mirror of
https://github.com/k2-fsa/icefall.git
synced 2025-08-08 09:32:20 +00:00
c3f6f28116
* init commit * Create README.md * handle code switching cases * misc. fixes * added manifest statistics * init commit for the zipformer recipe * added scripts for exporting model * added RESULTS.md * added scripts for streaming related stuff * doc str fixed
309 lines
8.8 KiB
Bash
Executable File
309 lines
8.8 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
|
|
|
|
stage=-1
|
|
stop_stage=100
|
|
perturb_speed=true
|
|
|
|
|
|
# We assume dl_dir (download dir) contains the following
|
|
# directories and files. If not, they will be downloaded
|
|
# by this script automatically.
|
|
#
|
|
# - $dl_dir/mdcc
|
|
# |-- README.md
|
|
# |-- audio/
|
|
# |-- clip_info_rthk.csv
|
|
# |-- cnt_asr_metadata_full.csv
|
|
# |-- cnt_asr_test_metadata.csv
|
|
# |-- cnt_asr_train_metadata.csv
|
|
# |-- cnt_asr_valid_metadata.csv
|
|
# |-- data_statistic.py
|
|
# |-- length
|
|
# |-- podcast_447_2021.csv
|
|
# |-- test.txt
|
|
# |-- transcription/
|
|
# `-- words_length
|
|
# You can download them from:
|
|
# https://drive.google.com/file/d/1epfYMMhXdBKA6nxPgUugb2Uj4DllSxkn/view?usp=drive_link
|
|
#
|
|
# - $dl_dir/musan
|
|
# This directory contains the following directories downloaded from
|
|
# http://www.openslr.org/17/
|
|
#
|
|
# - music
|
|
# - noise
|
|
# - speech
|
|
|
|
dl_dir=$PWD/download
|
|
|
|
. shared/parse_options.sh || exit 1
|
|
|
|
# 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/mdcc,
|
|
# you can create a symlink
|
|
#
|
|
# ln -sfv /path/to/mdcc $dl_dir/mdcc
|
|
#
|
|
# The directory structure is
|
|
# mdcc/
|
|
# |-- README.md
|
|
# |-- audio/
|
|
# |-- clip_info_rthk.csv
|
|
# |-- cnt_asr_metadata_full.csv
|
|
# |-- cnt_asr_test_metadata.csv
|
|
# |-- cnt_asr_train_metadata.csv
|
|
# |-- cnt_asr_valid_metadata.csv
|
|
# |-- data_statistic.py
|
|
# |-- length
|
|
# |-- podcast_447_2021.csv
|
|
# |-- test.txt
|
|
# |-- transcription/
|
|
# `-- words_length
|
|
|
|
if [ ! -d $dl_dir/mdcc/audio ]; then
|
|
lhotse download mdcc $dl_dir
|
|
|
|
# this will download and unzip dataset.zip to $dl_dir/
|
|
|
|
mv $dl_dir/dataset $dl_dir/mdcc
|
|
fi
|
|
|
|
# If you have pre-downloaded it to /path/to/musan,
|
|
# you can create a symlink
|
|
#
|
|
# ln -sfv /path/to/musan $dl_dir/musan
|
|
#
|
|
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 MDCC manifest"
|
|
# We assume that you have downloaded the MDCC corpus
|
|
# to $dl_dir/mdcc
|
|
if [ ! -f data/manifests/.mdcc_manifests.done ]; then
|
|
log "Might take 40 minutes to traverse the directory."
|
|
mkdir -p data/manifests
|
|
lhotse prepare mdcc $dl_dir/mdcc data/manifests
|
|
touch data/manifests/.mdcc_manifests.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 data/musan
|
|
if [ ! -f data/manifests/.musan_manifests.done ]; then
|
|
log "It may take 6 minutes"
|
|
mkdir -p data/manifests
|
|
lhotse prepare musan $dl_dir/musan data/manifests
|
|
touch data/manifests/.musan_manifests.done
|
|
fi
|
|
fi
|
|
|
|
if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
|
|
log "Stage 3: Compute fbank for MDCC"
|
|
if [ ! -f data/fbank/.mdcc.done ]; then
|
|
mkdir -p data/fbank
|
|
./local/compute_fbank_mdcc.py --perturb-speed ${perturb_speed}
|
|
touch data/fbank/.mdcc.done
|
|
fi
|
|
fi
|
|
|
|
if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
|
|
log "Stage 4: Compute fbank for musan"
|
|
if [ ! -f data/fbank/.msuan.done ]; then
|
|
mkdir -p data/fbank
|
|
./local/compute_fbank_musan.py
|
|
touch data/fbank/.msuan.done
|
|
fi
|
|
fi
|
|
|
|
lang_char_dir=data/lang_char
|
|
if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
|
|
log "Stage 5: Prepare char based lang"
