forked from kaldi-asr/kaldi
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Adding the reverberation script to AMI (kaldi-asr#1178)
Added reverberation based data augmentation recipe for AMI. Gives gains in IHM, SDM and MDM settings. (TDNN + Chain recipe checked in).
- Loading branch information
1 parent
49eb733
commit a2f5464
Showing
5 changed files
with
553 additions
and
2 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,283 @@ | ||
#!/bin/bash | ||
|
||
# This is a chain-training script with TDNN neural networks. | ||
# This script is based on local/chain/run_tdnn.sh, but adding | ||
# the reverberated IHM data into the train set. | ||
# This script obtains better results on both IHM and SDM tasks. | ||
|
||
# Please see RESULTS_* for examples of command lines invoking this script. | ||
|
||
# local/chain/multi_condition/run_tdnn.sh --mic ihm --train-set train_cleaned --gmm tri3_cleaned & | ||
# local/chain/multi_condition/run_tdnn.sh --mic sdm1 --use-ihm-ali true --train-set train_cleaned --gmm tri3_cleaned & | ||
# local/chain/multi_condition/run_tdnn.sh --mic mdm8 --use-ihm-ali true --train-set train_cleaned --gmm tri3_cleaned & | ||
|
||
|
||
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=1 | ||
mic=ihm | ||
nj=30 | ||
min_seg_len=1.55 | ||
use_ihm_ali=true | ||
train_set=train_cleaned | ||
gmm=tri3_cleaned # the gmm for the target data | ||
ihm_gmm=tri3_cleaned # the gmm for the IHM system (if --use-ihm-ali true). | ||
num_threads_ubm=32 | ||
num_data_reps=1 | ||
|
||
# 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= #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. | ||
|
||
# End configuration section. | ||
echo "$0 $@" # Print the command line for logging | ||
|
||
. cmd.sh | ||
. ./path.sh | ||
. ./utils/parse_options.sh | ||
|
||
if ! $use_ihm_ali; then | ||
[ "$mic" != "ihm" ] && \ | ||
echo "$0: you cannot specify --use-ihm-ali false if the microphone is not ihm." && \ | ||
exit 1; | ||
else | ||
[ "$mic" == "ihm" ] && \ | ||
echo "$0: you must specify --use-ihm-ali false if the microphone is ihm." && \ | ||
exit 1; | ||
fi | ||
|
||
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 | ||
|
||
nnet3_affix=_cleaned | ||
rvb_affix=_rvb | ||
|
||
|
||
if $use_ihm_ali; then | ||
gmm_dir=exp/ihm/${ihm_gmm} | ||
ali_dir=exp/${mic}/${ihm_gmm}_ali_${train_set}_sp_comb_ihmdata | ||
lores_train_data_dir=data/$mic/${train_set}_ihmdata_sp_comb | ||
tree_dir=exp/$mic/chain${nnet3_affix}/tree_bi${tree_affix}_ihmdata | ||
original_lat_dir=exp/$mic/chain${nnet3_affix}/${ihm_gmm}_${train_set}_sp_comb_lats_ihmdata | ||
lat_dir=exp/$mic/chain${nnet3_affix}${rvb_affix}/${ihm_gmm}_${train_set}_sp${rvb_affix}_comb_lats_ihmdata | ||
dir=exp/$mic/chain${nnet3_affix}${rvb_affix}/tdnn${tdnn_affix}_sp${rvb_affix}_bi_ihmali | ||
# note: the distinction between when we use the 'ihmdata' suffix versus | ||
# 'ihmali' is pretty arbitrary. | ||
else | ||
gmm_dir=exp/${mic}/$gmm | ||
ali_dir=exp/${mic}/${gmm}_ali_${train_set}_sp_comb | ||
lores_train_data_dir=data/$mic/${train_set}_sp_comb | ||
tree_dir=exp/$mic/chain${nnet3_affix}/tree_bi${tree_affix} | ||
original_lat_dir=exp/$mic/chain${nnet3_affix}/${gmm}_${train_set}_sp_comb_lats | ||
