '{print $2}' $dir/segments | sort -u | join - $dir/wav1.scp >  $dir/wav2.scp
 
 #replace path with an appropriate sox command that select single channel only
-awk '{print $1" sox -c 1 -t wavpcm -s "$2" -t wavpcm - |"}' $dir/wav2.scp > $dir/wav.scp
+awk '{print $1" sox -c 1 -t wavpcm -e signed-integer "$2" -t wavpcm - |"}' $dir/wav2.scp > $dir/wav.scp
 
 # (1d) reco2file_and_channel
 cat $dir/wav.scp \

diff --git a/egs/ami/s5/local/ami_ihm_scoring_data_prep.sh b/egs/ami/s5/local/ami_ihm_scoring_data_prep.sh
@@ -68,7 +68,7 @@ sed -e 's?.*/??' -e 's?.wav??' $dir/wav.flist | \
 awk '{print $2}' $dir/segments | sort -u | join - $dir/wav1.scp >  $dir/wav2.scp
 
 #replace path with an appropriate sox command that select single channel only
-awk '{print $1" sox -c 1 -t wavpcm -s "$2" -t wavpcm - |"}' $dir/wav2.scp > $dir/wav.scp
+awk '{print $1" sox -c 1 -t wavpcm -e signed-integer "$2" -t wavpcm - |"}' $dir/wav2.scp > $dir/wav.scp
 
 # (1d) reco2file_and_channel
 cat $dir/wav.scp \

diff --git a/egs/ami/s5/local/ami_mdm_data_prep.sh b/egs/ami/s5/local/ami_mdm_data_prep.sh
@@ -75,7 +75,7 @@ awk '{print $1}' $dir/wav2.scp | join -2 2 - $dir/segments | \
 awk '{print $1}' $dir/segments | join - $dir/text > $dir/t; mv $dir/t $dir/text
 
 #replace path with an appropriate sox command that select single channel only
-awk '{print $1" sox -c 1 -t wavpcm -s "$2" -t wavpcm - |"}' $dir/wav2.scp > $dir/wav.scp
+awk '{print $1" sox -c 1 -t wavpcm -e signed-integer "$2" -t wavpcm - |"}' $dir/wav2.scp > $dir/wav.scp
 
 #prep reco2file_and_channel
 cat $dir/wav.scp | \

diff --git a/egs/ami/s5/local/ami_mdm_scoring_data_prep.sh b/egs/ami/s5/local/ami_mdm_scoring_data_prep.sh
@@ -67,7 +67,7 @@ sed -e 's?.*/??' -e 's?.wav??' $tmpdir/wav.flist | \
 awk '{print $2}' $tmpdir/segments | sort -u | join - $tmpdir/wav1.scp >  $tmpdir/wav2.scp
 
 #replace path with an appropriate sox command that select single channel only
-awk '{print $1" sox -c 1 -t wavpcm -s "$2" -t wavpcm - |"}' $tmpdir/wav2.scp > $tmpdir/wav.scp
+awk '{print $1" sox -c 1 -t wavpcm -e signed-integer "$2" -t wavpcm - |"}' $tmpdir/wav2.scp > $tmpdir/wav.scp
 
 #prep reco2file_and_channel
 cat $tmpdir/wav.scp | \

diff --git a/egs/ami/s5/local/ami_sdm_data_prep.sh b/egs/ami/s5/local/ami_sdm_data_prep.sh
@@ -74,7 +74,7 @@ awk '{print $1}' $dir/wav2.scp | join -2 2 - $dir/segments | \
 awk '{print $1}' $dir/segments | join - $dir/text > $dir/t; mv $dir/t $dir/text
 
 #replace path with an appropriate sox command that select a single channel only
-awk '{print $1" sox -c 1 -t wavpcm -s "$2" -t wavpcm - |"}' $dir/wav2.scp > $dir/wav.scp
+awk '{print $1" sox -c 1 -t wavpcm -e signed-integer "$2" -t wavpcm - |"}' $dir/wav2.scp > $dir/wav.scp
 