|
|
mkdir -p $lang_char_dir
|
|
|
|
# Prepare text.
|
|
# Note: in Linux, you can install jq with the following command:
|
|
# 1. wget -O jq https://github.com/stedolan/jq/releases/download/jq-1.6/jq-linux64
|
|
# 2. chmod +x ./jq
|
|
# 3. cp jq /usr/bin
|
|
if [ ! -f $lang_char_dir/text ]; then
|
|
gunzip -c data/manifests/mdcc_supervisions_train.jsonl.gz \
|
|
|jq '.text' | sed 's/"//g' | ./local/text2token.py -t "char" \
|
|
> $lang_char_dir/train_text
|
|
|
|
cat $lang_char_dir/train_text > $lang_char_dir/text
|
|
|
|
gunzip -c data/manifests/mdcc_supervisions_test.jsonl.gz \
|
|
|jq '.text' | sed 's/"//g' | ./local/text2token.py -t "char" \
|
|
> $lang_char_dir/valid_text
|
|
|
|
cat $lang_char_dir/valid_text >> $lang_char_dir/text
|
|
|
|
gunzip -c data/manifests/mdcc_supervisions_valid.jsonl.gz \
|
|
|jq '.text' | sed 's/"//g' | ./local/text2token.py -t "char" \
|
|
> $lang_char_dir/test_text
|
|
|
|
cat $lang_char_dir/test_text >> $lang_char_dir/text
|
|
fi
|
|
|
|
if [ ! -f $lang_char_dir/text_words_segmentation ]; then
|
|
./local/preprocess_mdcc.py --input-file $lang_char_dir/text \
|
|
--output-dir $lang_char_dir
|
|
|
|
mv $lang_char_dir/text $lang_char_dir/_text
|
|
cp $lang_char_dir/text_words_segmentation $lang_char_dir/text
|
|
fi
|
|
|
|
if [ ! -f $lang_char_dir/tokens.txt ]; then
|
|
./local/prepare_char.py --lang-dir $lang_char_dir
|
|
fi
|
|
fi
|
|
|
|
if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
|
|
log "Stage 6: Prepare G"
|
|
|
|
mkdir -p data/lm
|
|
|
|
# Train LM on transcripts
|
|
if [ ! -f data/lm/3-gram.unpruned.arpa ]; then
|
|
python3 ./shared/make_kn_lm.py \
|
|
-ngram-order 3 \
|
|
-text $lang_char_dir/text_words_segmentation \
|
|
-lm data/lm/3-gram.unpruned.arpa
|
|
fi
|
|
|
|
# We assume you have installed kaldilm, if not, please install
|
|
# it using: pip install kaldilm
|
|
if [ ! -f data/lm/G_3_gram_char.fst.txt ]; then
|
|
# It is used in building HLG
|
|
python3 -m kaldilm \
|
|
--read-symbol-table="$lang_char_dir/words.txt" \
|
|
--disambig-symbol='#0' \
|
|
--max-order=3 \
|
|
data/lm/3-gram.unpruned.arpa > data/lm/G_3_gram_char.fst.txt
|
|
fi
|
|
|
|
if [ ! -f $lang_char_dir/HLG.fst ]; then
|
|
./local/prepare_lang_fst.py \
|
|
--lang-dir $lang_char_dir \
|
|
--ngram-G ./data/lm/G_3_gram_char.fst.txt
|
|
fi
|
|
fi
|
|
|
|
if [ $stage -le 7 ] && [ $stop_stage -ge 7 ]; then
|
|
log "Stage 7: Compile LG & HLG"
|
|
|
|
./local/compile_hlg.py --lang-dir $lang_char_dir --lm G_3_gram_char
|
|
./local/compile_lg.py --lang-dir $lang_char_dir --lm G_3_gram_char
|
|
fi
|
|
|
|
if [ $stage -le 8 ] && [ $stop_stage -ge 8 ]; then
|
|
log "Stage 8: Generate LM training data"
|
|
|
|
