mirror of
https://github.com/k2-fsa/icefall.git
synced 2025-09-03 14:14:19 +00:00
add md files and prepare.sh
This commit is contained in:
luomingshuang
parent
23dc6a0a84
commit
94810ced45
18
egs/tedlium3/ASR/README.md
Normal file
18
egs/tedlium3/ASR/README.md
Normal file
@ -0,0 +1,18 @@
|
||||
|
||||
# Introduction
|
||||
|
||||
This recipe includes some different ASR models trained with TedLium3.
|
||||
|
||||
# Transducers
|
||||
|
||||
There are various folders containing the name `transducer` in this folder.
|
||||
The following table lists the differences among them.
|
||||
|
||||
| | Encoder | Decoder |
|
||||
|------------------------|-----------|--------------------|
|
||||
| `transducer_stateless` | Conformer | Embedding + Conv1d |
|
||||
|
||||
|
||||
The decoder in `transducer_stateless` is modified from the paper
|
||||
[Rnn-Transducer with Stateless Prediction Network](https://ieeexplore.ieee.org/document/9054419/).
|
||||
We place an additional Conv1d layer right after the input embedding layer.
|
||||
68
egs/tedlium3/ASR/RESULTS.md
Normal file
68
egs/tedlium3/ASR/RESULTS.md
Normal file
@ -0,0 +1,68 @@
|
||||
## Results
|
||||
|
||||
### TedLium3 BPE training results (Transducer)
|
||||
|
||||
#### Conformer encoder + embedding decoder
|
||||
|
||||
Using the codes from this PR commit https://github.com/k2-fsa/icefall/pull/183/commits/536ad2252e2d406f24a681743d98bd5f90801b97.
|
||||
|
||||
Conformer encoder + non-current decoder. The decoder
|
||||
contains only an embedding layer and a Conv1d (with kernel size 2).
|
||||
|
||||
The WERs are
|
||||
|
||||
| | dev | test | comment |
|
||||
|------------------------------------|------------|------------|------------------------------------------|
|
||||
| greedy search | 7.19 | 6.57 | --epoch 29, --avg 16, --max-duration 100 |
|
||||
| beam search (beam size 4) | 7.12 | 6.37 | --epoch 29, --avg 16, --max-duration 100 |
|
||||
| modified beam search (beam size 4) | 7.00 | 6.19 | --epoch 29, --avg 16, --max-duration 100 |
|
||||
|
||||
The training command for reproducing is given below:
|
||||
|
||||
```
|
||||
export CUDA_VISIBLE_DEVICES="0,1,2,3"
|
||||
|
||||
./transducer_stateless/train.py \
|
||||
--world-size 4 \
|
||||
--num-epochs 30 \
|
||||
--start-epoch 0 \
|
||||
--exp-dir transducer_stateless/exp \
|
||||
--max-duration 200 \
|
||||
```
|
||||
|
||||
The tensorboard training log can be found at
|
||||
https://tensorboard.dev/experiment/DnRwoZF8RRyod4kkfG5q5Q/#scalars
|
||||
|
||||
The decoding command is:
|
||||
```
|
||||
epoch=29
|
||||
avg=15
|
||||
|
||||
## greedy search
|
||||
./transducer_stateless/decode.py \
|
||||
--epoch $epoch \
|
||||
--avg $avg \
|
||||
--exp-dir transducer_stateless/exp \
|
||||
--bpe-model ./data/lang_bpe_500/bpe.model \
|
||||
--max-duration 100
|
||||
|
||||
## beam search
|
||||
./transducer_stateless/decode.py \
|
||||
--epoch $epoch \
|
||||
--avg $avg \
|
||||
--exp-dir transducer_stateless/exp \
|
||||
--bpe-model ./data/lang_bpe_500/bpe.model \
|
||||
--max-duration 100 \
|
||||
--decoding-method beam_search \
|
||||
--beam-size 4
|
||||
|
||||
## modified beam search
|
||||
./transducer_stateless/decode.py \
|
||||
--epoch $epoch \
|
||||
--avg $avg \
|
||||
--exp-dir transducer_stateless/exp \
|
||||
--bpe-model ./data/lang_bpe_500/bpe.model \
|
||||
--max-duration 100 \
|
||||
--decoding-method beam_search \
|
||||
--beam-size 4
|
||||
```
|
||||
243
egs/tedlium3/ASR/prepare.sh
Normal file
243
egs/tedlium3/ASR/prepare.sh
Normal file
@ -0,0 +1,243 @@
|
||||
#!/usr/bin/env bash
|
||||
|
||||
set -eou pipefail
|
||||
|
||||
nj=15
|
||||
stage=-1
|
||||
stop_stage=100
|
||||
|
||||
# We assume dl_dir (download dir) contains the following
|
||||
# directories and files. If not, they will be downloaded
|
||||
# by this script automatically.
|
||||
#
|
||||
# - $dl_dir/tedlium3
|
||||
# You can find data, doc, legacy, LM, etc, inside it.
|
||||
# You can download them from https://www.openslr.org/51
|
||||
#
|
||||
# - $dl_dir/lm
|
||||
# This directory contains the language model(LM) downloaded from
|
||||
# https://huggingface.co/luomingshuang/tedlium3_lm. About how to get these LM files, you can know it
|
||||
# from https://github.com/luomingshuang/Train_LM_with_kaldilm.
|
||||
#
|
||||
# - lm_3_gram.arpa
|
||||
# - lm_4_gram.arpa
|
||||
#
|
||||
# - $dl_dir/musan
|
||||
# This directory contains the following directories downloaded from
|
||||
# http://www.openslr.org/17/
|
||||
#
|
||||
# - music
|
||||
# - noise
|
||||
# - speech
|
||||
dl_dir=$PWD/download
|
||||
|
||||
. shared/parse_options.sh || exit 1
|
||||
|
||||
# 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 "dl_dir: $dl_dir"
|
||||
|
||||
if [ $stage -le -1 ] && [ $stop_stage -ge -1 ]; then
|
||||
log "Stage -1: Download LM"
|
||||
# We assume that you have installed the git-lfs, if not, you could install it
|
||||
# using: `sudo apt-get install git-lfs && git-lfs install`
|
||||
[ ! -e $dl_dir/lm ] && mkdir -p $dl_dir/lm
|
||||
git clone https://huggingface.co/luomingshuang/tedlium3_lm $dl_dir/lm
|
||||
cd $dl_dir/lm && git lfs pull
|
||||
|
||||
# If you want to download Tedlium 4 gram language models
|
||||
# using the follow commands:
|
||||
#wget --continue http://kaldi-asr.org/models/5/4gram_small.arpa.gz -P $dl_dir/lm/ || exit 1
|
||||
#wget --continue http://kaldi-asr.org/models/5/4gram_big.arpa.gz -P $dl_dir/lm/ || exit 1
|
||||
fi
|
||||
|
||||
if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
|
||||
log "Stage 0: Download data"
|
||||
|
||||
# If you have pre-downloaded it to /path/to/LibriSpeech,
|
||||
# you can create a symlink
|
||||
#
|
||||
# ln -sfv /path/to/tedlium3 $dl_dir/tedlium3
|
||||
#
|
||||
if [ ! -d $dl_dir/tedlium ]; then
|
||||
lhotse download tedlium $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/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 tedlium3 manifest"
|
||||
# We assume that you have downloaded the tedlium3 corpus
|
||||
# to $dl_dir/tedlium3
|
||||
mkdir -p data/manifests
|
||||
lhotse prepare tedlium $dl_dir/tedlium3 data/manifests
|
||||
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
|
||||
mkdir -p data/manifests
|
||||
lhotse prepare musan $dl_dir/musan data/manifests
|
||||
fi
|
||||
|
||||
if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
|
||||
log "Stage 3: Compute fbank for tedlium3"