 # this file reco2file_and_channel maps recording-id
 cat $dir/wav.scp | \

diff --git a/egs/ami/s5/local/ami_sdm_scoring_data_prep.sh b/egs/ami/s5/local/ami_sdm_scoring_data_prep.sh
@@ -72,7 +72,7 @@ sed -e 's?.*/??' -e 's?.wav??' $tmpdir/wav.flist | \
 awk '{print $2}' $tmpdir/segments | sort -u | join - $tmpdir/wav1.scp > $tmpdir/wav2.scp
 
 #replace path with an appropriate sox command that select single channel only
-awk '{print $1" sox -c 1 -t wavpcm -s "$2" -t wavpcm - |"}' $tmpdir/wav2.scp > $tmpdir/wav.scp
+awk '{print $1" sox -c 1 -t wavpcm -e signed-integer "$2" -t wavpcm - |"}' $tmpdir/wav2.scp > $tmpdir/wav.scp
 
 #prep reco2file_and_channel
 cat $tmpdir/wav.scp | \

diff --git a/egs/ami/s5/local/online/run_nnet2_ms_perturbed.sh b/egs/ami/s5/local/online/run_nnet2_ms_perturbed.sh
@@ -8,7 +8,7 @@
 # This example script demonstrates how speed perturbation of the data helps the nnet training in the SWB setup.
 
 . ./cmd.sh
-set -e 
+set -e
 stage=1
 train_stage=-10
 use_gpu=true
@@ -27,13 +27,13 @@ fix_nnet=false
 
 if $use_gpu; then
   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 
+    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.  Otherwise, call this script with --use-gpu false
 EOF
   fi
-  parallel_opts="--gpu 1" 
+  parallel_opts="--gpu 1"
   num_threads=1
   minibatch_size=512
   if [[ $(hostname -f) == *.clsp.jhu.edu ]]; then
@@ -52,7 +52,7 @@ else
   # almost the same, but this may be a little bit slow.
   num_threads=16
   minibatch_size=128
-  parallel_opts="-pe smp $num_threads" 
+  parallel_opts="--num-threads $num_threads"
 fi
 
 dir=exp/$mic/nnet2_online/nnet_ms_sp${affix:+_$affix}
@@ -155,7 +155,7 @@ fi
 wait;
 
 if [ $stage -le 12 ]; then
-  # do the actual online decoding with iVectors, carrying info forward from 
+  # do the actual online decoding with iVectors, carrying info forward from
   # previous utterances of the same speaker.
   for decode_set in dev eval; do
     (
@@ -191,7 +191,7 @@ if [ $stage -le 14 ]; then
       steps/online/nnet2/decode.sh --config conf/decode.conf --cmd "$decode_cmd" --nj $num_jobs \
         --per-utt true --online false $graph_dir data/$mic/${decode_set}_hires \
           $decode_dir || exit 1;
-    ) & 
+    ) &
   done
 fi
 wait;

diff --git a/egs/ami/s5/local/online/run_nnet2_ms_sp_disc.sh b/egs/ami/s5/local/online/run_nnet2_ms_sp_disc.sh
@@ -6,7 +6,7 @@
 # note: this relies on having a cluster that has plenty of CPUs as well as GPUs,
 # since the lattice generation runs in about real-time, so takes of the order of
 # 1000 hours of CPU time.
-# 
+#
 # Note: rather than using any features we have dumped on disk, this script
 # regenerates them from the wav data three times-- when we do lattice
 # generation, numerator alignment and discriminative training.  This made the
@@ -42,19 +42,19 @@ set -e
 