log "Processing char based data"
|
|
out_dir=data/lm_training_char
|
|
mkdir -p $out_dir $dl_dir/lm
|
|
|
|
if [ ! -f $dl_dir/lm/mdcc-train-word.txt ]; then
|
|
./local/text2segments.py --input-file $lang_char_dir/train_text \
|
|
--output-file $dl_dir/lm/mdcc-train-word.txt
|
|
fi
|
|
|
|
# training words
|
|
./local/prepare_char_lm_training_data.py \
|
|
--lang-char data/lang_char \
|
|
--lm-data $dl_dir/lm/mdcc-train-word.txt \
|
|
--lm-archive $out_dir/lm_data.pt
|
|
|
|
# valid words
|
|
if [ ! -f $dl_dir/lm/mdcc-valid-word.txt ]; then
|
|
./local/text2segments.py --input-file $lang_char_dir/valid_text \
|
|
--output-file $dl_dir/lm/mdcc-valid-word.txt
|
|
fi
|
|
|
|
./local/prepare_char_lm_training_data.py \
|
|
--lang-char data/lang_char \
|
|
--lm-data $dl_dir/lm/mdcc-valid-word.txt \
|
|
--lm-archive $out_dir/lm_data_valid.pt
|
|
|
|
# test words
|
|
if [ ! -f $dl_dir/lm/mdcc-test-word.txt ]; then
|
|
./local/text2segments.py --input-file $lang_char_dir/test_text \
|
|
--output-file $dl_dir/lm/mdcc-test-word.txt
|
|
fi
|
|
|
|
./local/prepare_char_lm_training_data.py \
|
|
--lang-char data/lang_char \
|
|
--lm-data $dl_dir/lm/mdcc-test-word.txt \
|
|
--lm-archive $out_dir/lm_data_test.pt
|
|
fi
|
|
|
|
if [ $stage -le 9 ] && [ $stop_stage -ge 9 ]; then
|
|
log "Stage 9: Sort LM training data"
|
|
# Sort LM training data by sentence length in descending order
|
|
# for ease of training.
|
|
#
|
|
# Sentence length equals to the number of tokens
|
|
# in a sentence.
|
|
|
|
out_dir=data/lm_training_char
|
|
mkdir -p $out_dir
|
|
ln -snf ../../../librispeech/ASR/local/sort_lm_training_data.py local/
|
|
|
|
./local/sort_lm_training_data.py \
|
|
--in-lm-data $out_dir/lm_data.pt \
|
|
--out-lm-data $out_dir/sorted_lm_data.pt \
|
|
--out-statistics $out_dir/statistics.txt
|
|
|
|
./local/sort_lm_training_data.py \
|
|
--in-lm-data $out_dir/lm_data_valid.pt \
|
|
--out-lm-data $out_dir/sorted_lm_data-valid.pt \
|
|
--out-statistics $out_dir/statistics-valid.txt
|
|
|
|
./local/sort_lm_training_data.py \
|
|
--in-lm-data $out_dir/lm_data_test.pt \
|
|
--out-lm-data $out_dir/sorted_lm_data-test.pt \
|
|
--out-statistics $out_dir/statistics-test.txt
|
|
fi
|
|
|
|
if [ $stage -le 12 ] && [ $stop_stage -ge 12 ]; then
|
|
log "Stage 12: Train RNN LM model"
|
|
python ../../../icefall/rnn_lm/train.py \
|
|
--start-epoch 0 \
|
|
--world-size 1 \
|
|
--num-epochs 20 \
|
|
--use-fp16 0 \
|
|
--embedding-dim 512 \
|
|
--hidden-dim 512 \
|
|
--num-layers 2 \
|
|
--batch-size 400 \
|
|
--exp-dir rnnlm_char/exp \
|
|
--lm-data $out_dir/sorted_lm_data.pt \
|
|
--lm-data-valid $out_dir/sorted_lm_data-valid.pt \
|
|
--vocab-size 4336 \
|
|
--master-port 12345
|
|
fi
|