Skip to content
Snippets Groups Projects
Commit 356538a1 authored by Dan Povey's avatar Dan Povey
Browse files

Adding some scripts (train_sgmm.sh not finished yet!) 

git-svn-id: https://svn.code.sf.net/p/kaldi/code/trunk@904 5e6a8d80-dfce-4ca6-a32a-6e07a63d50c8
parent 143fd26d
No related branches found
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
egs/wsj/s5/steps/train_quick.sh 0 → 100755
+ 186
− 0
View file @ 356538a1
#!/bin/bash
# Copyright 2012 Daniel Povey. Apache 2.0.
# Train a model on top of existing features (no feature-space learning of any
# kind is done). This script initializes the model from each stage of the
# previous system's model, judging the similarities based on overlap of counts
# in the tree stats.
# Begin configuration..
cmd=run.pl
scale_opts="--transition-scale=1.0 --acoustic-scale=0.1 --self-loop-scale=0.1"
realign_iters="10 15"; # Only realign twice.
num_iters=20 # Number of iterations of training
maxiterinc=15 # Last iter to increase #Gauss on.
batch_size=750 # batch size to use while compiling graphs... memory/speed tradeoff.
beam=10 # alignment beam.
retry_beam=40
stage=-5
# End configuration section.
[ -f path.sh ] && . ./path.sh
. parse_options.sh || exit 1;
if [ $# != 6 ]; then
echo "Usage: steps/train_quick.sh <num-leaves> <num-gauss> <data> <lang> <ali-dir> <exp-dir>"
echo " e.g.: steps/train_quick.sh 2500 15000 data/train_si284 data/lang exp/tri3c_ali_si284 exp/tri4b"
echo "Main options (for others, see top of script file)"
echo " --cmd (utils/run.pl|utils/queue.pl <queue opts>) # how to run jobs."
echo " --config <config-file> # config containing options"
echo " --stage <stage> # stage to do partial re-run from."
exit 1;
fi
numleaves=$1
totgauss=$2
data=$3
lang=$4
alidir=$5
dir=$6
for f in $data/feats.scp $lang/L.fst $alidir/ali.1.gz $alidir/final.mdl; do
[ ! -f $f ] && echo "$0: no such file $f" && exit 1;
done
# Set various variables.
oov=`cat $lang/oov.int`
silphonelist=`cat $lang/phones/silence.csl`
ciphonelist=`cat $lang/phones/context_indep.csl`
numgauss=$[totgauss/2] # Start with half the total number of Gaussians. We won't have
# to mix up much probably, as we're initializing with the old (already mixed-up) pdf's.
[ $numgauss -lt $numleaves ] && numgauss=$numleaves
incgauss=$[($totgauss-$numgauss)/$maxiterinc] # per-iter increment for #Gauss
nj=`cat $alidir/num_jobs` || exit 1;
sdata=$data/split$nj
mkdir -p $dir/log
echo $nj >$dir/num_jobs
[[ -d $sdata && $data/feats.scp -ot $sdata ]] || split_data.sh $data $nj || exit 1;
## Set up features.
if [ -f $alidir/final.mat ]; then feat_type=lda; else feat_type=delta; fi
echo "$0: feature type is $feat_type"
case $feat_type in
delta) sifeats="ark,s,cs:apply-cmvn --norm-vars=false --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | add-deltas ark:- ark:- |";;
lda) sifeats="ark,s,cs:apply-cmvn --norm-vars=false --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | splice-feats ark:- ark:- | transform-feats $alidir/final.mat ark:- ark:- |"
cp $alidir/final.mat $dir
;;
*) echo "Invalid feature type $feat_type" && exit 1;
esac
if [ -f $alidir/1.trans ]; then
echo "$0: using transforms from $alidir"
ln.pl $alidir/*.trans $dir # Link them to dest dir.
