icefall/egs/mls/ASR/prepare.sh
2024-02-28 12:10:37 +08:00

228 lines
6.2 KiB
Bash
Executable File

#!/usr/bin/env bash
# fix segmentation fault reported in https://github.com/k2-fsa/icefall/issues/674
export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python
set -eou pipefail
nj=15
# run step 0 to step 5 by default
stage=0
stop_stage=5
# Note: This script just prepare the minimal requirements that needed by a
# transducer training with bpe units.
#
# If you want to use ngram or nnlm, please continue running prepare_lm.sh after
# you succeed running this script.
#
# This script also contains the steps to generate phone based units, but they
# will not run automatically, you can generate the phone based units by
# bash prepare.sh --stage -1 --stop-stage -1
# bash prepare.sh --stage 6 --stop-stage 6
# We assume dl_dir (download dir) contains the following
# directories and files. If not, they will be downloaded
# by this script automatically.
#
# - $dl_dir/LibriSpeech
# You can find BOOKS.TXT, test-clean, train-clean-360, etc, inside it.
# You can download them from https://www.openslr.org/12
#
# - $dl_dir/musan
# This directory contains the following directories downloaded from
# http://www.openslr.org/17/
#
# - music
# - noise
# - speech
num_per_split=4000
fbank_dir=data/fbank_mls
dl_dir=$PWD/download
. 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=(
# 5000
2000
1000
# 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 "Running prepare.sh"
log "dl_dir: $dl_dir"
log "fbank_dir: $fbank_dir"
languages=(
english
german
dutch
spanish
italian
french
polish
portuguese
)
if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
log "Stage 0: Download data"
# If you have pre-downloaded it to /path/to/MLS,
# you can create a symlink
#
# ln -sfv /path/to/MLS $dl_dir/MLS
#
if [ ! -d $dl_dir/MLS/train-other-500 ]; then
lhotse download mls --full $dl_dir
fi
# 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 MLS manifest"
# We assume that you have downloaded the MLS corpus
# to $dl_dir/MLS
mkdir -p data/manifests
if [ ! -e data/manifests/.mls.done ]; then
lhotse prepare mls -j $nj $dl_dir/MLS data/manifests
touch data/manifests/.mls.done
fi
fi
if [ $stage -le 2 ] && [ $stop_stage -ge 2 ]; then
log "Stage 2: Prepare musan manifest"
# We assume that you have downloaded the musan corpus
# to $dl_dir/musan
mkdir -p data/manifests
if [ ! -e data/manifests/.musan.done ]; then
lhotse prepare musan $dl_dir/musan data/manifests
touch data/manifests/.musan.done
fi
fi
if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
log "Stage 3: Split english subset into pieces (may take 30 minutes)"
split_dir=${fbank_dir}/english_split
if [ ! -f $split_dir/.split_completed ]; then
lhotse split-lazy ${fbank_dir}/mls-english_train_raw.jsonl.gz $split_dir $num_per_split
touch $split_dir/.split_completed
fi
fi
if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
log "Stage 4: Compute fbank for MLS (except English)"
mkdir -p ${fbank_dir}
if [ ! -e ${fbank_dir}/.mls.done ]; then
./local/compute_fbank_mls.py
touch ${fbank_dir}/.mls.done
fi
fi
if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
log "Stage 5: Compute fbank for English split of MLS"
if [ ! -e ${fbank_dir}/.mls-english.done ]; then
num_splits=$(find ${fbank_dir}/english_split -name "mls-english_train_raw.*.jsonl.gz" | wc -l)
./local/compute_fbank_mls_splits.py \
--fbank-dir $fbank_dir \
--num-workers 20 \
--language english \
--num-splits $num_splits \
touch ${fbank_dir}/.mls-english.done
fi
if [ ! -e ${fbank_dir}/mls-english_train.jsonl.gz ]; then
pieces=$(find ${fbank_dir}/english_split -name "mls-english_train.*.jsonl.gz")
lhotse combine $pieces ${fbank_dir}/mls-english_train.jsonl.gz
fi
fi
if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
log "Stage 6: Validate the manifest of MLS"
if [ ! -e ${fbank_dir}/.mls-validated.done ]; then
log "Validating the fbank features for MLS"
parts=(
train
dev
test
)
for lan in ${languages[@]}; do
for part in ${parts[@]}; do
python3 ./local/validate_manifest.py \
${fbank_dir}/mls-${lan}_${part}.jsonl.gz
done
done
touch ${fbank_dir}/.mls-validated.done
fi
fi
if [ $stage -le 7 ] && [ $stop_stage -ge 7 ]; then
log "Stage 7: Compute fbank for musan"
mkdir -p ${fbank_dir}
if [ ! -e ${fbank_dir}/.musan.done ]; then
./local/compute_fbank_musan.py
touch ${fbank_dir}/.musan.done
fi
fi
if [ $stage -le 8 ] && [ $stop_stage -ge 8 ]; 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
if [ ! -f $lang_dir/transcript_words.txt ]; then
log "Generate data for BPE training"
files=(
"$dl_dir/MLS/mls_english/train/transcripts.txt"
"$dl_dir/MLS/mls_german/train/transcripts.txt"
"$dl_dir/MLS/mls_dutch/train/transcripts.txt"
"$dl_dir/MLS/mls_french/train/transcripts.txt"
"$dl_dir/MLS/mls_spanish/train/transcripts.txt"
"$dl_dir/MLS/mls_italian/train/transcripts.txt"
"$dl_dir/MLS/mls_portuguese/train/transcripts.txt"
"$dl_dir/MLS/mls_polish/train/transcripts.txt"
)
for f in ${files[@]}; do
head -n 1000000 $f | cut -d " " -f 2-
done > $lang_dir/transcript_words.txt
fi
if [ ! -f $lang_dir/bpe.model ]; then
./local/train_bpe_model.py \
--lang-dir $lang_dir \
--vocab-size $vocab_size \
--character-coverage 0.999 \
--transcript $lang_dir/transcript_words.txt \
--byte-fallback
fi
done
fi