 if $use_gpu; then
   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 
+    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.  Otherwise, call this script with --use-gpu false
 EOF
   fi
-  parallel_opts=" -l gpu=1,hostname='!g01*&!g02*' " #we want to submit to all.q as we use multiple GPUs for this 
+  parallel_opts="--gpu 1" #we want to submit to all.q as we use multiple GPUs for this
   num_threads=1
 else
   # Use 4 nnet jobs just like run_4d_gpu.sh so the results should be
   # almost the same, but this may be a little bit slow.
   num_threads=16
-  parallel_opts="-pe smp $num_threads" 
+  parallel_opts="--num-threads $num_threads"
 fi
 
 if [ -z $srcdir ]; then
@@ -76,7 +76,7 @@ if [ $stage -le 1 ]; then
   num_threads_denlats=6
   subsplit=40 # number of jobs that run per job (but 2 run at a time, so total jobs is 80, giving
               # max total slots = 80 * 6 = 480.
-  steps/nnet2/make_denlats.sh --cmd "$decode_cmd -l mem_free=1G,ram_free=1G -pe smp $num_threads_denlats" \
+  steps/nnet2/make_denlats.sh --cmd "$decode_cmd --mem 1G --num-threads $num_threads_denlats" \
       --online-ivector-dir exp/$mic/nnet2_online/ivectors_train_hires_sp2 \
       --nj $nj --sub-split $subsplit --num-threads "$num_threads_denlats" --config conf/decode.conf \
      data/$mic/train_hires_sp data/lang $srcdir ${srcdir}_denlats || exit 1;
@@ -142,7 +142,7 @@ if [ $stage -le 5 ]; then
       (
         num_jobs=`cat data/$mic/${decode_set}_hires/utt2spk|cut -d' ' -f2|sort -u|wc -l`
         decode_dir=$dir/decode_epoch${epoch}_${decode_set}_utt
-        
+
         steps/online/nnet2/decode.sh --config conf/decode.conf --cmd "$decode_cmd" --nj $num_jobs \
         --per-utt true  --iter epoch$epoch $graph_dir data/$mic/${decode_set}_hires $decode_dir || exit 1
       ) &
@@ -154,13 +154,13 @@ if [ $stage -le 5 ]; then
       (
         num_jobs=`cat data/$mic/${decode_set}_hires/utt2spk|cut -d' ' -f2|sort -u|wc -l`
         decode_dir=$dir/decode_epoch${epoch}_${decode_set}_utt_offline
-        
+
         steps/online/nnet2/decode.sh --config conf/decode.conf --cmd "$decode_cmd" --nj $num_jobs \
         --per-utt true --online false --iter epoch$epoch $graph_dir data/$mic/${decode_set}_hires $decode_dir || exit 1
       ) &
     done
   done
-  
+
   wait
 fi
 

diff --git a/egs/ami/s5b/README.txt b/egs/ami/s5b/README.txt
@@ -5,7 +5,7 @@ many components removed.
  Before running run.sh, please run run_prepare_shared.sh.
 
  Afterwards, you can run:
-    run.sh --mic ihm    #  builds system for independent headset microphone
+    run.sh --mic ihm    # builds system for independent headset microphone
     run.sh --mic sdm1   # single distant micropophone
     run.sh --mic mdm8   # multiple distant microphones + beamforming.
 

diff --git a/egs/ami/s5b/RESULTS_ihm b/egs/ami/s5b/RESULTS_ihm
@@ -40,7 +40,6 @@
 %WER 24.0 | 13098 94470 | 79.4 12.1 8.5 3.4 24.0 57.1 | -0.153 | exp/ihm/nnet3_cleaned/tdnn_sp/decode_dev/ascore_12/dev_hires.ctm.filt.sys
 %WER 25.5 | 12643 89984 | 77.7 14.2 8.2 3.2 25.5 56.4 | -0.139 | exp/ihm/nnet3_cleaned/tdnn_sp/decode_eval/ascore_11/eval_hires.ctm.filt.sys
 