feats="$sifeats transform-feats --utt2spk=ark:$sdata/JOB/utt2spk ark,s,cs:$dir/JOB.trans ark:- ark:- |"
fi
##
if [ $stage -le -5 ]; then
echo "$0: accumulating tree stats"
$cmd JOB=1:$nj $dir/log/acc_tree.JOB.log \
acc-tree-stats --ci-phones=$ciphonelist $alidir/final.mdl "$feats" \
"ark:gunzip -c $alidir/ali.JOB.gz|" $dir/JOB.treeacc || exit 1;
[ "`ls $dir/*.treeacc | wc -w`" -ne "$nj" ] && echo "$0: Wrong #tree-stats" && exit 1;
sum-tree-stats $dir/treeacc $dir/*.treeacc 2>$dir/log/sum_tree_acc.log || exit 1;
rm $dir/*.treeacc
fi
if [ $stage -le -4 ]; then
echo "$0: Getting questions for tree clustering."
# preparing questions, roots file...
cluster-phones $dir/treeacc $lang/phones/sets.int $dir/questions.int 2> $dir/log/questions.log || exit 1;
cat $lang/phones/extra_questions.int >> $dir/questions.int
compile-questions $lang/topo $dir/questions.int $dir/questions.qst 2>$dir/log/compile_questions.log || exit 1;
echo "$0: Building the tree"
$cmd $dir/log/build_tree.log \
build-tree --verbose=1 --max-leaves=$numleaves \
$dir/treeacc $lang/phones/roots.int \
$dir/questions.qst $lang/topo $dir/tree || exit 1;
fi
if [ $stage -le -3 ]; then
echo "$0: Initializing the model"
# The gmm-init-model command (with more than the normal # of command-line args)
# will initialize the p.d.f.'s to the p.d.f.'s in the alignment model.
gmm-init-model --write-occs=$dir/1.occs \
$dir/tree $dir/treeacc $lang/topo $dir/tmp.mdl $alidir/tree $alidir/final.mdl \
2>$dir/log/init_model.log || exit 1;
grep 'no stats' $dir/log/init_model.log && echo "$0: This is a bad warning.";
rm $dir/treeacc
fi
if [ $stage -le -2 ]; then
echo "$0: mixing up old model."
# We do both mixing-down and mixing-up to get the target #Gauss in each state,
# since the initial model may have either more or fewer Gaussians than we want.
gmm-mixup --mix-down=$numgauss --mix-up=$numgauss $dir/tmp.mdl $dir/1.occs $dir/1.mdl \
2> $dir/log/mixup.log || exit 1;
rm $dir/tmp.mdl
fi
# Convert alignments to the new tree.
if [ $stage -le -1 ]; then
echo "$0: converting old alignments"
$cmd JOB=1:$nj $dir/log/convert.JOB.log \
convert-ali $alidir/final.mdl $dir/1.mdl $dir/tree \
"ark:gunzip -c $alidir/ali.JOB.gz|" "ark:|gzip -c >$dir/ali.JOB.gz" || exit 1;
fi
if [ $stage -le 0 ]; then
echo "$0: compiling training graphs"
$cmd JOB=1:$nj $dir/log/compile_graphs.JOB.log \
compile-train-graphs --batch-size=$batch_size $dir/tree $dir/1.mdl $lang/L.fst \
"ark:sym2int.pl --map-oov $oov -f 2- $lang/words.txt < $sdata/JOB/text |" \
"ark:|gzip -c >$dir/fsts.JOB.gz" || exit 1;
fi
x=1
while [ $x -lt $num_iters ]; do
echo "$0: pass $x"
if echo $realign_iters | grep -w $x >/dev/null && [ $stage -le $x ]; then