lat_dir=exp/$mic/chain${nnet3_affix}${rvb_affix}/${gmm}_${train_set}_sp${rvb_affix}_comb_lats | ||
dir=exp/$mic/chain${nnet3_affix}${rvb_affix}/tdnn${tdnn_affix}_sp${rvb_affix}_bi | ||
fi | ||
|
||
|
||
local/nnet3/multi_condition/run_ivector_common.sh --stage $stage \ | ||
--mic $mic \ | ||
--nj $nj \ | ||
--min-seg-len $min_seg_len \ | ||
--train-set $train_set \ | ||
--gmm $gmm \ | ||
--num-threads-ubm $num_threads_ubm \ | ||
--num-data-reps $num_data_reps \ | ||
--nnet3-affix "$nnet3_affix" | ||
|
||
|
||
# Note: the first stage of the following script is stage 8. | ||
local/nnet3/prepare_lores_feats.sh --stage $stage \ | ||
--mic $mic \ | ||
--nj $nj \ | ||
--min-seg-len $min_seg_len \ | ||
--use-ihm-ali $use_ihm_ali \ | ||
--train-set $train_set | ||
|
||
|
||
train_data_dir=data/$mic/${train_set}_sp${rvb_affix}_hires_comb | ||
train_ivector_dir=exp/$mic/nnet3${nnet3_affix}${rvb_affix}/ivectors_${train_set}_sp${rvb_affix}_hires_comb | ||
final_lm=`cat data/local/lm/final_lm` | ||
LM=$final_lm.pr1-7 | ||
|
||
|
||
for f in $gmm_dir/final.mdl $lores_train_data_dir/feats.scp \ | ||
$train_data_dir/feats.scp $train_ivector_dir/ivector_online.scp; do | ||
[ ! -f $f ] && echo "$0: expected file $f to exist" && exit 1 | ||
done | ||
|
||
|
||
if [ $stage -le 11 ]; then | ||
if [ -f $ali_dir/ali.1.gz ]; then | ||
echo "$0: alignments in $ali_dir appear to already exist. Please either remove them " | ||
echo " ... or use a later --stage option." | ||
exit 1 | ||
fi | ||
echo "$0: aligning perturbed, short-segment-combined ${maybe_ihm}data" | ||
steps/align_fmllr.sh --nj $nj --cmd "$train_cmd" \ | ||
${lores_train_data_dir} data/lang $gmm_dir $ali_dir | ||
fi | ||
|
||
[ ! -f $ali_dir/ali.1.gz ] && echo "$0: expected $ali_dir/ali.1.gz to exist" && exit 1 | ||
|
||
if [ $stage -le 12 ]; 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 [ $stage -le 13 ]; then | ||
# Get the alignments as lattices (gives the chain training more freedom). | ||
# use the same num-jobs as the alignments | ||
steps/align_fmllr_lats.sh --nj 100 --cmd "$train_cmd" ${lores_train_data_dir} \ | ||
data/lang $gmm_dir $original_lat_dir | ||
rm $original_lat_dir/fsts.*.gz # save space | ||
|
||
lat_dir_ihmdata=exp/ihm/chain${nnet3_affix}/${gmm}_${train_set}_sp_comb_lats | ||
|
||
mkdir -p $lat_dir/temp/ | ||
mkdir -p $lat_dir/temp2/ | ||
lattice-copy "ark:gunzip -c $original_lat_dir/lat.*.gz |" ark,scp:$lat_dir/temp/lats.ark,$lat_dir/temp/lats.scp | ||
lattice-copy "ark:gunzip -c $lat_dir_ihmdata/lat.*.gz |" ark,scp:$lat_dir/temp2/lats.ark,$lat_dir/temp2/lats.scp | ||
|
||
# copy the lattices for the reverberated data | ||
rm -f $lat_dir/temp/combined_lats.scp | ||
touch $lat_dir/temp/combined_lats.scp | ||
cat $lat_dir/temp/lats.scp >> $lat_dir/temp/combined_lats.scp | ||
for i in `seq 1 $num_data_reps`; do | ||
cat $lat_dir/temp2/lats.scp | sed -e "s/^/rev${i}_/" >> $lat_dir/temp/combined_lats.scp | ||
done | ||
sort -u $lat_dir/temp/combined_lats.scp > $lat_dir/temp/combined_lats_sorted.scp | ||
|
||
lattice-copy scp:$lat_dir/temp/combined_lats_sorted.scp "ark:|gzip -c >$lat_dir/lat.1.gz" || exit 1; | ||
echo "1" > $lat_dir/num_jobs | ||
|
||
# copy other files from original lattice dir | ||
for f in cmvn_opts final.mdl splice_opts tree; do | ||
cp $original_lat_dir/$f $lat_dir/$f | ||
