refactor kaldi_decode

1 files changed, 134 insertions, 0 deletions

diff --git a/kaldi_decode/decode_with_nerv.sh b/kaldi_decode/decode_with_nerv.sh
new file mode 100755
index 0000000..5554b2e
--- /dev/null
+++ b/kaldi_decode/decode_with_nerv.sh
@@ -0,0 +1,134 @@
+#!/bin/bash
+# Copyright 2012-2013 Karel Vesely, Daniel Povey
+# Apache 2.0
+
+# Begin configuration section. 
+nnet=               # non-default location of DNN (optional)
+feature_transform=  # non-default location of feature_transform (optional)
+model=              # non-default location of transition model (optional)
+class_frame_counts= # non-default location of PDF counts (optional)
+srcdir=             # non-default location of DNN-dir (decouples model dir from decode dir)
+
+stage=0 # stage=1 skips lattice generation
+nj=4
+cmd=run.pl
+
+acwt=0.10 # note: only really affects pruning (scoring is on lattices).
+beam=13.0
+lattice_beam=8.0
+min_active=200
+max_active=7000 # limit of active tokens
+max_mem=50000000 # approx. limit to memory consumption during minimization in bytes
+nnet_forward_opts="--apply-log=true" # IMPORTANT, to apply log before to substract log-prior, and to know the modified 'nnet-forward' removed '--no-softmax' option
+
+skip_scoring=false
+scoring_opts="--min-lmwt 4 --max-lmwt 15"
+
+num_threads=1 # if >1, will use latgen-faster-parallel
+parallel_opts=   # Ignored now.
+use_gpu="no" # yes|no|optionaly
+
+cmvn_opts=
+splice_opts=
+delta_opts=
+
+asr_propagator=
+nnet_forward=
+# End configuration section.
+
+echo "$0 $@"  # Print the command line for logging
+
+[ -f ./path.sh ] && . ./path.sh; # source the path.
+. parse_options.sh || exit 1;
+
+if [ $# != 4 ]; then
+   echo "Usage: $0 [options] <graph-dir> <data-dir> <nerv-model-config> <decode-dir>"
+   echo "... where <decode-dir> is assumed to be a sub-directory of the directory"
+   echo " where the DNN and transition model is."
+   echo "e.g.: $0 exp/dnn1/graph_tgpr data/test config.lua exp/dnn1/decode_tgpr"
+   echo ""
+   echo "This script works on plain or modified features (CMN,delta+delta-delta),"
+   echo "which are then sent through feature-transform. It works out what type"
+   echo "of features you used from content of srcdir."
+   echo ""
+   echo "main options (for others, see top of script file)"
+   echo "  --config <config-file>                           # config containing options"
+   echo "  --nj <nj>                                        # number of parallel jobs"
+   echo "  --cmd (utils/run.pl|utils/queue.pl <queue opts>) # how to run jobs."
+   echo ""
+   echo "  --srcdir <dir>                                   # non-default dir with DNN/models, can be different"
+   echo "                                                   # from parent dir of <decode-dir>' (opt.)"
+   echo ""
+   echo "  --acwt <float>                                   # select acoustic scale for decoding"
+   echo "  --scoring-opts <opts>                            # options forwarded to local/score.sh"
+   echo "  --num-threads <N>                                # N>1: run multi-threaded decoder"
+   exit 1;
+fi
+
+
+graphdir=$1
+data=$2
+model_conf=$3
+dir=$4
+
+[ -z $srcdir ] && srcdir=`dirname $dir`; # Default model directory one level up from decoding directory.
+sdata=$data/split$nj;
+
+mkdir -p $dir/log
+
+[[ -d $sdata && $data/feats.scp -ot $sdata ]] || split_data.sh $data $nj || exit 1;
+echo $nj > $dir/num_jobs
+
+# Select default locations to model files (if not already set externally)
+[ -z "$model" ] && model=$srcdir/final.mdl
+#
+[ -z "$class_frame_counts" -a -f $srcdir/prior_counts ] && class_frame_counts=$srcdir/prior_counts # priority,
+[ -z "$class_frame_counts" ] && class_frame_counts=$srcdir/ali_train_pdf.counts
+
+# Check that files exist
+for f in $sdata/1/feats.scp $model $class_frame_counts $graphdir/HCLG.fst; do
+  [ ! -f $f ] && echo "$0: missing file $f" && exit 1;
+done
+
+# Possibly use multi-threaded decoder
+thread_string=
+[ $num_threads -gt 1 ] && thread_string="-parallel --num-threads=$num_threads" 
+
+
+# PREPARE FEATURE EXTRACTION PIPELINE
+# import config,
+D=$srcdir
+[ -e $D/norm_vars ] && cmvn_opts="--norm-means=true --norm-vars=$(cat $D/norm_vars)" # Bwd-compatibility,
+[ -e $D/cmvn_opts ] && cmvn_opts=$(cat $D/cmvn_opts)
+[ -e $D/splice_opts ] && splice_opts=$(cat $D/splice_opts)
+[ -e $D/delta_order ] && delta_opts="--delta-order=$(cat $D/delta_order)" # Bwd-compatibility,
+[ -e $D/delta_opts ] && delta_opts=$(cat $D/delta_opts)
+#
+# Create the feature stream,
+feats="ark,s,cs:copy-feats scp:$sdata/JOB/feats.scp ark:- |"
+# apply-cmvn (optional),
+[ ! -z "$cmvn_opts" -a ! -f $sdata/1/cmvn.scp ] && echo "$0: Missing $sdata/1/cmvn.scp" && exit 1
+[ ! -z "$cmvn_opts" ] && feats="$feats apply-cmvn $cmvn_opts --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp ark:- ark:- |"
+# splice-opts (optional),
+[ ! -z "$splice_opts" ] && feats="$feats splice-feats $splice_opts ark:- ark:- |"
+# add-deltas (optional),
+[ ! -z "$delta_opts" ] && feats="$feats add-deltas $delta_opts ark:- ark:- |"
+#
+# Run the decoding in the queue,
+if [ $stage -le 0 ]; then
+#  $cmd --num-threads $((num_threads+1)) JOB=1:$nj $dir/log/decode.JOB.log \
+# remove multi-threads to avoid smp requirement
+  $cmd --num-threads $((num_threads)) JOB=1:$nj $dir/log/decode.JOB.log \
+    $nnet_forward $nnet_forward_opts --class-frame-counts=$class_frame_counts --use-gpu=$use_gpu $model_conf "$feats" ark:- $asr_propagator \| \
+    latgen-faster-mapped$thread_string --min-active=$min_active --max-active=$max_active --max-mem=$max_mem --beam=$beam \
+    --lattice-beam=$lattice_beam --acoustic-scale=$acwt --allow-partial=true --word-symbol-table=$graphdir/words.txt \
+    $model $graphdir/HCLG.fst ark:- "ark:|gzip -c > $dir/lat.JOB.gz" || exit 1;
+fi
+# Run the scoring
+if ! $skip_scoring ; then
+  [ ! -x local/score.sh ] && \
+    echo "Not scoring because local/score.sh does not exist or not executable." && exit 1;
+  local/score.sh $scoring_opts --cmd "$cmd" $data $graphdir $dir || exit 1;
+fi
+
+exit 0;