echo "$0: aligning data"
$cmd JOB=1:$nj $dir/log/align.$x.JOB.log \
gmm-align-compiled $scale_opts --beam=$beam --retry-beam=$retry_beam $dir/$x.mdl \
"ark:gunzip -c $dir/fsts.JOB.gz|" "$feats" "ark:|gzip -c >$dir/ali.JOB.gz" \
|| exit 1;
fi
if [ $stage -le $x ]; then
$cmd JOB=1:$nj $dir/log/acc.$x.JOB.log \
gmm-acc-stats-ali $dir/$x.mdl "$feats" \
"ark,s,cs:gunzip -c $dir/ali.JOB.gz|" $dir/$x.JOB.acc || exit 1;
[ "`ls $dir/$x.*.acc | wc -w`" -ne "$nj" ] && echo "$0: wrong #accs" && exit 1;
$cmd $dir/log/update.$x.log \
gmm-est --write-occs=$dir/$[$x+1].occs --mix-up=$numgauss $dir/$x.mdl \
"gmm-sum-accs - $dir/$x.*.acc |" $dir/$[$x+1].mdl || exit 1;
rm $dir/$x.mdl $dir/$x.*.acc $dir/$x.occs
fi
[[ $x -le $maxiterinc ]] && numgauss=$[$numgauss+$incgauss];
x=$[$x+1];
done
if [ -f $alidir/1.trans ]; then
echo "$0: estimating alignment model"
$cmd JOB=1:$nj $dir/log/acc_alimdl.JOB.log \
ali-to-post "ark:gunzip -c $dir/ali.JOB.gz|" ark:- \| \
gmm-acc-stats-twofeats $dir/$x.mdl "$feats" "$sifeats" \
ark,s,cs:- $dir/$x.JOB.acc || exit 1;
[ "`ls $dir/$x.*.acc | wc -w`" -ne "$nj" ] && echo "$0: wrong #accs" && exit 1;
$cmd $dir/log/est_alimdl.log \
gmm-est --write-occs=$dir/final.occs --remove-low-count-gaussians=false $dir/$x.mdl \
"gmm-sum-accs - $dir/$x.*.acc|" $dir/$x.alimdl || exit 1;
rm $dir/$x.*.acc
rm $dir/final.alimdl 2>/dev/null
ln -s $x.alimdl $dir/final.alimdl
fi
rm $dir/final.mdl 2>/dev/null
ln -s $x.mdl $dir/final.mdl
echo Done
egs/wsj/s5/steps/train_sgmm.sh 0 → 100755
+ 311
− 0
View file @ 356538a1
#!/bin/bash
# Copyright 2012 Daniel Povey. Apache 2.0.
# SGMM training, with speaker vectors. This script would normally be called on
# top of fMLLR features obtained from a conventional system, but it also works
# on top of any type of speaker-independent features (based on
# deltas+delta-deltas or LDA+MLLT). For more info on SGMMs, see the paper "The
# subspace Gaussian mixture model--A structured model for speech recognition".
# (Computer Speech and Language, 2011).
# Begin configuration section.
nj=4
cmd=scripts/run.pl
stage=-5
context_opts= # e.g. set it to "--context-width=5 --central-position=2" for a
# quinphone system.
scale_opts="--transition-scale=1.0 --acoustic-scale=0.1 --self-loop-scale=0.1"
num_iters=25 # Total number of iterations
num_iters_alimdl=3 # Number of iterations for estimating alignment model.
maxiterinc=15 # Last iter to increase #substates on.
realign_iters="5 10 15"; # Iters to realign on.
spkvec_iters="5 8 12 17" # Iters to estimate speaker vectors on.
add_dim_iters="6 8 10 12"; # Iters on which to increase phn dim and/or spk dim,
rand_prune=0.1 # Randomized-pruning parameter for posteriors, to speed up training.
