mirror of
https://github.com/k2-fsa/icefall.git
synced 2025-08-09 18:12:19 +00:00
301 lines
11 KiB
Bash
Executable File
301 lines
11 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
. ./path.sh
|
|
set -eou pipefail
|
|
|
|
nj=15
|
|
stage=0
|
|
stop_stage=500
|
|
|
|
# We assume dl_dir (download dir) contains the following
|
|
# directories and files. Most of them can't be downloaded automatically
|
|
# as they are not publically available and require a license purchased
|
|
# from the LDC.
|
|
#
|
|
# - $dl_dir/{LDC2004S13,LDC2004T19,LDC2005S13,LDC2005T19}
|
|
# Fisher LDC packages.
|
|
#
|
|
# - $dl_dir/LDC97S62
|
|
# Switchboard LDC audio package (transcripts are auto-downloaded)
|
|
#
|
|
# - $dl_dir/{LDC2002S09,LDC2002T43}
|
|
# Eval2000 audio and transcript
|
|
#
|
|
# - $dl_dir/musan
|
|
# This directory contains the following directories downloaded from
|
|
# http://www.openslr.org/17/
|
|
#
|
|
# - music
|
|
# - noise
|
|
# - speech
|
|
|
|
dl_dir=$PWD/download
|
|
mkdir -p $dl_dir
|
|
|
|
. shared/parse_options.sh || exit 1
|
|
|
|
# vocab size for sentence piece models.
|
|
# It will generate data/lang_bpe_xxx,
|
|
# data/lang_bpe_yyy if the array contains xxx, yyy
|
|
vocab_sizes=(
|
|
500
|
|
)
|
|
|
|
# All files generated by this script are saved in "data".
|
|
# You can safely remove "data" and rerun this script to regenerate it.
|
|
mkdir -p data
|
|
|
|
log() {
|
|
# This function is from espnet
|
|
local fname=${BASH_SOURCE[1]##*/}
|
|
echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
|
|
}
|
|
|
|
log "dl_dir: $dl_dir"
|
|
|
|
if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
|
|
log "Stage 0: Download data"
|
|
|
|
# If you have pre-downloaded it to /path/to/fisher and /path/to/swbd,
|
|
# you can create a symlink
|
|
#
|
|
# ln -sfv /path/to/fisher $dl_dir/fisher
|
|
#
|
|
|
|
# TODO: remove
|
|
LDC_ROOT=/nas/data4/DATA
|
|
for pkg in LDC2004S13 LDC2004T19 LDC2005S13 LDC2005T19 LDC97S62 LDC2002S09 LDC2002T43; do
|
|
ln -sfv $LDC_ROOT/$pkg $dl_dir/
|
|
done
|
|
|
|
# If you have pre-downloaded it to /path/to/musan,
|
|
# you can create a symlink
|
|
#
|
|
# ln -sfv /path/to/musan $dl_dir/
|
|
#
|
|
if [ ! -d $dl_dir/musan ]; then
|
|
lhotse download musan $dl_dir
|
|
fi
|
|
fi
|
|
|
|
if [ $stage -le 1 ] && [ $stop_stage -ge 1 ] ; then
|
|
log "Stage 1: Prepare Fisher manifests"
|
|
mkdir -p data/manifests/fisher
|
|
lhotse prepare fisher-english --absolute-paths 1 $dl_dir data/manifests/fisher
|
|
local/normalize_and_filter_supervisions.py data/manifests/fisher/supervisions.jsonl.gz data/manifests/supervisions_fisher.jsonl.gz
|
|
cp data/manifests/fisher/recordings.jsonl.gz data/manifests/recordings_fisher.jsonl.gz
|
|
gzip -d data/manifests/supervisions_fisher.jsonl.gz
|
|
gzip -d data/manifests/recordings_fisher.jsonl.gz
|
|
fi
|
|
|
|
if [ $stage -le 2 ] && [ $stop_stage -ge 2 ]; then
|
|
log "Stage 2: Prepare SWBD manifests"
|
|
mkdir -p data/manifests/swbd
|
|
lhotse prepare switchboard --absolute-paths 1 --omit-silence $dl_dir/LDC97S62 data/manifests/swbd
|
|
python3 local/normalize_and_filter_supervisions.py data/manifests/swbd/swbd_supervisions_all.jsonl.gz data/manifests/supervisions_swbd.jsonl.gz