|
||||
mkdir -p data/fbank
|
||||
./local/compute_fbank_tedlium.py
|
||||
fi
|
||||
|
||||
if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
|
||||
log "Stage 4: Compute fbank for musan"
|
||||
mkdir -p data/fbank
|
||||
./local/compute_fbank_musan.py
|
||||
fi
|
||||
|
||||
if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
|
||||
log "Stage 5: Prepare phone based lang"
|
||||
lang_dir=data/lang_phone
|
||||
mkdir -p $lang_dir
|
||||
|
||||
if [ ! -f $lang_dir/train.text ]; then
|
||||
./local/prepare_transcripts.py \
|
||||
--lang-dir $lang_dir \
|
||||
--manifests-dir data/manifests
|
||||
|
||||
cat download/tedlium3/TEDLIUM.152k.dic |
|
||||
grep -v -w "<s>" |
|
||||
grep -v -w "</s>" |
|
||||
grep -v -w "<unk>" |
|
||||
LANG= LC_ALL= sort |
|
||||
sed 's:([0-9])::g' > $lang_dir/lexicon_words.txt
|
||||
|
||||
(echo '<UNK> <UNK>'; ) |
|
||||
cat - $lang_dir/lexicon_words.txt |
|
||||
sort | uniq > $lang_dir/lexicon.txt
|
||||
|
||||
if [ ! -f $lang_dir/L_disambig.pt ]; then
|
||||
./local/prepare_lang.py --lang-dir $lang_dir
|
||||
fi
|
||||
fi
|
||||
|
||||
if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
|
||||
log "Stage 6: 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
|
||||
|
||||
if [ ! -f $lang_dir/transcript_words.txt ]; then
|
||||
log "Generate data for BPE training"
|
||||
cat data/lang_phone/train.text | cut -d " " -f 2-
|
||||
> $lang_dir/transcript_words.txt
|
||||
# remove the <unk> for transcript_words.txt
|
||||
sed -i 's/ <unk>//g' $lang_dir/transcript_words.txt
|
||||
sed -i 's/<unk> //g' $lang_dir/transcript_words.txt
|
||||
sed -i 's/<unk>//g' $lang_dir/transcript_words.txt
|
||||
fi
|
||||
|
||||
./local/train_bpe_model.py \
|
||||
--lang-dir $lang_dir \
|
||||
--vocab-size $vocab_size \
|
||||
--transcript $lang_dir/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 7 ] && [ $stop_stage -ge 7 ]; then
|
||||
log "Stage 7: Prepare bigram P"
|
||||
|
||||
for vocab_size in ${vocab_sizes[@]}; do
|
||||
lang_dir=data/lang_bpe_${vocab_size}
|
||||
|
||||
if [ ! -f $lang_dir/transcript_tokens.txt ]; then
|
||||
./local/convert_transcript_words_to_tokens.py \
|
||||
--lexicon $lang_dir/lexicon.txt \
|
||||
--transcript $lang_dir/transcript_words.txt \
|
||||
--oov "<UNK>" \
|
||||
> $lang_dir/transcript_tokens.txt
|
||||
fi
|
||||
|
||||
if [ ! -f $lang_dir/P.arpa ]; then
|
||||
./shared/make_kn_lm.py \
|
||||
-ngram-order 2 \
|
||||
-text $lang_dir/transcript_tokens.txt \
|
||||
-lm $lang_dir/P.arpa
|
||||
fi
|
||||
|
||||
if [ ! -f $lang_dir/P.fst.txt ]; then
|
||||
python3 -m kaldilm \
|
||||
--read-symbol-table="$lang_dir/tokens.txt" \
|
||||
--disambig-symbol='#0' \
|
||||
--max-order=2 \
|
||||
$lang_dir/P.arpa > $lang_dir/P.fst.txt
|
||||
fi
|
||||
done
|
||||
fi
|
||||
|
||||
if [ $stage -le 8 ] && [ $stop_stage -ge 8 ]; then
|
||||
log "Stage 8: Prepare G"
|
||||
# We assume you have install kaldilm, if not, please install
|
||||
# it using: pip install kaldilm
|
||||
|
||||
mkdir -p data/lm
|
||||
if [ ! -f data/lm/G_3_gram.fst.txt ]; then
|
||||
# It is used in building HLG
|
||||
python3 -m kaldilm \
|
||||
--read-symbol-table="data/lang_phone/words.txt" \
|
||||
--disambig-symbol='#0' \
|
||||
--max-order=3 \
|
||||
data/lm/lm_3_gram.arpa > data/lm/G_3_gram.fst.txt
|
||||
fi
|
||||
|
||||
if [ ! -f data/lm/G_4_gram.fst.txt ]; then
|
||||
# It is used for LM rescoring
|
||||
python3 -m kaldilm \
|
||||
--read-symbol-table="data/lang_phone/words.txt" \
|
||||
--disambig-symbol='#0' \
|
||||
--max-order=4 \
|
||||
data/lm/lm_4_gram.arpa > data/lm/G_4_gram.fst.txt
|
||||
fi
|
||||
fi
|
||||
echo 'completing the G building....'
|
||||
if [ $stage -le 9 ] && [ $stop_stage -ge 9 ]; then
|
||||
log "Stage 9: 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
|
||||
Loading…
x
Reference in New Issue
Block a user