phn_dim= # You can use this to set the phonetic subspace dim. [default: feat-dim+1]
spk_dim= # You can use this to set the speaker subspace dim. [default: feat-dim]
# End configuration section.
if [ -f path.sh ]; then . ./path.sh; fi
. parse_options.sh || exit 1;
if [ $# != 7 ]; then
echo "Usage: steps/train_sgmm.sh <num-leaves> <num-substates> <data> <lang> <ali-dir> <ubm-dir> <exp-dir>"
echo " e.g.: steps/train_sgmm.sh 3500 10000 data/train_si84 data/lang \\"
echo " exp/tri3b_ali_si84 exp/ubm4a/final.ubm exp/sgmm4a"
echo "main options (for others, see top of script file)"
echo " --config <config-file> # config containing options"
echo " --cmd (utils/run.pl|utils/queue.pl <queue opts>) # how to run jobs."
echo " --silence-weight <sil-weight> # weight for silence (e.g. 0.5 or 0.0)"
echo " --num-iters <#iters> # Number of iterations of E-M"
exit 1;
fi
num_leaves=$1
totsubstates=$2
data=$3
lang=$4
alidir=$5
ubm=$6
dir=$7
# Check some files.
for f in $data/feats.scp $lang/L.fst $alidir/ali.1.gz $alidir/final.mdl $ubm/final.ubm; do
[ ! -f $f ] && echo "$0: no such file $f" && exit 1;
done
cp $transform_dir/final.mat $dir/final.mat || exit 1;
# Set some variables.
oov=`cat $lang/oov.int`
silphonelist=`cat $lang/phones/silence.csl`
numsubstates=$num_leaves # Initial #-substates.
incsubstates=$[($totsubstates-$numsubstates)/$maxiterinc] # per-iter increment for #substates
feat_dim=`gmm-info $alidir/final.model | awk '/feature dimension/{print $NF}'` || exit 1;
[ $feat_dim -eq $feat_dim ] || exit 1; # make sure it's numeric.
[ -z $phn_dim ] && phn_dim=$[$feat_dim+1]
[ -z $spk_dim ] && spk_dim=$feat_dim
nj=`cat $alidir/num_jobs` || exit 1;
mkdir -p $dir/log
echo $nj > $dir/num_jobs
sdata=$data/split$nj;
[[ -d $sdata && $data/feats.scp -ot $sdata ]] || split_data.sh $data $nj || exit 1;
for n in `get_splits.pl $nj`; do
# Initially don't have speaker vectors, but change this after we estimate them.
spkvecs_opt[$n]=
gselect_opt[$n]="--gselect=ark,s,cs:gunzip -c $dir/$n.gselect.gz|"
done
n1=`get_splits.pl $nj | awk '{print $1}'`
[ -f $transform_dir/$n1.trans ] && echo "Using speaker transforms from $transform_dir"
for n in `get_splits.pl $nj`; do
featspart[$n]="ark,s,cs:apply-cmvn --norm-vars=false --utt2spk=ark:$data/split$nj/$n/utt2spk ark:$transform_dir/$n.cmvn scp:$data/split$nj/$n/feats.scp ark:- | splice-feats ark:- ark:- | transform-feats $dir/final.mat ark:- ark:- |"
if [ -f $transform_dir/$n1.trans ]; then
featspart[$n]="${featspart[$n]} transform-feats --utt2spk=ark:$data/split$nj/$n/utt2spk ark:$transform_dir/$n.trans ark:- ark:- |"
fi
done
if [ ! -f $ubm ]; then
echo "No UBM in $ubm"
exit 1;
fi
if [ $stage -le -5 ]; then
# This stage assumes we won't need the context of silence, which
# assumes something about $lang/roots.txt, but it seems pretty safe.
echo "Accumulating tree stats"
rm $dir/.error 2>/dev/null
for n in `get_splits.pl $nj`; do
$cmd $dir/log/acc_tree.$n.log \
acc-tree-stats $context_opts --ci-phones=$silphonelist $alidir/final.mdl "${featspart[$n]}" \
"ark:gunzip -c $alidir/$n.ali.gz|" $dir/$n.treeacc || touch $dir/.error &
done
wait
[ -f $dir/.error ] && echo Error accumulating tree stats && exit 1;
sum-tree-stats $dir/treeacc $dir/*.treeacc 2>$dir/log/sum_tree_acc.log || exit 1;
rm $dir/*.treeacc
fi
if [ $stage -le -4 ]; then
echo "Computing questions for tree clustering"