|
|
cp data/manifests/swbd/swbd_recordings_all.jsonl.gz data/manifests/recordings_swbd.jsonl.gz
|
|
gzip -d data/manifests/supervisions_swbd.jsonl.gz
|
|
gzip -d data/manifests/recordings_swbd.jsonl.gz
|
|
fi
|
|
|
|
if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
|
|
mkdir -p data/manifests/eval2000
|
|
lhotse prepare eval2000 --absolute-paths 1 $dl_dir data/manifests/eval2000
|
|
python3 local/normalize_eval2000.py data/manifests/eval2000/eval2000_supervisions_unnorm.jsonl.gz data/manifests/eval2000/supervisions_eval2000.jsonl.gz
|
|
lhotse fix data/manifests/eval2000/eval2000_recordings_all.jsonl.gz data/manifests/eval2000/supervisions_eval2000.jsonl.gz data/manifests
|
|
mv data/manifests/eval2000_recordings_all.jsonl.gz data/manifests/recordings_eval2000.jsonl.gz
|
|
gzip -d data/manifests/recordings_eval2000.jsonl.gz
|
|
gzip -d data/manifests/supervisions_eval2000.jsonl.gz
|
|
fi
|
|
|
|
if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
|
|
mkdir -p data/fbank
|
|
python3 local/compute_fbank_fisher_swbd_eval2000.py
|
|
fi
|
|
|
|
if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
|
|
#####################################
|
|
#fisher
|
|
#####################################
|
|
|
|
gzip -d data/fbank/cuts_fisher.json.gz
|
|
jq -c '.[]' data/fbank/cuts_fisher.json > data/fbank/cuts_fisher.jsonl
|
|
gzip -c data/fbank/cuts_fisher.jsonl > data/fbank/cuts_fisher.jsonl.gz
|
|
|
|
# extract list of sph
|
|
python3 local/extract_list_of_sph.py data/fbank/cuts_fisher.jsonl | sort | uniq > data/fbank/cuts_fisher_sph.list
|
|
|
|
num_fisher_total_session=$(wc -l <data/fbank/cuts_fisher_sph.list)
|
|
num_fisher_dev_session=10
|
|
num_fisher_train_session=$(($num_fisher_total_session - $num_fisher_dev_session))
|
|
head -n $num_fisher_dev_session data/fbank/cuts_fisher_sph.list >data/fbank/cuts_fisher_sph_dev.list
|
|
tail -n $num_fisher_train_session data/fbank/cuts_fisher_sph.list >data/fbank/cuts_fisher_sph_train.list
|
|
|
|
# extarct dev json
|
|
python3 local/extract_json_cuts.py data/fbank/cuts_fisher_sph_dev.list data/fbank/cuts_fisher.jsonl data/fbank/dev_cuts_fisher.jsonl
|
|
gzip -c data/fbank/dev_cuts_fisher.jsonl > data/fbank/dev_cuts_fisher.jsonl.gz
|
|
|
|
# extract train json
|
|
python3 local/extract_json_cuts.py data/fbank/cuts_fisher_sph_train.list data/fbank/cuts_fisher.jsonl data/fbank/train_cuts_fisher.jsonl
|
|
gzip -c data/fbank/train_cuts_fisher.jsonl > data/fbank/train_cuts_fisher.jsonl.gz
|
|
|
|
# describe cut
|
|
lhotse cut describe data/fbank/train_cuts_fisher.jsonl.gz
|
|
lhotse cut describe data/fbank/dev_cuts_fisher.jsonl.gz
|
|
|
|
# extract dev supervision
|
|
python local/extract_json_supervision.py data/fbank/cuts_fisher_sph_dev.list data/manifests/supervisions_fisher.jsonl data/manifests/dev_supervisions_fisher.jsonl
|
|
python local/extract_json_supervision.py data/fbank/cuts_fisher_sph_train.list data/manifests/supervisions_fisher.jsonl data/manifests/train_supervisions_fisher.jsonl
|
|
|
|
######################################
|
|
#swbd
|
|
######################################
|
|
|
|
gzip -d data/fbank/cuts_swbd.json.gz
|
|