done | ||
fi | ||
|
||
|
||
if [ $stage -le 14 ]; 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: $tree_dir/final.mdl already exists, refusing to overwrite it." | ||
exit 1; | ||
fi | ||
steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 \ | ||
--context-opts "--context-width=2 --central-position=1" \ | ||
--leftmost-questions-truncate -1 \ | ||
--cmd "$train_cmd" 4200 ${lores_train_data_dir} data/lang_chain $ali_dir $tree_dir | ||
fi | ||
|
||
if [ $stage -le 15 ]; then | ||
mkdir -p $dir | ||
|
||
echo "$0: creating neural net configs"; | ||
|
||
steps/nnet3/tdnn/make_configs.py \ | ||
--self-repair-scale-nonlinearity 0.00001 \ | ||
--feat-dir data/$mic/${train_set}_sp_hires_comb \ | ||
--ivector-dir $train_ivector_dir \ | ||
--tree-dir $tree_dir \ | ||
--relu-dim 450 \ | ||
--splice-indexes "-1,0,1 -1,0,1,2 -3,0,3 -3,0,3 -3,0,3 -6,-3,0 0" \ | ||
--use-presoftmax-prior-scale false \ | ||
--xent-regularize 0.1 \ | ||
--xent-separate-forward-affine true \ | ||
--include-log-softmax false \ | ||
--final-layer-normalize-target 1.0 \ | ||
$dir/configs || exit 1; | ||
fi | ||
|
||
if [ $stage -le 16 ]; 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-rvb$(date +'%m_%d_%H_%M')/s5b/$dir/egs/storage $dir/egs/storage | ||
fi | ||
|
||
touch $dir/egs/.nodelete # keep egs around when that run dies. | ||
|
||
steps/nnet3/chain/train.py --stage $train_stage \ | ||
--cmd "$decode_cmd" \ | ||
--feat.online-ivector-dir $train_ivector_dir \ | ||
--feat.cmvn-opts "--norm-means=false --norm-vars=false" \ | ||
--chain.xent-regularize 0.1 \ | ||
--chain.leaky-hmm-coefficient 0.1 \ | ||
--chain.l2-regularize 0.00005 \ | ||
--chain.apply-deriv-weights false \ | ||
--chain.lm-opts="--num-extra-lm-states=2000" \ | ||
--egs.dir "$common_egs_dir" \ | ||
--egs.opts "--frames-overlap-per-eg 0" \ | ||
--egs.chunk-width 150 \ | ||
--trainer.num-chunk-per-minibatch 128 \ | ||
--trainer.frames-per-iter 1500000 \ | ||
--trainer.num-epochs 4 \ | ||
--trainer.optimization.num-jobs-initial 2 \ | ||
--trainer.optimization.num-jobs-final 12 \ | ||
--trainer.optimization.initial-effective-lrate 0.001 \ | ||
--trainer.optimization.final-effective-lrate 0.0001 \ | ||
--trainer.max-param-change 2.0 \ | ||
--cleanup.remove-egs true \ | ||
--feat-dir $train_data_dir \ | ||
--tree-dir $tree_dir \ | ||
--lat-dir $lat_dir \ | ||
--dir $dir | ||
fi | ||
|
||
|
||
graph_dir=$dir/graph_${LM} | ||
if [ $stage -le 17 ]; then | ||
# Note: it might appear that this data/lang_chain directory is mismatched, and it is as | ||
# far as the 'topo' is concerned, but this script doesn't read the 'topo' from | ||
# the lang directory. | ||
utils/mkgraph.sh --left-biphone --self-loop-scale 1.0 data/lang_${LM} $dir $graph_dir | ||
fi | ||
|
||
if [ $stage -le 18 ]; then | ||
rm $dir/.error 2>/dev/null || true | ||
for decode_set in dev eval; do | ||
( | ||
steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \ | ||
--nj $nj --cmd "$decode_cmd" \ | ||
--online-ivector-dir exp/$mic/nnet3${nnet3_affix}${rvb_affix}/ivectors_${decode_set}_hires \ | ||
--scoring-opts "--min-lmwt 5 " \ | ||
$graph_dir data/$mic/${decode_set}_hires $dir/decode_${decode_set} || 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 |
Oops, something went wrong.