# preparing questions, roots file...
sym2int.pl $lang/phones.txt $lang/phonesets_cluster.txt > $dir/phonesets.txt || exit 1;
cluster-phones $context_opts $dir/treeacc $dir/phonesets.txt $dir/questions.txt 2> $dir/log/questions.log || exit 1;
sym2int.pl $lang/phones.txt $lang/extra_questions.txt >> $dir/questions.txt
compile-questions $context_opts $lang/topo $dir/questions.txt $dir/questions.qst 2>$dir/log/compile_questions.log || exit 1;
sym2int.pl --ignore-oov $lang/phones.txt $lang/roots.txt > $dir/roots.txt
echo "Building tree"
$cmd $dir/log/train_tree.log \
build-tree $context_opts --verbose=1 --max-leaves=$num_leaves \
$dir/treeacc $dir/roots.txt \
$dir/questions.qst $lang/topo $dir/tree || exit 1;
# The next line is a bit of a hack to work out the feature dim. The program
# feat-to-len returns the #rows of each matrix, which for the transform matrix,
# is the feature dim.
featdim=`feat-to-len "scp:echo foo $transform_dir/final.mat|" ark,t:- 2>/dev/null | awk '{print $2}'`
# Note: if phn_dim and/or spk_dim are higher than you can initialize with,
# sgmm-init will just make them as high as it can (later we'll increase)
$cmd $dir/log/init_sgmm.log \
sgmm-init --phn-space-dim=$phn_dim --spk-space-dim=$spk_dim $lang/topo $dir/tree $ubm \
$dir/0.mdl || exit 1;
fi
rm $dir/.error 2>/dev/null
if [ $stage -le -3 ]; then
echo "Doing Gaussian selection"
for n in `get_splits.pl $nj`; do
$cmd $dir/log/gselect$n.log \
sgmm-gselect $dir/0.mdl "${featspart[$n]}" "ark,t:|gzip -c > $dir/$n.gselect.gz" \
|| touch $dir/.error &
done
wait;
[ -f $dir/.error ] && echo "Error doing Gaussian selection" && exit 1;
fi
if [ $stage -le -2 ]; then
echo "Compiling training graphs"
for n in `get_splits.pl $nj`; do
$cmd $dir/log/compile_graphs$n.log \
compile-train-graphs $dir/tree $dir/0.mdl $lang/L.fst \
"ark:sym2int.pl --map-oov $oov --ignore-first-field $lang/words.txt < $data/split$nj/$n/text |" \
"ark:|gzip -c >$dir/$n.fsts.gz" || touch $dir/.error &
done
wait;
[ -f $dir/.error ] && echo "Error compiling training graphs" && exit 1;
fi
if [ $stage -le -1 ]; then
echo "Converting alignments" # don't bother parallelizing; very fast.