jq -c '.[]' data/fbank/cuts_swbd.json > data/fbank/cuts_swbd.jsonl
|
|
gzip -c data/fbank/cuts_swbd.jsonl > data/fbank/cuts_swbd.jsonl.gz
|
|
|
|
python3 local/extract_list_of_sph.py data/fbank/cuts_swbd.jsonl| sort | uniq > data/fbank/cuts_swbd_sph.list
|
|
num_swbd_total_session=$(wc -l <data/fbank/cuts_swbd_sph.list)
|
|
num_swbd_dev_session=10
|
|
num_swbd_train_session=$(($num_swbd_total_session - $num_swbd_dev_session))
|
|
|
|
head -n $num_swbd_dev_session data/fbank/cuts_swbd_sph.list >data/fbank/cuts_swbd_sph_dev.list
|
|
tail -n $num_swbd_train_session data/fbank/cuts_swbd_sph.list >data/fbank/cuts_swbd_sph_train.list
|
|
|
|
# extarct dev json
|
|
python3 local/extract_json_cuts.py data/fbank/cuts_swbd_sph_dev.list data/fbank/cuts_swbd.jsonl data/fbank/dev_cuts_swbd.jsonl
|
|
gzip -c data/fbank/dev_cuts_swbd.jsonl > data/fbank/dev_cuts_swbd.jsonl.gz
|
|
|
|
python3 local/extract_json_cuts.py data/fbank/cuts_swbd_sph_train.list data/fbank/cuts_swbd.jsonl data/fbank/train_cuts_swbd.jsonl
|
|
gzip -c data/fbank/train_cuts_swbd.jsonl > data/fbank/train_cuts_swbd.jsonl.gz
|
|
|
|
# describe cut
|
|
lhotse cut describe data/fbank/train_cuts_swbd.jsonl.gz
|
|
lhotse cut describe data/fbank/dev_cuts_swbd.jsonl.gz
|
|
|
|
# extract dev supervision
|
|
python local/extract_json_supervision.py data/fbank/cuts_swbd_sph_dev.list data/manifests/supervisions_swbd.jsonl data/manifests/dev_supervisions_swbd.jsonl
|
|
python local/extract_json_supervision.py data/fbank/cuts_swbd_sph_train.list data/manifests/supervisions_swbd.jsonl data/manifests/train_supervisions_swbd.jsonl
|
|
fi
|
|
|
|
if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
|
|
log "Stage 3: Prepare musan manifest"
|
|
# We assume that you have downloaded the musan corpus
|
|
# to data/musan
|
|
mkdir -p data/manifests/musan
|
|
lhotse prepare musan $dl_dir/musan data/manifests/musan
|
|
fi
|
|
|
|
if [ $stage -le 7 ] && [ $stop_stage -ge 7 ]; then
|
|
python3 local/compute_fbank_musan.py
|
|
fi
|
|
|
|
|
|
if [ $stage -le 8 ] && [ $stop_stage -ge 8 ]; then
|
|
log "Stage 6: Dump transcripts for LM training"
|
|
mkdir -p data/lm
|
|
cat data/manifests/supervisions_fisher.jsonl data/manifests/supervisions_swbd.jsonl \
|
|
| jq '.text' \
|
|
| sed 's:"::g' \
|
|
> data/lm/transcript_words.txt
|
|
|
|
cat data/manifests/train_supervisions_fisher.jsonl data/manifests/train_supervisions_swbd.jsonl \
|
|
| jq '.text' \
|
|
| sed 's:"::g' \
|
|
> data/lm/train_transcript_words.txt
|
|
|
|
cat data/manifests/dev_supervisions_fisher.jsonl data/manifests/dev_supervisions_swbd.jsonl \
|
|
| jq '.text' \
|
|
| sed 's:"::g' \
|
|
> data/lm/dev_transcript_words.txt
|
|
fi
|
|
|
|
if [ $stage -le 9 ] && [ $stop_stage -ge 9 ]; then
|
|
log "Stage 7: Prepare lexicon using g2p_en"
|
|
lang_dir=data/lang_phone
|
|
mkdir -p $lang_dir
|
|
|
|
# Add special words to words.txt
|
|
echo "<eps> 0" > $lang_dir/words.txt
|
|
echo "!SIL 1" >> $lang_dir/words.txt
|
|
echo "[UNK] 2" >> $lang_dir/words.txt
|
|
|
|
# Add regular words to words.txt
|
|
#gunzip -c data/manifests/fisher-swbd_supervisions_norm.jsonl.gz \
|
|
cat data/manifests/supervisions_fisher.jsonl data/manifests/supervisions_swbd.jsonl \