-
 # local/nnet3/run_tdnn.sh --mic ihm --train-set train --gmm tri3 --nnet3-affix ""
 # nnet3 xent TDNN without data cleaning [cleaning makes very small and
 #  inconsistent difference on this dat]
@@ -55,21 +54,25 @@
 %WER 22.4 | 12643 89977 | 80.3 12.5 7.2 2.7 22.4 53.6 | -0.503 | exp/ihm/nnet3_cleaned/lstm_bidirectional_sp/decode_eval/ascore_10/eval_hires.ctm.filt.sys
 
 ############################################
-
-# local/chain/run_tdnn.sh --mic ihm --stage 12 &
-# cleanup + chain TDNN model
-# for d in exp/ihm/chain_cleaned/tdnn_sp_bi/decode_*; do grep Sum $d/*sc*/*ys | utils/best_wer.sh; done
-%WER 22.5 | 13098 94490 | 80.6 10.8 8.6 3.1 22.5 55.0 | 0.072 | exp/ihm/chain_cleaned/tdnn_sp_bi/decode_dev/ascore_10/dev_hires.ctm.filt.sys
-%WER 22.5 | 12643 89978 | 80.3 12.5 7.2 2.7 22.5 53.1 | 0.149 | exp/ihm/chain_cleaned/tdnn_sp_bi/decode_eval/ascore_10/eval_hires.ctm.filt.sys
-
-
-# local/chain/run_tdnn.sh --mic ihm --train-set train --gmm tri3 --nnet3-affix "" --stage 12
+# cleanup + chain TDNN model.
+# local/chain/run_tdnn.sh --mic ihm --stage 4 &
+# for d in exp/ihm/chain_cleaned/tdnn1d_sp_bi/decode_*; do grep Sum $d/*sc*/*ys | utils/best_wer.sh; done
+%WER 21.7 | 13098 94488 | 81.1 10.4 8.4 2.8 21.7 54.4 | 0.096 | exp/ihm/chain_cleaned/tdnn1d_sp_bi/decode_dev/ascore_10/dev_hires.ctm.filt.sys
+%WER 22.1 | 12643 89979 | 80.5 12.1 7.4 2.6 22.1 52.8 | 0.185 | exp/ihm/chain_cleaned/tdnn1d_sp_bi/decode_eval/ascore_10/eval_hires.ctm.filt.sys
+
+# cleanup + chain TDNN model. Uses LDA instead of PCA for ivector features.
+# local/chain/tuning/run_tdnn_1b.sh --mic ihm --stage 4 &
+# for d in exp/ihm/chain_cleaned/tdnn1b_sp_bi/decode_*; do grep Sum $d/*sc*/*ys | utils/best_wer.sh; done
+%WER 22.0 | 13098 94488 | 80.8 10.2 9.0 2.8 22.0 54.7 | 0.102 | exp/ihm/chain_cleaned/tdnn1b_sp_bi/decode_dev/ascore_10/dev_hires.ctm.filt.sys
+%WER 22.2 | 12643 89968 | 80.3 12.1 7.6 2.6 22.2 52.9 | 0.170 | exp/ihm/chain_cleaned/tdnn1b_sp_bi/decode_eval/ascore_10/eval_hires.ctm.filt.sys
+
+# local/chain/run_tdnn.sh --mic ihm --train-set train --gmm tri3 --nnet3-affix "" --stage 4
 # chain TDNN model without cleanup [note: cleanup helps very little on this IHM data.]
-for d in exp/ihm/chain/tdnn_sp_bi/decode_*; do grep Sum $d/*sc*/*ys | utils/best_wer.sh; done
-%WER 22.4 | 13098 94476 | 80.4 10.4 9.2 2.8 22.4 54.6 | 0.069 | exp/ihm/chain/tdnn_sp_bi/decode_dev/ascore_10/dev_hires.ctm.filt.sys
-%WER 22.5 | 12643 89974 | 80.0 12.1 7.9 2.6 22.5 52.8 | 0.157 | exp/ihm/chain/tdnn_sp_bi/decode_eval/ascore_10/eval_hires.ctm.filt.sys
+# for d in exp/ihm/chain/tdnn1d_sp_bi/decode_*; do grep Sum $d/*sc*/*ys | utils/best_wer.sh; done
+%WER 21.8 | 13098 94484 | 80.7 9.7 9.6 2.5 21.8 54.2 | 0.114 | exp/ihm/chain/tdnn1d_sp_bi/decode_dev/ascore_10/dev_hires.ctm.filt.sys
+%WER 22.1 | 12643 89965 | 80.2 11.5 8.3 2.3 22.1 52.5 | 0.203 | exp/ihm/chain/tdnn1d_sp_bi/decode_eval/ascore_10/eval_hires.ctm.filt.sy
 
-# local/chain/multi_condition/run_tdnn.sh --mic ihm &
+# local/chain/multi_condition/run_tdnn.sh --mic ihm
 # cleanup + chain TDNN model + IHM reverberated data
 # for d in exp/ihm/chain_cleaned_rvb/tdnn_sp_bi/decode_*; do grep Sum $d/*sc*/*ys | utils/best_wer.sh; done
 %WER 21.5 | 13098 94486 | 81.8 11.0 7.2 3.3 21.5 54.6 | 0.090 | exp/ihm/chain_cleaned_rvb/tdnn_sp_rvb_bi/decode_dev/ascore_10/dev_hires.ctm.filt.sys
@@ -80,3 +83,15 @@ for d in exp/ihm/chain/tdnn_sp_bi/decode_*; do grep Sum $d/*sc*/*ys | utils/best
 # cleanup + chain TDNN+LSTM model
 %WER 20.8 | 13098 94489 | 82.0 10.0 8.0 2.8 20.8 53.2 | -0.096 | exp/ihm/chain_cleaned/tdnn_lstm1i_sp_bi_ld5/decode_dev/ascore_11/dev_hires.ctm.filt.sys
 %WER 20.7 | 12643 89980 | 81.7 11.5 6.8 2.5 20.7 51.8 | 0.015 | exp/ihm/chain_cleaned/tdnn_lstm1i_sp_bi_ld5/decode_eval/ascore_11/eval_hires.ctm.filt.sys
+
+# local/chain/tuning/run_tdnn_lstm_1l.sh --mic ihm  --train-set train_cleaned  --gmm tri3_cleaned
+# same as local/chain/tuning/run_tdnn_lstm_1i.sh, except that dropout is adopted
+# cleanup + chain TDNN+LSTM model + per-frame dropout
+%WER 19.8 | 13098 94475 | 83.1 9.6 7.4 2.8 19.8 51.8 | -0.041 | exp/ihm/chain_cleaned/tdnn_lstm1l_sp_bi_ld5/decode_dev/ascore_10/dev_hires.ctm.filt.sys
+%WER 19.2 | 12643 89964 | 83.2 10.7 6.1 2.5 19.2 49.7 | 0.079 | exp/ihm/chain_cleaned/tdnn_lstm1l_sp_bi_ld5/decode_eval/ascore_10/eval_hires.ctm.filt.sys
+
+# local/chain/multi_condition/tuning/run_tdnn_lstm_1a.sh --mic ihm
+# cleanup + chain TDNN+LSTM model + IHM reverberated data
+%WER 19.4 | 13098 94479 | 83.8 10.0 6.1 3.2 19.4 51.8 | -0.168 | exp/ihm/chain_cleaned_rvb/tdnn_lstm1i_sp_rvb_bi/decode_dev/ascore_10/dev_hires.ctm.filt.sys
+%WER 19.3 | 12643 89977 | 83.3 11.0 5.7 2.6 19.3 49.6 | -0.046 | exp/ihm/chain_cleaned_rvb/tdnn_lstm1i_sp_rvb_bi/decode_eval/ascore_10/eval_hires.ctm.filt.sys
+