for n in `get_splits.pl $nj`; do
convert-ali $alidir/final.mdl $dir/0.mdl $dir/tree "ark:gunzip -c $alidir/$n.ali.gz|" \
"ark:|gzip -c >$dir/$n.ali.gz" 2>$dir/log/convert$n.log
done
fi
x=0
while [ $x -lt $num_iters ]; do
echo "Pass $x ... "
if echo $realign_iters | grep -w $x >/dev/null; then
if [ $stage -le $x ]; then
echo "Aligning data"
for n in `get_splits.pl $nj`; do
$cmd $dir/log/align.$x.$n.log \
sgmm-align-compiled ${spkvecs_opt[$n]} $scale_opts "${gselect_opt[$n]}" \
--utt2spk=ark:$data/split$nj/$n/utt2spk --beam=8 --retry-beam=40 \
$dir/$x.mdl "ark:gunzip -c $dir/$n.fsts.gz|" "${featspart[$n]}" \
"ark:|gzip -c >$dir/$n.ali.gz" || touch $dir/.error &
done
wait;
[ -f $dir/.error ] && echo "Error realigning data on iter $x" && exit 1;
fi
fi
if [ $spk_dim -gt 0 ] && echo $spkvec_iters | grep -w $x >/dev/null; then
for n in `get_splits.pl $nj`; do
if [ $stage -le $x ]; then
$cmd $dir/log/spkvecs.$x.$n.log \
ali-to-post "ark:gunzip -c $dir/$n.ali.gz|" ark:- \| \
weight-silence-post 0.01 $silphonelist $dir/$x.mdl ark:- ark:- \| \
sgmm-est-spkvecs --spk2utt=ark:$data/split$nj/$n/spk2utt \
${spkvecs_opt[$n]} "${gselect_opt[$n]}" $dir/$x.mdl \
"${featspart[$n]}" ark,s,cs:- ark:$dir/tmp$n.vecs \
&& mv $dir/tmp$n.vecs $dir/$n.vecs || touch $dir/.error &
fi
spkvecs_opt[$n]="--spk-vecs=ark:$dir/$n.vecs"
done
wait;
[ -f $dir/.error ] && echo "Error computing speaker vectors on iter $x" && exit 1;
fi
if [ $x -eq 0 ]; then
flags=vwcSt # On first iter, don't update M or N.
elif [ $spk_dim -gt 0 -a $[$x%2] -eq 1 -a $x -ge `echo $spkvec_iters | awk '{print $1}'` ]; then
# Update N if we have spk-space and x is odd, and we're at least at 1st spkvec iter.
flags=vNwcSt
else # Else update M but not N.
flags=vMwcSt
fi
if [ $stage -le $x ]; then
for n in `get_splits.pl $nj`; do
$cmd $dir/log/acc.$x.$n.log \
sgmm-acc-stats ${spkvecs_opt[$n]} --utt2spk=ark:$data/split$nj/$n/utt2spk \
--update-flags=$flags "${gselect_opt[$n]}" --rand-prune=$randprune \
$dir/$x.mdl "${featspart[$n]}" "ark,s,cs:ali-to-post 'ark:gunzip -c $dir/$n.ali.gz|' ark:-|" \
$dir/$x.$n.acc || touch $dir/.error &
done
wait;
[ -f $dir/.error ] && echo "Error accumulating stats on iter $x" && exit 1;
fi
add_dim_opts=
if echo $add_dim_iters | grep -w $x >/dev/null; then
add_dim_opts="--increase-phn-dim=$phn_dim --increase-spk-dim=$spk_dim"
fi
if [ $stage -le $x ]; then
$cmd $dir/log/update.$x.log \
sgmm-est --update-flags=$flags --split-substates=$numsubstates $add_dim_opts \
--write-occs=$dir/$[$x+1].occs $dir/$x.mdl "sgmm-sum-accs - $dir/$x.*.acc|" \
$dir/$[$x+1].mdl || exit 1;
rm $dir/$x.mdl $dir/$x.*.acc
rm $dir/$x.occs
fi
if [ $x -lt $maxiterinc ]; then
numsubstates=$[$numsubstates+$incsubstates]
fi
x=$[$x+1];
done
( cd $dir; rm final.mdl final.occs 2>/dev/null;
ln -s $x.mdl final.mdl;
ln -s $x.occs final.occs )
if [ $spk_dim -gt 0 ]; then
# If we have speaker vectors, we need an alignment model.
# The point of this last phase of accumulation is to get Gaussian-level
# alignments with the speaker vectors but accumulate stats without
# any speaker vectors; we re-estimate M, w, c and S to get a model
# that's compatible with not having speaker vectors.
# We do this for a few iters, in this recipe.
cur_alimdl=$dir/$x.mdl
y=0;
while [ $y -lt $num_iters_alimdl ]; do
echo "Pass $y of building alignment model"
if [ $y -eq 0 ]; then
flags=MwcS # First time don't update v...