|
|
| jq '.text' \
|
|
| sed 's:"::g' \
|
|
| sed 's: :\n:g' \
|
|
| sort \
|
|
| uniq \
|
|
| awk '{print $0,NR+2}' \
|
|
>> $lang_dir/words.txt
|
|
|
|
# Add remaining special word symbols expected by LM scripts.
|
|
num_words=$(cat $lang_dir/words.txt | wc -l)
|
|
echo "<s> ${num_words}" >> $lang_dir/words.txt
|
|
num_words=$(cat $lang_dir/words.txt | wc -l)
|
|
echo "</s> ${num_words}" >> $lang_dir/words.txt
|
|
num_words=$(cat $lang_dir/words.txt | wc -l)
|
|
echo "#0 ${num_words}" >> $lang_dir/words.txt
|
|
|
|
if [ ! -f $lang_dir/L_disambig.pt ]; then
|
|
# We discard SWBD's lexicon and just use g2p_en
|
|
# It was trained on CMUdict and looks it up before
|
|
# resorting to an LSTM G2P model.
|
|
pip install g2p_en
|
|
./local/prepare_lang_g2pen.py --lang-dir $lang_dir
|
|
fi
|
|
fi
|
|
|
|
if [ $stage -le 10 ] && [ $stop_stage -ge 10 ]; then
|
|
log "Stage 8: Prepare BPE based lang"
|
|
|
|
for vocab_size in ${vocab_sizes[@]}; do
|
|
lang_dir=data/lang_bpe_${vocab_size}
|
|
mkdir -p $lang_dir
|
|
# We reuse words.txt from phone based lexicon
|
|
# so that the two can share G.pt later.
|
|
cp data/lang_phone/words.txt $lang_dir
|
|
|
|
./local/train_bpe_model.py \
|
|
--lang-dir $lang_dir \
|
|
--vocab-size $vocab_size \
|
|
--transcript data/lm/transcript_words.txt
|
|
|
|
if [ ! -f $lang_dir/L_disambig.pt ]; then
|
|
./local/prepare_lang_bpe.py --lang-dir $lang_dir
|
|
fi
|
|
done
|
|
fi
|
|
|
|
if [ $stage -le 11 ] && [ $stop_stage -ge 11 ]; then
|
|
log "Stage 9: Train LM"
|
|
lm_dir=data/lm
|
|
|
|
if [ ! -f $lm_dir/G.arpa ]; then
|
|
./shared/make_kn_lm.py \
|
|
-ngram-order 3 \
|
|
-text $lm_dir/transcript_words.txt \
|
|
-lm $lm_dir/G.arpa
|
|
fi
|
|
|
|
if [ ! -f $lm_dir/G_3_gram.fst.txt ]; then
|
|
python3 -m kaldilm \
|
|
--read-symbol-table="data/lang_phone/words.txt" \
|
|
--disambig-symbol='#0' \
|
|
--max-order=3 \
|
|
$lm_dir/G.arpa > $lm_dir/G_3_gram.fst.txt
|
|
fi
|
|
fi
|
|
|
|
if [ $stage -le 12 ] && [ $stop_stage -ge 12 ]; then
|
|
log "Stage 10: Compile HLG"
|
|
./local/compile_hlg.py --lang-dir data/lang_phone
|
|
|
|
for vocab_size in ${vocab_sizes[@]}; do
|
|
lang_dir=data/lang_bpe_${vocab_size}
|
|
./local/compile_hlg.py --lang-dir $lang_dir
|
|
done
|
|
fi
|