else
flags=vMwcS # don't update transitions-- will probably share graph with normal model.
fi
if [ $stage -le $[$y+100] ]; then
for n in `get_splits.pl $nj`; do
$cmd $dir/log/acc_ali.$y.$n.log \
ali-to-post "ark:gunzip -c $dir/$n.ali.gz|" ark:- \| \
sgmm-post-to-gpost ${spkvecs_opt[$n]} "${gselect_opt[$n]}" \
--utt2spk=ark:$data/split$nj/$n/utt2spk $dir/$x.mdl "${featspart[$n]}" ark,s,cs:- ark:- \| \
sgmm-acc-stats-gpost --update-flags=$flags $cur_alimdl "${featspart[$n]}" \
ark,s,cs:- $dir/$y.$n.aliacc || touch $dir/.error &
done
wait;
[ -f $dir/.error ] && echo "Error accumulating stats for alignment model on iter $y" && exit 1;
$cmd $dir/log/update_ali.$y.log \
sgmm-est --update-flags=$flags --remove-speaker-space=true $cur_alimdl \
"sgmm-sum-accs - $dir/$y.*.aliacc|" $dir/$[$y+1].alimdl || exit 1;
rm $dir/$y.*.aliacc || exit 1;
[ $y -gt 0 ] && rm $dir/$y.alimdl
fi
cur_alimdl=$dir/$[$y+1].alimdl
y=$[$y+1]
done
(cd $dir; rm final.alimdl 2>/dev/null; ln -s $y.alimdl final.alimdl )
fi
# Print out summary of the warning messages.
for x in $dir/log/*.log; do
n=`grep WARNING $x | wc -l`;
if [ $n -ne 0 ]; then echo $n warnings in $x; fi;
done
echo Done
egs/wsj/s5/steps/train_ubm.sh 0 → 100755
+ 125
− 0
View file @ 356538a1
#!/bin/bash
# Copyright 2012 Daniel Povey. Apache 2.0.
# This trains a UBM (i.e. a mixture of Gaussians), by clustering
# the Gaussians from a trained HMM/GMM system and then doing a few
# iterations of UBM training.
# We mostly use this for SGMM systems.
# Begin configuration section.
nj=4
cmd=run.pl
silence_weight= # You can set it to e.g. 0.0, to weight down silence in training.
stage=-2
num_gselect1=50 # first stage of Gaussian-selection
num_gselect2=25 # second stage.
intermediate_num_gauss=2000
num_iters=3
# End configuration section.
if [ -f path.sh ]; then . ./path.sh; fi
. parse_options.sh || exit 1;
if [ $# != 5 ]; then
echo "Usage: steps/train_ubm.sh <num-gauss> <data> <lang> <ali-dir> <exp>"
echo " e.g.: steps/train_ubm.sh 400 data/train_si84 data/lang exp/tri2b_ali_si84 exp/ubm3c"
echo "main options (for others, see top of script file)"
echo " --config <config-file> # config containing options"
echo " --cmd (utils/run.pl|utils/queue.pl <queue opts>) # how to run jobs."
echo " --silence-weight <sil-weight> # weight for silence (e.g. 0.5 or 0.0)"
echo " --num-iters <#iters> # Number of iterations of E-M"
exit 1;
fi
num_gauss=$1
data=$2
lang=$3
alidir=$4
dir=$5
for f in $data/feats.scp $lang/L.fst $alidir/ali.1.gz $alidir/final.mdl; do
[ ! -f $f ] && echo "No such file $f" && exit 1;
done
if [ $[$num_gauss*2] -gt $intermediate_num_gauss ]; then
echo "intermediate_num_gauss was too small $intermediate_num_gauss"
intermediate_num_gauss=$[$num_gauss*2];
echo "setting it to $intermediate_num_gauss"
fi
# Set various variables.
silphonelist=`cat $lang/phones/silence.csl` || exit 1;
nj=`cat $alidir/num_jobs` || exit 1;
mkdir -p $dir/log
echo $nj > $dir/num_jobs
sdata=$data/split$nj;
[[ -d $sdata && $data/feats.scp -ot $sdata ]] || split_data.sh $data $nj || exit 1;
## Set up features.
if [ -f $alidir/final.mat ]; then feat_type=lda; else feat_type=delta; fi
echo "$0: feature type is $feat_type"
case $feat_type in
delta) sifeats="ark,s,cs:apply-cmvn --norm-vars=false --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | add-deltas ark:- ark:- |";;
lda) sifeats="ark,s,cs:apply-cmvn --norm-vars=false --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | splice-feats ark:- ark:- | transform-feats $alidir/final.mat ark:- ark:- |"
cp $alidir/final.mat $dir
;;
*) echo "$0: invalid feature type $feat_type" && exit 1;
esac
if [ -f $alidir/1.trans ]; then
echo "$0: using transforms from $alidir"
feats="$sifeats transform-feats --utt2spk=ark:$sdata/JOB/utt2spk ark,s,cs:$alidir/JOB.trans ark:- ark:- |"
fi
##
if [ ! -z "$silence_weight" ]; then
weights_opt="--weights='gunzip -c $alidir/ali.JOB.gz | ali-to-post ark:- ark:- | weight-silence-post $silence_weight $silphonelist $alidir/final.mdl ark:- ark:- | post-to-weights ark:- ark:- |'"
else
weights_opt=
fi
if [ $stage -le -2 ]; then
echo "$0: clustering model $alidir/final.mdl to get initial UBM"
$cmd $dir/log/cluster.log \
init-ubm --intermediate-num-gauss=$intermediate_num_gauss --ubm-num-gauss=$num_gauss \
--verbose=2 --fullcov-ubm=true $alidir/final.mdl $alidir/final.occs \
$dir/0.ubm || exit 1;
fi
# Do initial phase of Gaussian selection and save it to disk -- later on we'll
# do more Gaussian selection to further prune, as the model changes.
if [ $stage -le -1 ]; then
echo "$0: doing Gaussian selection"
$cmd JOB=1:$nj $dir/log/gselect.JOB.log \
gmm-gselect --n=$num_gselect1 "fgmm-global-to-gmm $dir/0.ubm - |" "$feats" \
"ark:|gzip -c >$dir/gselect.JOB.gz" || exit 1;
fi
x=0
while [ $x -lt $num_iters ]; do
echo "Pass $x"
$cmd JOB=1:$nj $dir/log/acc.$x.JOB.log \
gmm-gselect --n=$num_gselect2 "--gselect=ark,s,cs:gunzip -c $dir/gselect.JOB.gz|" \
"fgmm-global-to-gmm $dir/$x.ubm - |" "$feats" ark:- \| \
fgmm-global-acc-stats $weights_opt --gselect=ark,s,cs:- $dir/$x.ubm "$feats" \
$dir/$x.JOB.acc || exit 1;
lowcount_opt="--remove-low-count-gaussians=false"
[ $[$x+1] -eq $num_iters ] && lowcount_opt= # Only remove low-count Gaussians
# on last iter-- we can't do it earlier, or the Gaussian-selection info would
# be mismatched.
$cmd $dir/log/update.$x.log \
fgmm-global-est $lowcount_opt --verbose=2 $dir/$x.ubm "fgmm-global-sum-accs - $dir/$x.*.acc |" \
$dir/$[$x+1].ubm || exit 1;
rm $dir/$x.*.acc $dir/$x.ubm
x=$[$x+1]
done
rm $dir/gselect.*.gz
rm $dir/final.ubm 2>/dev/null
mv $dir/$x.ubm $dir/final.ubm || exit 1;
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment