Merge branch 'k2-fsa:master' into fisher_swbd
5
.flake8
@ -1,7 +1,7 @@
|
|||||||
[flake8]
|
[flake8]
|
||||||
show-source=true
|
show-source=true
|
||||||
statistics=true
|
statistics=true
|
||||||
max-line-length = 80
|
max-line-length = 88
|
||||||
per-file-ignores =
|
per-file-ignores =
|
||||||
# line too long
|
# line too long
|
||||||
icefall/diagnostics.py: E501,
|
icefall/diagnostics.py: E501,
|
||||||
@ -11,7 +11,8 @@ per-file-ignores =
|
|||||||
egs/*/ASR/*/scaling.py: E501,
|
egs/*/ASR/*/scaling.py: E501,
|
||||||
egs/librispeech/ASR/lstm_transducer_stateless*/*.py: E501, E203
|
egs/librispeech/ASR/lstm_transducer_stateless*/*.py: E501, E203
|
||||||
egs/librispeech/ASR/conv_emformer_transducer_stateless*/*.py: E501, E203
|
egs/librispeech/ASR/conv_emformer_transducer_stateless*/*.py: E501, E203
|
||||||
egs/librispeech/ASR/conformer_ctc2/*py: E501,
|
egs/librispeech/ASR/conformer_ctc*/*py: E501,
|
||||||
|
egs/librispeech/ASR/zipformer_mmi/*.py: E501, E203
|
||||||
egs/librispeech/ASR/RESULTS.md: E999,
|
egs/librispeech/ASR/RESULTS.md: E999,
|
||||||
|
|
||||||
# invalid escape sequence (cause by tex formular), W605
|
# invalid escape sequence (cause by tex formular), W605
|
||||||
|
3
.git-blame-ignore-revs
Normal file
@ -0,0 +1,3 @@
|
|||||||
|
# Migrate to 88 characters per line (see: https://github.com/lhotse-speech/lhotse/issues/890)
|
||||||
|
107df3b115a58f1b68a6458c3f94a130004be34c
|
||||||
|
d31db010371a4128856480382876acdc0d1739ed
|
@ -4,6 +4,8 @@
|
|||||||
# The computed features are saved to ~/tmp/fbank-libri and are
|
# The computed features are saved to ~/tmp/fbank-libri and are
|
||||||
# cached for later runs
|
# cached for later runs
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
export PYTHONPATH=$PWD:$PYTHONPATH
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
echo $PYTHONPATH
|
echo $PYTHONPATH
|
||||||
|
|
||||||
|
@ -6,6 +6,8 @@
|
|||||||
# You will find directories `~/tmp/giga-dev-dataset-fbank` after running
|
# You will find directories `~/tmp/giga-dev-dataset-fbank` after running
|
||||||
# this script.
|
# this script.
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
mkdir -p ~/tmp
|
mkdir -p ~/tmp
|
||||||
cd ~/tmp
|
cd ~/tmp
|
||||||
|
|
||||||
|
@ -7,6 +7,8 @@
|
|||||||
# You will find directories ~/tmp/download/LibriSpeech after running
|
# You will find directories ~/tmp/download/LibriSpeech after running
|
||||||
# this script.
|
# this script.
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
mkdir ~/tmp/download
|
mkdir ~/tmp/download
|
||||||
cd egs/librispeech/ASR
|
cd egs/librispeech/ASR
|
||||||
ln -s ~/tmp/download .
|
ln -s ~/tmp/download .
|
||||||
|
2
.github/scripts/install-kaldifeat.sh
vendored
@ -3,6 +3,8 @@
|
|||||||
# This script installs kaldifeat into the directory ~/tmp/kaldifeat
|
# This script installs kaldifeat into the directory ~/tmp/kaldifeat
|
||||||
# which is cached by GitHub actions for later runs.
|
# which is cached by GitHub actions for later runs.
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
mkdir -p ~/tmp
|
mkdir -p ~/tmp
|
||||||
cd ~/tmp
|
cd ~/tmp
|
||||||
git clone https://github.com/csukuangfj/kaldifeat
|
git clone https://github.com/csukuangfj/kaldifeat
|
||||||
|
@ -4,6 +4,8 @@
|
|||||||
# to egs/librispeech/ASR/download/LibriSpeech and generates manifest
|
# to egs/librispeech/ASR/download/LibriSpeech and generates manifest
|
||||||
# files in egs/librispeech/ASR/data/manifests
|
# files in egs/librispeech/ASR/data/manifests
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
cd egs/librispeech/ASR
|
cd egs/librispeech/ASR
|
||||||
[ ! -e download ] && ln -s ~/tmp/download .
|
[ ! -e download ] && ln -s ~/tmp/download .
|
||||||
mkdir -p data/manifests
|
mkdir -p data/manifests
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
@ -40,7 +42,7 @@ for sym in 1 2 3; do
|
|||||||
--lang-dir $repo/data/lang_char \
|
--lang-dir $repo/data/lang_char \
|
||||||
$repo/test_wavs/BAC009S0764W0121.wav \
|
$repo/test_wavs/BAC009S0764W0121.wav \
|
||||||
$repo/test_wavs/BAC009S0764W0122.wav \
|
$repo/test_wavs/BAC009S0764W0122.wav \
|
||||||
$rep/test_wavs/BAC009S0764W0123.wav
|
$repo/test_wavs/BAC009S0764W0123.wav
|
||||||
done
|
done
|
||||||
|
|
||||||
for method in modified_beam_search beam_search fast_beam_search; do
|
for method in modified_beam_search beam_search fast_beam_search; do
|
||||||
@ -53,7 +55,7 @@ for method in modified_beam_search beam_search fast_beam_search; do
|
|||||||
--lang-dir $repo/data/lang_char \
|
--lang-dir $repo/data/lang_char \
|
||||||
$repo/test_wavs/BAC009S0764W0121.wav \
|
$repo/test_wavs/BAC009S0764W0121.wav \
|
||||||
$repo/test_wavs/BAC009S0764W0122.wav \
|
$repo/test_wavs/BAC009S0764W0122.wav \
|
||||||
$rep/test_wavs/BAC009S0764W0123.wav
|
$repo/test_wavs/BAC009S0764W0123.wav
|
||||||
done
|
done
|
||||||
|
|
||||||
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
123
.github/scripts/run-librispeech-conformer-ctc3-2022-11-28.sh
vendored
Executable file
@ -0,0 +1,123 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
|
log() {
|
||||||
|
# This function is from espnet
|
||||||
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
|
||||||
|
}
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
repo_url=https://huggingface.co/Zengwei/icefall-asr-librispeech-conformer-ctc3-2022-11-27
|
||||||
|
|
||||||
|
log "Downloading pre-trained model from $repo_url"
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
|
||||||
|
repo=$(basename $repo_url)
|
||||||
|
|
||||||
|
log "Display test files"
|
||||||
|
tree $repo/
|
||||||
|
soxi $repo/test_wavs/*.wav
|
||||||
|
ls -lh $repo/test_wavs/*.wav
|
||||||
|
|
||||||
|
pushd $repo/exp
|
||||||
|
git lfs pull --include "data/lang_bpe_500/HLG.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/L.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/LG.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/Linv.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/bpe.model"
|
||||||
|
git lfs pull --include "data/lm/G_4_gram.pt"
|
||||||
|
git lfs pull --include "exp/jit_trace.pt"
|
||||||
|
git lfs pull --include "exp/pretrained.pt"
|
||||||
|
ln -s pretrained.pt epoch-99.pt
|
||||||
|
ls -lh *.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
log "Decode with models exported by torch.jit.trace()"
|
||||||
|
|
||||||
|
for m in ctc-decoding 1best; do
|
||||||
|
./conformer_ctc3/jit_pretrained.py \
|
||||||
|
--model-filename $repo/exp/jit_trace.pt \
|
||||||
|
--words-file $repo/data/lang_bpe_500/words.txt \
|
||||||
|
--HLG $repo/data/lang_bpe_500/HLG.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--G $repo/data/lm/G_4_gram.pt \
|
||||||
|
--method $m \
|
||||||
|
--sample-rate 16000 \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
log "Export to torchscript model"
|
||||||
|
|
||||||
|
./conformer_ctc3/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--lang-dir $repo/data/lang_bpe_500 \
|
||||||
|
--jit-trace 1 \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--use-averaged-model 0
|
||||||
|
|
||||||
|
ls -lh $repo/exp/*.pt
|
||||||
|
|
||||||
|
log "Decode with models exported by torch.jit.trace()"
|
||||||
|
|
||||||
|
for m in ctc-decoding 1best; do
|
||||||
|
./conformer_ctc3/jit_pretrained.py \
|
||||||
|
--model-filename $repo/exp/jit_trace.pt \
|
||||||
|
--words-file $repo/data/lang_bpe_500/words.txt \
|
||||||
|
--HLG $repo/data/lang_bpe_500/HLG.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--G $repo/data/lm/G_4_gram.pt \
|
||||||
|
--method $m \
|
||||||
|
--sample-rate 16000 \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
for m in ctc-decoding 1best; do
|
||||||
|
./conformer_ctc3/pretrained.py \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--words-file $repo/data/lang_bpe_500/words.txt \
|
||||||
|
--HLG $repo/data/lang_bpe_500/HLG.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--G $repo/data/lm/G_4_gram.pt \
|
||||||
|
--method $m \
|
||||||
|
--sample-rate 16000 \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
||||||
|
echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
|
||||||
|
if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode" ]]; then
|
||||||
|
mkdir -p conformer_ctc3/exp
|
||||||
|
ln -s $PWD/$repo/exp/pretrained.pt conformer_ctc3/exp/epoch-999.pt
|
||||||
|
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
||||||
|
|
||||||
|
ls -lh data
|
||||||
|
ls -lh conformer_ctc3/exp
|
||||||
|
|
||||||
|
log "Decoding test-clean and test-other"
|
||||||
|
|
||||||
|
# use a small value for decoding with CPU
|
||||||
|
max_duration=100
|
||||||
|
|
||||||
|
for method in ctc-decoding 1best; do
|
||||||
|
log "Decoding with $method"
|
||||||
|
./conformer_ctc3/decode.py \
|
||||||
|
--epoch 999 \
|
||||||
|
--avg 1 \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--exp-dir conformer_ctc3/exp/ \
|
||||||
|
--max-duration $max_duration \
|
||||||
|
--decoding-method $method \
|
||||||
|
--lm-dir data/lm
|
||||||
|
done
|
||||||
|
|
||||||
|
rm conformer_ctc3/exp/*.pt
|
||||||
|
fi
|
79
.github/scripts/run-librispeech-conv-emformer-transducer-stateless2-2022-12-05.sh
vendored
Executable file
@ -0,0 +1,79 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
#
|
||||||
|
set -e
|
||||||
|
|
||||||
|
log() {
|
||||||
|
# This function is from espnet
|
||||||
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
|
||||||
|
}
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
repo_url=https://huggingface.co/Zengwei/icefall-asr-librispeech-conv-emformer-transducer-stateless2-2022-07-05
|
||||||
|
|
||||||
|
log "Downloading pre-trained model from $repo_url"
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
|
||||||
|
repo=$(basename $repo_url)
|
||||||
|
pushd $repo
|
||||||
|
git lfs pull --include "exp/pretrained-epoch-30-avg-10-averaged.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/bpe.model"
|
||||||
|
cd exp
|
||||||
|
ln -s pretrained-epoch-30-avg-10-averaged.pt epoch-99.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
log "Display test files"
|
||||||
|
tree $repo/
|
||||||
|
soxi $repo/test_wavs/*.wav
|
||||||
|
ls -lh $repo/test_wavs/*.wav
|
||||||
|
|
||||||
|
log "Install ncnn and pnnx"
|
||||||
|
|
||||||
|
# We are using a modified ncnn here. Will try to merge it to the official repo
|
||||||
|
# of ncnn
|
||||||
|
git clone https://github.com/csukuangfj/ncnn
|
||||||
|
pushd ncnn
|
||||||
|
git submodule init
|
||||||
|
git submodule update python/pybind11
|
||||||
|
python3 setup.py bdist_wheel
|
||||||
|
ls -lh dist/
|
||||||
|
pip install dist/*.whl
|
||||||
|
cd tools/pnnx
|
||||||
|
mkdir build
|
||||||
|
cd build
|
||||||
|
cmake -D Python3_EXECUTABLE=/opt/hostedtoolcache/Python/3.8.14/x64/bin/python3 ..
|
||||||
|
make -j4 pnnx
|
||||||
|
|
||||||
|
./src/pnnx || echo "pass"
|
||||||
|
|
||||||
|
popd
|
||||||
|
|
||||||
|
log "Test exporting to pnnx format"
|
||||||
|
|
||||||
|
./conv_emformer_transducer_stateless2/export-for-ncnn.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
\
|
||||||
|
--num-encoder-layers 12 \
|
||||||
|
--chunk-length 32 \
|
||||||
|
--cnn-module-kernel 31 \
|
||||||
|
--left-context-length 32 \
|
||||||
|
--right-context-length 8 \
|
||||||
|
--memory-size 32
|
||||||
|
|
||||||
|
./ncnn/tools/pnnx/build/src/pnnx $repo/exp/encoder_jit_trace-pnnx.pt
|
||||||
|
./ncnn/tools/pnnx/build/src/pnnx $repo/exp/decoder_jit_trace-pnnx.pt
|
||||||
|
./ncnn/tools/pnnx/build/src/pnnx $repo/exp/joiner_jit_trace-pnnx.pt
|
||||||
|
|
||||||
|
./conv_emformer_transducer_stateless2/streaming-ncnn-decode.py \
|
||||||
|
--tokens $repo/data/lang_bpe_500/tokens.txt \
|
||||||
|
--encoder-param-filename $repo/exp/encoder_jit_trace-pnnx.ncnn.param \
|
||||||
|
--encoder-bin-filename $repo/exp/encoder_jit_trace-pnnx.ncnn.bin \
|
||||||
|
--decoder-param-filename $repo/exp/decoder_jit_trace-pnnx.ncnn.param \
|
||||||
|
--decoder-bin-filename $repo/exp/decoder_jit_trace-pnnx.ncnn.bin \
|
||||||
|
--joiner-param-filename $repo/exp/joiner_jit_trace-pnnx.ncnn.param \
|
||||||
|
--joiner-bin-filename $repo/exp/joiner_jit_trace-pnnx.ncnn.bin \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav
|
@ -1,4 +1,6 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
#
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
@ -14,6 +16,7 @@ log "Downloading pre-trained model from $repo_url"
|
|||||||
git lfs install
|
git lfs install
|
||||||
git clone $repo_url
|
git clone $repo_url
|
||||||
repo=$(basename $repo_url)
|
repo=$(basename $repo_url)
|
||||||
|
abs_repo=$(realpath $repo)
|
||||||
|
|
||||||
log "Display test files"
|
log "Display test files"
|
||||||
tree $repo/
|
tree $repo/
|
||||||
@ -103,6 +106,47 @@ log "Decode with models exported by torch.jit.trace()"
|
|||||||
$repo/test_wavs/1221-135766-0001.wav \
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
$repo/test_wavs/1221-135766-0002.wav
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
|
||||||
|
log "Test exporting to ONNX"
|
||||||
|
|
||||||
|
./lstm_transducer_stateless2/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--onnx 1
|
||||||
|
|
||||||
|
log "Decode with ONNX models "
|
||||||
|
|
||||||
|
./lstm_transducer_stateless2/streaming-onnx-decode.py \
|
||||||
|
--bpe-model-filename $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--encoder-model-filename $repo//exp/encoder.onnx \
|
||||||
|
--decoder-model-filename $repo/exp/decoder.onnx \
|
||||||
|
--joiner-model-filename $repo/exp/joiner.onnx \
|
||||||
|
--joiner-encoder-proj-model-filename $repo/exp/joiner_encoder_proj.onnx \
|
||||||
|
--joiner-decoder-proj-model-filename $repo/exp/joiner_decoder_proj.onnx \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav
|
||||||
|
|
||||||
|
./lstm_transducer_stateless2/streaming-onnx-decode.py \
|
||||||
|
--bpe-model-filename $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--encoder-model-filename $repo//exp/encoder.onnx \
|
||||||
|
--decoder-model-filename $repo/exp/decoder.onnx \
|
||||||
|
--joiner-model-filename $repo/exp/joiner.onnx \
|
||||||
|
--joiner-encoder-proj-model-filename $repo/exp/joiner_encoder_proj.onnx \
|
||||||
|
--joiner-decoder-proj-model-filename $repo/exp/joiner_decoder_proj.onnx \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav
|
||||||
|
|
||||||
|
./lstm_transducer_stateless2/streaming-onnx-decode.py \
|
||||||
|
--bpe-model-filename $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--encoder-model-filename $repo//exp/encoder.onnx \
|
||||||
|
--decoder-model-filename $repo/exp/decoder.onnx \
|
||||||
|
--joiner-model-filename $repo/exp/joiner.onnx \
|
||||||
|
--joiner-encoder-proj-model-filename $repo/exp/joiner_encoder_proj.onnx \
|
||||||
|
--joiner-decoder-proj-model-filename $repo/exp/joiner_decoder_proj.onnx \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
for sym in 1 2 3; do
|
for sym in 1 2 3; do
|
||||||
log "Greedy search with --max-sym-per-frame $sym"
|
log "Greedy search with --max-sym-per-frame $sym"
|
||||||
|
|
||||||
@ -131,7 +175,89 @@ done
|
|||||||
|
|
||||||
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
||||||
echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
|
echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
|
||||||
if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"ncnn" ]]; then
|
|
||||||
|
if [[ x"${GITHUB_EVENT_LABEL_NAME}" == x"shallow-fusion" ]]; then
|
||||||
|
lm_repo_url=https://huggingface.co/ezerhouni/icefall-librispeech-rnn-lm
|
||||||
|
log "Download pre-trained RNN-LM model from ${lm_repo_url}"
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $lm_repo_url
|
||||||
|
lm_repo=$(basename $lm_repo_url)
|
||||||
|
pushd $lm_repo
|
||||||
|
git lfs pull --include "exp/pretrained.pt"
|
||||||
|
mv exp/pretrained.pt exp/epoch-88.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
mkdir -p lstm_transducer_stateless2/exp
|
||||||
|
ln -sf $PWD/$repo/exp/pretrained.pt lstm_transducer_stateless2/exp/epoch-999.pt
|
||||||
|
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
||||||
|
|
||||||
|
ls -lh data
|
||||||
|
ls -lh lstm_transducer_stateless2/exp
|
||||||
|
|
||||||
|
log "Decoding test-clean and test-other"
|
||||||
|
|
||||||
|
./lstm_transducer_stateless2/decode.py \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--epoch 999 \
|
||||||
|
--avg 1 \
|
||||||
|
--exp-dir lstm_transducer_stateless2/exp \
|
||||||
|
--max-duration 600 \
|
||||||
|
--decoding-method modified_beam_search_rnnlm_shallow_fusion \
|
||||||
|
--beam 4 \
|
||||||
|
--rnn-lm-scale 0.3 \
|
||||||
|
--rnn-lm-exp-dir $lm_repo/exp \
|
||||||
|
--rnn-lm-epoch 88 \
|
||||||
|
--rnn-lm-avg 1 \
|
||||||
|
--rnn-lm-num-layers 3 \
|
||||||
|
--rnn-lm-tie-weights 1
|
||||||
|
fi
|
||||||
|
|
||||||
|
if [[ x"${GITHUB_EVENT_LABEL_NAME}" == x"LODR" ]]; then
|
||||||
|
bigram_repo_url=https://huggingface.co/marcoyang/librispeech_bigram
|
||||||
|
log "Download bi-gram LM from ${bigram_repo_url}"
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $bigram_repo_url
|
||||||
|
bigramlm_repo=$(basename $bigram_repo_url)
|
||||||
|
pushd $bigramlm_repo
|
||||||
|
git lfs pull --include "2gram.fst.txt"
|
||||||
|
cp 2gram.fst.txt $abs_repo/data/lang_bpe_500/.
|
||||||
|
popd
|
||||||
|
|
||||||
|
lm_repo_url=https://huggingface.co/ezerhouni/icefall-librispeech-rnn-lm
|
||||||
|
log "Download pre-trained RNN-LM model from ${lm_repo_url}"
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $lm_repo_url
|
||||||
|
lm_repo=$(basename $lm_repo_url)
|
||||||
|
pushd $lm_repo
|
||||||
|
git lfs pull --include "exp/pretrained.pt"
|
||||||
|
mv exp/pretrained.pt exp/epoch-88.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
mkdir -p lstm_transducer_stateless2/exp
|
||||||
|
ln -sf $PWD/$repo/exp/pretrained.pt lstm_transducer_stateless2/exp/epoch-999.pt
|
||||||
|
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
||||||
|
|
||||||
|
ls -lh data
|
||||||
|
ls -lh lstm_transducer_stateless2/exp
|
||||||
|
|
||||||
|
log "Decoding test-clean and test-other"
|
||||||
|
|
||||||
|
./lstm_transducer_stateless2/decode.py \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--epoch 999 \
|
||||||
|
--avg 1 \
|
||||||
|
--exp-dir lstm_transducer_stateless2/exp \
|
||||||
|
--max-duration 600 \
|
||||||
|
--decoding-method modified_beam_search_rnnlm_LODR \
|
||||||
|
--beam 4 \
|
||||||
|
--rnn-lm-scale 0.3 \
|
||||||
|
--rnn-lm-exp-dir $lm_repo/exp \
|
||||||
|
--rnn-lm-epoch 88 \
|
||||||
|
--rnn-lm-avg 1 \
|
||||||
|
--rnn-lm-num-layers 3 \
|
||||||
|
--rnn-lm-tie-weights 1 \
|
||||||
|
--tokens-ngram 2 \
|
||||||
|
--ngram-lm-scale -0.16
|
||||||
|
fi
|
||||||
|
|
||||||
|
if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" ]]; then
|
||||||
mkdir -p lstm_transducer_stateless2/exp
|
mkdir -p lstm_transducer_stateless2/exp
|
||||||
ln -s $PWD/$repo/exp/pretrained.pt lstm_transducer_stateless2/exp/epoch-999.pt
|
ln -s $PWD/$repo/exp/pretrained.pt lstm_transducer_stateless2/exp/epoch-999.pt
|
||||||
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
@ -11,10 +13,14 @@ cd egs/librispeech/ASR
|
|||||||
repo_url=https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless2-2022-04-29
|
repo_url=https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless2-2022-04-29
|
||||||
|
|
||||||
log "Downloading pre-trained model from $repo_url"
|
log "Downloading pre-trained model from $repo_url"
|
||||||
git lfs install
|
GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
|
||||||
git clone $repo_url
|
|
||||||
repo=$(basename $repo_url)
|
repo=$(basename $repo_url)
|
||||||
|
|
||||||
|
pushd $repo
|
||||||
|
git lfs pull --include "data/lang_bpe_500/bpe.model"
|
||||||
|
git lfs pull --include "exp/pretrained-epoch-38-avg-10.pt"
|
||||||
|
popd
|
||||||
|
|
||||||
log "Display test files"
|
log "Display test files"
|
||||||
tree $repo/
|
tree $repo/
|
||||||
soxi $repo/test_wavs/*.wav
|
soxi $repo/test_wavs/*.wav
|
||||||
@ -77,4 +83,5 @@ if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" ==
|
|||||||
done
|
done
|
||||||
|
|
||||||
rm pruned_transducer_stateless2/exp/*.pt
|
rm pruned_transducer_stateless2/exp/*.pt
|
||||||
|
rm -r data/lang_bpe_500
|
||||||
fi
|
fi
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
@ -11,9 +13,12 @@ cd egs/librispeech/ASR
|
|||||||
repo_url=https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-04-29
|
repo_url=https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-04-29
|
||||||
|
|
||||||
log "Downloading pre-trained model from $repo_url"
|
log "Downloading pre-trained model from $repo_url"
|
||||||
git lfs install
|
GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
|
||||||
git clone $repo_url
|
|
||||||
repo=$(basename $repo_url)
|
repo=$(basename $repo_url)
|
||||||
|
pushd $repo
|
||||||
|
git lfs pull --include "data/lang_bpe_500/bpe.model"
|
||||||
|
git lfs pull --include "exp/pretrained-epoch-25-avg-6.pt"
|
||||||
|
popd
|
||||||
|
|
||||||
log "Display test files"
|
log "Display test files"
|
||||||
tree $repo/
|
tree $repo/
|
||||||
@ -77,4 +82,5 @@ if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" ==
|
|||||||
done
|
done
|
||||||
|
|
||||||
rm pruned_transducer_stateless3/exp/*.pt
|
rm pruned_transducer_stateless3/exp/*.pt
|
||||||
|
rm -r data/lang_bpe_500
|
||||||
fi
|
fi
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
@ -58,17 +60,17 @@ log "Decode with ONNX models"
|
|||||||
--jit-filename $repo/exp/cpu_jit.pt \
|
--jit-filename $repo/exp/cpu_jit.pt \
|
||||||
--onnx-encoder-filename $repo/exp/encoder.onnx \
|
--onnx-encoder-filename $repo/exp/encoder.onnx \
|
||||||
--onnx-decoder-filename $repo/exp/decoder.onnx \
|
--onnx-decoder-filename $repo/exp/decoder.onnx \
|
||||||
--onnx-joiner-filename $repo/exp/joiner.onnx
|
--onnx-joiner-filename $repo/exp/joiner.onnx \
|
||||||
|
--onnx-joiner-encoder-proj-filename $repo/exp/joiner_encoder_proj.onnx \
|
||||||
./pruned_transducer_stateless3/onnx_check_all_in_one.py \
|
--onnx-joiner-decoder-proj-filename $repo/exp/joiner_decoder_proj.onnx
|
||||||
--jit-filename $repo/exp/cpu_jit.pt \
|
|
||||||
--onnx-all-in-one-filename $repo/exp/all_in_one.onnx
|
|
||||||
|
|
||||||
./pruned_transducer_stateless3/onnx_pretrained.py \
|
./pruned_transducer_stateless3/onnx_pretrained.py \
|
||||||
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
--encoder-model-filename $repo/exp/encoder.onnx \
|
--encoder-model-filename $repo/exp/encoder.onnx \
|
||||||
--decoder-model-filename $repo/exp/decoder.onnx \
|
--decoder-model-filename $repo/exp/decoder.onnx \
|
||||||
--joiner-model-filename $repo/exp/joiner.onnx \
|
--joiner-model-filename $repo/exp/joiner.onnx \
|
||||||
|
--joiner-encoder-proj-model-filename $repo/exp/joiner_encoder_proj.onnx \
|
||||||
|
--joiner-decoder-proj-model-filename $repo/exp/joiner_decoder_proj.onnx \
|
||||||
$repo/test_wavs/1089-134686-0001.wav \
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
$repo/test_wavs/1221-135766-0001.wav \
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
$repo/test_wavs/1221-135766-0002.wav
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
107
.github/scripts/run-librispeech-pruned-transducer-stateless7-2022-11-11.sh
vendored
Executable file
@ -0,0 +1,107 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
|
log() {
|
||||||
|
# This function is from espnet
|
||||||
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
|
||||||
|
}
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
repo_url=https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11
|
||||||
|
|
||||||
|
log "Downloading pre-trained model from $repo_url"
|
||||||
|
git lfs install
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
|
||||||
|
repo=$(basename $repo_url)
|
||||||
|
|
||||||
|
log "Display test files"
|
||||||
|
tree $repo/
|
||||||
|
soxi $repo/test_wavs/*.wav
|
||||||
|
ls -lh $repo/test_wavs/*.wav
|
||||||
|
|
||||||
|
pushd $repo/exp
|
||||||
|
git lfs pull --include "data/lang_bpe_500/bpe.model"
|
||||||
|
git lfs pull --include "exp/cpu_jit.pt"
|
||||||
|
git lfs pull --include "exp/pretrained.pt"
|
||||||
|
ln -s pretrained.pt epoch-99.pt
|
||||||
|
ls -lh *.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
log "Export to torchscript model"
|
||||||
|
./pruned_transducer_stateless7/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--use-averaged-model false \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--jit 1
|
||||||
|
|
||||||
|
ls -lh $repo/exp/*.pt
|
||||||
|
|
||||||
|
log "Decode with models exported by torch.jit.script()"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless7/jit_pretrained.py \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--nn-model-filename $repo/exp/cpu_jit.pt \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
|
||||||
|
for sym in 1 2 3; do
|
||||||
|
log "Greedy search with --max-sym-per-frame $sym"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless7/pretrained.py \
|
||||||
|
--method greedy_search \
|
||||||
|
--max-sym-per-frame $sym \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
for method in modified_beam_search beam_search fast_beam_search; do
|
||||||
|
log "$method"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless7/pretrained.py \
|
||||||
|
--method $method \
|
||||||
|
--beam-size 4 \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
||||||
|
echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
|
||||||
|
if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode" ]]; then
|
||||||
|
mkdir -p pruned_transducer_stateless7/exp
|
||||||
|
ln -s $PWD/$repo/exp/pretrained.pt pruned_transducer_stateless7/exp/epoch-999.pt
|
||||||
|
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
||||||
|
|
||||||
|
ls -lh data
|
||||||
|
ls -lh pruned_transducer_stateless7/exp
|
||||||
|
|
||||||
|
log "Decoding test-clean and test-other"
|
||||||
|
|
||||||
|
# use a small value for decoding with CPU
|
||||||
|
max_duration=100
|
||||||
|
|
||||||
|
for method in greedy_search fast_beam_search modified_beam_search; do
|
||||||
|
log "Decoding with $method"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless7/decode.py \
|
||||||
|
--decoding-method $method \
|
||||||
|
--epoch 999 \
|
||||||
|
--avg 1 \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--max-duration $max_duration \
|
||||||
|
--exp-dir pruned_transducer_stateless7/exp
|
||||||
|
done
|
||||||
|
|
||||||
|
rm pruned_transducer_stateless7/exp/*.pt
|
||||||
|
fi
|
151
.github/scripts/run-librispeech-pruned-transducer-stateless7-ctc-2022-12-01.sh
vendored
Executable file
@ -0,0 +1,151 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
|
log() {
|
||||||
|
# This function is from espnet
|
||||||
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
|
||||||
|
}
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
repo_url=https://huggingface.co/Zengwei/icefall-asr-librispeech-pruned-transducer-stateless7-ctc-2022-12-01
|
||||||
|
|
||||||
|
log "Downloading pre-trained model from $repo_url"
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
|
||||||
|
repo=$(basename $repo_url)
|
||||||
|
|
||||||
|
log "Display test files"
|
||||||
|
tree $repo/
|
||||||
|
soxi $repo/test_wavs/*.wav
|
||||||
|
ls -lh $repo/test_wavs/*.wav
|
||||||
|
|
||||||
|
pushd $repo/exp
|
||||||
|
git lfs pull --include "data/lang_bpe_500/HLG.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/L.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/LG.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/Linv.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/bpe.model"
|
||||||
|
git lfs pull --include "data/lm/G_4_gram.pt"
|
||||||
|
git lfs pull --include "exp/cpu_jit.pt"
|
||||||
|
git lfs pull --include "exp/pretrained.pt"
|
||||||
|
ln -s pretrained.pt epoch-99.pt
|
||||||
|
ls -lh *.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
log "Export to torchscript model"
|
||||||
|
./pruned_transducer_stateless7_ctc/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--use-averaged-model false \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--jit 1
|
||||||
|
|
||||||
|
ls -lh $repo/exp/*.pt
|
||||||
|
|
||||||
|
log "Decode with models exported by torch.jit.script()"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless7_ctc/jit_pretrained.py \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--nn-model-filename $repo/exp/cpu_jit.pt \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
|
||||||
|
for m in ctc-decoding 1best; do
|
||||||
|
./pruned_transducer_stateless7_ctc/jit_pretrained_ctc.py \
|
||||||
|
--model-filename $repo/exp/cpu_jit.pt \
|
||||||
|
--words-file $repo/data/lang_bpe_500/words.txt \
|
||||||
|
--HLG $repo/data/lang_bpe_500/HLG.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--G $repo/data/lm/G_4_gram.pt \
|
||||||
|
--method $m \
|
||||||
|
--sample-rate 16000 \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
for sym in 1 2 3; do
|
||||||
|
log "Greedy search with --max-sym-per-frame $sym"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless7_ctc/pretrained.py \
|
||||||
|
--method greedy_search \
|
||||||
|
--max-sym-per-frame $sym \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
for method in modified_beam_search beam_search fast_beam_search; do
|
||||||
|
log "$method"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless7_ctc/pretrained.py \
|
||||||
|
--method $method \
|
||||||
|
--beam-size 4 \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
for m in ctc-decoding 1best; do
|
||||||
|
./pruned_transducer_stateless7_ctc/pretrained_ctc.py \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--words-file $repo/data/lang_bpe_500/words.txt \
|
||||||
|
--HLG $repo/data/lang_bpe_500/HLG.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--G $repo/data/lm/G_4_gram.pt \
|
||||||
|
--method $m \
|
||||||
|
--sample-rate 16000 \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
||||||
|
echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
|
||||||
|
if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode" ]]; then
|
||||||
|
mkdir -p pruned_transducer_stateless7_ctc/exp
|
||||||
|
ln -s $PWD/$repo/exp/pretrained.pt pruned_transducer_stateless7_ctc/exp/epoch-999.pt
|
||||||
|
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
||||||
|
|
||||||
|
ls -lh data
|
||||||
|
ls -lh pruned_transducer_stateless7_ctc/exp
|
||||||
|
|
||||||
|
log "Decoding test-clean and test-other"
|
||||||
|
|
||||||
|
# use a small value for decoding with CPU
|
||||||
|
max_duration=100
|
||||||
|
|
||||||
|
for method in greedy_search fast_beam_search modified_beam_search; do
|
||||||
|
log "Decoding with $method"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless7_ctc/decode.py \
|
||||||
|
--decoding-method $method \
|
||||||
|
--epoch 999 \
|
||||||
|
--avg 1 \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--max-duration $max_duration \
|
||||||
|
--exp-dir pruned_transducer_stateless7_ctc/exp
|
||||||
|
done
|
||||||
|
|
||||||
|
for m in ctc-decoding 1best; do
|
||||||
|
./pruned_transducer_stateless7_ctc/ctc_decode.py \
|
||||||
|
--epoch 999 \
|
||||||
|
--avg 1 \
|
||||||
|
--exp-dir ./pruned_transducer_stateless7_ctc/exp \
|
||||||
|
--max-duration $max_duration \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--decoding-method $m \
|
||||||
|
--hlg-scale 0.6 \
|
||||||
|
--lm-dir data/lm
|
||||||
|
done
|
||||||
|
|
||||||
|
rm pruned_transducer_stateless7_ctc/exp/*.pt
|
||||||
|
fi
|
116
.github/scripts/run-librispeech-pruned-transducer-stateless8-2022-11-14.sh
vendored
Executable file
@ -0,0 +1,116 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
|
log() {
|
||||||
|
# This function is from espnet
|
||||||
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
|
||||||
|
}
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
repo_url=https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless8-2022-11-14
|
||||||
|
|
||||||
|
log "Downloading pre-trained model from $repo_url"
|
||||||
|
git lfs install
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
|
||||||
|
repo=$(basename $repo_url)
|
||||||
|
|
||||||
|
log "Display test files"
|
||||||
|
tree $repo/
|
||||||
|
soxi $repo/test_wavs/*.wav
|
||||||
|
ls -lh $repo/test_wavs/*.wav
|
||||||
|
|
||||||
|
pushd $repo/exp
|
||||||
|
git lfs pull --include "data/lang_bpe_500/bpe.model"
|
||||||
|
git lfs pull --include "exp/cpu_jit.pt"
|
||||||
|
git lfs pull --include "exp/pretrained.pt"
|
||||||
|
ln -s pretrained.pt epoch-99.pt
|
||||||
|
ls -lh *.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
log "Decode with models exported by torch.jit.script()"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless8/jit_pretrained.py \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--nn-model-filename $repo/exp/cpu_jit.pt \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
|
||||||
|
log "Export to torchscript model"
|
||||||
|
./pruned_transducer_stateless8/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--use-averaged-model false \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--jit 1
|
||||||
|
|
||||||
|
ls -lh $repo/exp/*.pt
|
||||||
|
|
||||||
|
log "Decode with models exported by torch.jit.script()"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless8/jit_pretrained.py \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--nn-model-filename $repo/exp/cpu_jit.pt \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
|
||||||
|
for sym in 1 2 3; do
|
||||||
|
log "Greedy search with --max-sym-per-frame $sym"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless8/pretrained.py \
|
||||||
|
--method greedy_search \
|
||||||
|
--max-sym-per-frame $sym \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
for method in modified_beam_search beam_search fast_beam_search; do
|
||||||
|
log "$method"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless8/pretrained.py \
|
||||||
|
--method $method \
|
||||||
|
--beam-size 4 \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
||||||
|
echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
|
||||||
|
if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode" ]]; then
|
||||||
|
mkdir -p pruned_transducer_stateless8/exp
|
||||||
|
ln -s $PWD/$repo/exp/pretrained.pt pruned_transducer_stateless8/exp/epoch-999.pt
|
||||||
|
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
||||||
|
|
||||||
|
ls -lh data
|
||||||
|
ls -lh pruned_transducer_stateless8/exp
|
||||||
|
|
||||||
|
log "Decoding test-clean and test-other"
|
||||||
|
|
||||||
|
# use a small value for decoding with CPU
|
||||||
|
max_duration=100
|
||||||
|
|
||||||
|
for method in greedy_search fast_beam_search modified_beam_search; do
|
||||||
|
log "Decoding with $method"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless8/decode.py \
|
||||||
|
--decoding-method $method \
|
||||||
|
--epoch 999 \
|
||||||
|
--avg 1 \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--max-duration $max_duration \
|
||||||
|
--exp-dir pruned_transducer_stateless8/exp
|
||||||
|
done
|
||||||
|
|
||||||
|
rm pruned_transducer_stateless8/exp/*.pt
|
||||||
|
fi
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
103
.github/scripts/run-librispeech-zipformer-mmi-2022-12-08.sh
vendored
Executable file
@ -0,0 +1,103 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
|
log() {
|
||||||
|
# This function is from espnet
|
||||||
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
|
||||||
|
}
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
repo_url=https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-mmi-2022-12-08
|
||||||
|
|
||||||
|
log "Downloading pre-trained model from $repo_url"
|
||||||
|
GIT_LFS_SKIP_SMUDGE=1 git clone $repo_url
|
||||||
|
repo=$(basename $repo_url)
|
||||||
|
|
||||||
|
log "Display test files"
|
||||||
|
tree $repo/
|
||||||
|
soxi $repo/test_wavs/*.wav
|
||||||
|
ls -lh $repo/test_wavs/*.wav
|
||||||
|
|
||||||
|
pushd $repo/exp
|
||||||
|
git lfs pull --include "data/lang_bpe_500/3gram.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/4gram.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/L.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/LG.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/Linv.pt"
|
||||||
|
git lfs pull --include "data/lang_bpe_500/bpe.model"
|
||||||
|
git lfs pull --include "exp/cpu_jit.pt"
|
||||||
|
git lfs pull --include "exp/pretrained.pt"
|
||||||
|
ln -s pretrained.pt epoch-99.pt
|
||||||
|
ls -lh *.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
log "Export to torchscript model"
|
||||||
|
./zipformer_mmi/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--use-averaged-model false \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--jit 1
|
||||||
|
|
||||||
|
ls -lh $repo/exp/*.pt
|
||||||
|
|
||||||
|
log "Decode with models exported by torch.jit.script()"
|
||||||
|
|
||||||
|
./zipformer_mmi/jit_pretrained.py \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
--nn-model-filename $repo/exp/cpu_jit.pt \
|
||||||
|
--lang-dir $repo/data/lang_bpe_500 \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
|
||||||
|
for method in 1best nbest nbest-rescoring-LG nbest-rescoring-3-gram nbest-rescoring-4-gram; do
|
||||||
|
log "$method"
|
||||||
|
|
||||||
|
./zipformer_mmi/pretrained.py \
|
||||||
|
--method $method \
|
||||||
|
--checkpoint $repo/exp/pretrained.pt \
|
||||||
|
--lang-dir $repo/data/lang_bpe_500 \
|
||||||
|
--bpe-model $repo/data/lang_bpe_500/bpe.model \
|
||||||
|
$repo/test_wavs/1089-134686-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0001.wav \
|
||||||
|
$repo/test_wavs/1221-135766-0002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
|
||||||
|
echo "GITHUB_EVENT_NAME: ${GITHUB_EVENT_NAME}"
|
||||||
|
echo "GITHUB_EVENT_LABEL_NAME: ${GITHUB_EVENT_LABEL_NAME}"
|
||||||
|
if [[ x"${GITHUB_EVENT_NAME}" == x"schedule" || x"${GITHUB_EVENT_LABEL_NAME}" == x"run-decode" ]]; then
|
||||||
|
mkdir -p zipformer_mmi/exp
|
||||||
|
ln -s $PWD/$repo/exp/pretrained.pt zipformer_mmi/exp/epoch-999.pt
|
||||||
|
ln -s $PWD/$repo/data/lang_bpe_500 data/
|
||||||
|
|
||||||
|
ls -lh data
|
||||||
|
ls -lh zipformer_mmi/exp
|
||||||
|
|
||||||
|
log "Decoding test-clean and test-other"
|
||||||
|
|
||||||
|
# use a small value for decoding with CPU
|
||||||
|
max_duration=100
|
||||||
|
|
||||||
|
for method in 1best nbest nbest-rescoring-LG nbest-rescoring-3-gram nbest-rescoring-4-gram; do
|
||||||
|
log "Decoding with $method"
|
||||||
|
|
||||||
|
./zipformer_mmi/decode.py \
|
||||||
|
--decoding-method $method \
|
||||||
|
--epoch 999 \
|
||||||
|
--avg 1 \
|
||||||
|
--use-averaged-model 0 \
|
||||||
|
--nbest-scale 1.2 \
|
||||||
|
--hp-scale 1.0 \
|
||||||
|
--max-duration $max_duration \
|
||||||
|
--lang-dir $repo/data/lang_bpe_500 \
|
||||||
|
--exp-dir zipformer_mmi/exp
|
||||||
|
done
|
||||||
|
|
||||||
|
rm zipformer_mmi/exp/*.pt
|
||||||
|
fi
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
@ -10,7 +12,6 @@ cd egs/librispeech/ASR
|
|||||||
|
|
||||||
repo_url=https://github.com/csukuangfj/icefall-asr-conformer-ctc-bpe-500
|
repo_url=https://github.com/csukuangfj/icefall-asr-conformer-ctc-bpe-500
|
||||||
git lfs install
|
git lfs install
|
||||||
git clone $repo
|
|
||||||
|
|
||||||
log "Downloading pre-trained model from $repo_url"
|
log "Downloading pre-trained model from $repo_url"
|
||||||
git clone $repo_url
|
git clone $repo_url
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
@ -1,5 +1,7 @@
|
|||||||
#!/usr/bin/env bash
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
log() {
|
log() {
|
||||||
# This function is from espnet
|
# This function is from espnet
|
||||||
local fname=${BASH_SOURCE[1]##*/}
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
124
.github/scripts/run-wenetspeech-pruned-transducer-stateless2.sh
vendored
Executable file
@ -0,0 +1,124 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
set -e
|
||||||
|
|
||||||
|
log() {
|
||||||
|
# This function is from espnet
|
||||||
|
local fname=${BASH_SOURCE[1]##*/}
|
||||||
|
echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
|
||||||
|
}
|
||||||
|
|
||||||
|
cd egs/wenetspeech/ASR
|
||||||
|
|
||||||
|
repo_url=https://huggingface.co/luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2
|
||||||
|
|
||||||
|
log "Downloading pre-trained model from $repo_url"
|
||||||
|
git lfs install
|
||||||
|
git clone $repo_url
|
||||||
|
repo=$(basename $repo_url)
|
||||||
|
|
||||||
|
|
||||||
|
log "Display test files"
|
||||||
|
tree $repo/
|
||||||
|
soxi $repo/test_wavs/*.wav
|
||||||
|
ls -lh $repo/test_wavs/*.wav
|
||||||
|
|
||||||
|
pushd $repo/exp
|
||||||
|
ln -s pretrained_epoch_10_avg_2.pt pretrained.pt
|
||||||
|
ln -s pretrained_epoch_10_avg_2.pt epoch-99.pt
|
||||||
|
popd
|
||||||
|
|
||||||
|
log "Test exporting to ONNX format"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--lang-dir $repo/data/lang_char \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--onnx 1
|
||||||
|
|
||||||
|
log "Export to torchscript model"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--lang-dir $repo/data/lang_char \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--jit 1
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/export.py \
|
||||||
|
--exp-dir $repo/exp \
|
||||||
|
--lang-dir $repo/data/lang_char \
|
||||||
|
--epoch 99 \
|
||||||
|
--avg 1 \
|
||||||
|
--jit-trace 1
|
||||||
|
|
||||||
|
ls -lh $repo/exp/*.onnx
|
||||||
|
ls -lh $repo/exp/*.pt
|
||||||
|
|
||||||
|
log "Decode with ONNX models"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/onnx_check.py \
|
||||||
|
--jit-filename $repo/exp/cpu_jit.pt \
|
||||||
|
--onnx-encoder-filename $repo/exp/encoder.onnx \
|
||||||
|
--onnx-decoder-filename $repo/exp/decoder.onnx \
|
||||||
|
--onnx-joiner-filename $repo/exp/joiner.onnx \
|
||||||
|
--onnx-joiner-encoder-proj-filename $repo/exp/joiner_encoder_proj.onnx \
|
||||||
|
--onnx-joiner-decoder-proj-filename $repo/exp/joiner_decoder_proj.onnx
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/onnx_pretrained.py \
|
||||||
|
--tokens $repo/data/lang_char/tokens.txt \
|
||||||
|
--encoder-model-filename $repo/exp/encoder.onnx \
|
||||||
|
--decoder-model-filename $repo/exp/decoder.onnx \
|
||||||
|
--joiner-model-filename $repo/exp/joiner.onnx \
|
||||||
|
--joiner-encoder-proj-model-filename $repo/exp/joiner_encoder_proj.onnx \
|
||||||
|
--joiner-decoder-proj-model-filename $repo/exp/joiner_decoder_proj.onnx \
|
||||||
|
$repo/test_wavs/DEV_T0000000000.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000001.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000002.wav
|
||||||
|
|
||||||
|
log "Decode with models exported by torch.jit.trace()"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/jit_pretrained.py \
|
||||||
|
--tokens $repo/data/lang_char/tokens.txt \
|
||||||
|
--encoder-model-filename $repo/exp/encoder_jit_trace.pt \
|
||||||
|
--decoder-model-filename $repo/exp/decoder_jit_trace.pt \
|
||||||
|
--joiner-model-filename $repo/exp/joiner_jit_trace.pt \
|
||||||
|
$repo/test_wavs/DEV_T0000000000.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000001.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000002.wav
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/jit_pretrained.py \
|
||||||
|
--tokens $repo/data/lang_char/tokens.txt \
|
||||||
|
--encoder-model-filename $repo/exp/encoder_jit_script.pt \
|
||||||
|
--decoder-model-filename $repo/exp/decoder_jit_script.pt \
|
||||||
|
--joiner-model-filename $repo/exp/joiner_jit_script.pt \
|
||||||
|
$repo/test_wavs/DEV_T0000000000.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000001.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000002.wav
|
||||||
|
|
||||||
|
for sym in 1 2 3; do
|
||||||
|
log "Greedy search with --max-sym-per-frame $sym"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/pretrained.py \
|
||||||
|
--checkpoint $repo/exp/epoch-99.pt \
|
||||||
|
--lang-dir $repo/data/lang_char \
|
||||||
|
--decoding-method greedy_search \
|
||||||
|
--max-sym-per-frame $sym \
|
||||||
|
$repo/test_wavs/DEV_T0000000000.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000001.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000002.wav
|
||||||
|
done
|
||||||
|
|
||||||
|
for method in modified_beam_search beam_search fast_beam_search; do
|
||||||
|
log "$method"
|
||||||
|
|
||||||
|
./pruned_transducer_stateless2/pretrained.py \
|
||||||
|
--decoding-method $method \
|
||||||
|
--beam-size 4 \
|
||||||
|
--checkpoint $repo/exp/epoch-99.pt \
|
||||||
|
--lang-dir $repo/data/lang_char \
|
||||||
|
$repo/test_wavs/DEV_T0000000000.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000001.wav \
|
||||||
|
$repo/test_wavs/DEV_T0000000002.wav
|
||||||
|
done
|
4
.github/workflows/build-doc.yml
vendored
@ -26,6 +26,10 @@ on:
|
|||||||
pull_request:
|
pull_request:
|
||||||
types: [labeled]
|
types: [labeled]
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: build_doc-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
build-doc:
|
build-doc:
|
||||||
if: github.event.label.name == 'doc' || github.event_name == 'push'
|
if: github.event.label.name == 'doc' || github.event_name == 'push'
|
||||||
|
4
.github/workflows/run-aishell-2022-06-20.yml
vendored
@ -34,6 +34,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_aishell_2022_06_20-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_aishell_2022_06_20:
|
run_aishell_2022_06_20:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -33,6 +33,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_gigaspeech_2022_05_13-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_gigaspeech_2022_05_13:
|
run_gigaspeech_2022_05_13:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -33,6 +33,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_2022_03_12-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_librispeech_2022_03_12:
|
run_librispeech_2022_03_12:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -33,6 +33,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_2022_04_29-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_librispeech_2022_04_29:
|
run_librispeech_2022_04_29:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -33,6 +33,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_2022_05_13-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_librispeech_2022_05_13:
|
run_librispeech_2022_05_13:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
159
.github/workflows/run-librispeech-2022-11-11-stateless7.yml
vendored
Normal file
@ -0,0 +1,159 @@
|
|||||||
|
# Copyright 2022 Fangjun Kuang (csukuangfj@gmail.com)
|
||||||
|
|
||||||
|
# See ../../LICENSE for clarification regarding multiple authors
|
||||||
|
#
|
||||||
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
# you may not use this file except in compliance with the License.
|
||||||
|
# You may obtain a copy of the License at
|
||||||
|
#
|
||||||
|
# http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
#
|
||||||
|
# Unless required by applicable law or agreed to in writing, software
|
||||||
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
# See the License for the specific language governing permissions and
|
||||||
|
# limitations under the License.
|
||||||
|
|
||||||
|
name: run-librispeech-2022-11-11-stateless7
|
||||||
|
# zipformer
|
||||||
|
|
||||||
|
on:
|
||||||
|
push:
|
||||||
|
branches:
|
||||||
|
- master
|
||||||
|
pull_request:
|
||||||
|
types: [labeled]
|
||||||
|
|
||||||
|
schedule:
|
||||||
|
# minute (0-59)
|
||||||
|
# hour (0-23)
|
||||||
|
# day of the month (1-31)
|
||||||
|
# month (1-12)
|
||||||
|
# day of the week (0-6)
|
||||||
|
# nightly build at 15:50 UTC time every day
|
||||||
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_2022_11_11_zipformer-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
|
jobs:
|
||||||
|
run_librispeech_2022_11_11_zipformer:
|
||||||
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
runs-on: ${{ matrix.os }}
|
||||||
|
strategy:
|
||||||
|
matrix:
|
||||||
|
os: [ubuntu-latest]
|
||||||
|
python-version: [3.8]
|
||||||
|
|
||||||
|
fail-fast: false
|
||||||
|
|
||||||
|
steps:
|
||||||
|
- uses: actions/checkout@v2
|
||||||
|
with:
|
||||||
|
fetch-depth: 0
|
||||||
|
|
||||||
|
- name: Setup Python ${{ matrix.python-version }}
|
||||||
|
uses: actions/setup-python@v2
|
||||||
|
with:
|
||||||
|
python-version: ${{ matrix.python-version }}
|
||||||
|
cache: 'pip'
|
||||||
|
cache-dependency-path: '**/requirements-ci.txt'
|
||||||
|
|
||||||
|
- name: Install Python dependencies
|
||||||
|
run: |
|
||||||
|
grep -v '^#' ./requirements-ci.txt | xargs -n 1 -L 1 pip install
|
||||||
|
pip uninstall -y protobuf
|
||||||
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
- name: Cache kaldifeat
|
||||||
|
id: my-cache
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/kaldifeat
|
||||||
|
key: cache-tmp-${{ matrix.python-version }}-2022-09-25
|
||||||
|
|
||||||
|
- name: Install kaldifeat
|
||||||
|
if: steps.my-cache.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/install-kaldifeat.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other datasets
|
||||||
|
id: libri-test-clean-and-test-other-data
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/download
|
||||||
|
key: cache-libri-test-clean-and-test-other
|
||||||
|
|
||||||
|
- name: Download LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-data.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/download-librispeech-test-clean-and-test-other-dataset.sh
|
||||||
|
|
||||||
|
- name: Prepare manifests for LibriSpeech test-clean and test-other
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/prepare-librispeech-test-clean-and-test-other-manifests.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other fbank features
|
||||||
|
id: libri-test-clean-and-test-other-fbank
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/fbank-libri
|
||||||
|
key: cache-libri-fbank-test-clean-and-test-other-v2
|
||||||
|
|
||||||
|
- name: Compute fbank for LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-fbank.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/compute-fbank-librispeech-test-clean-and-test-other.sh
|
||||||
|
|
||||||
|
- name: Inference with pre-trained model
|
||||||
|
shell: bash
|
||||||
|
env:
|
||||||
|
GITHUB_EVENT_NAME: ${{ github.event_name }}
|
||||||
|
GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name }}
|
||||||
|
run: |
|
||||||
|
mkdir -p egs/librispeech/ASR/data
|
||||||
|
ln -sfv ~/tmp/fbank-libri egs/librispeech/ASR/data/fbank
|
||||||
|
ls -lh egs/librispeech/ASR/data/*
|
||||||
|
|
||||||
|
sudo apt-get -qq install git-lfs tree sox
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
||||||
|
|
||||||
|
.github/scripts/run-librispeech-pruned-transducer-stateless7-2022-11-11.sh
|
||||||
|
|
||||||
|
- name: Display decoding results for librispeech pruned_transducer_stateless7
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
cd egs/librispeech/ASR/
|
||||||
|
tree ./pruned_transducer_stateless7/exp
|
||||||
|
|
||||||
|
cd pruned_transducer_stateless7
|
||||||
|
echo "results for pruned_transducer_stateless7"
|
||||||
|
echo "===greedy search==="
|
||||||
|
find exp/greedy_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/greedy_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===fast_beam_search==="
|
||||||
|
find exp/fast_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/fast_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===modified beam search==="
|
||||||
|
find exp/modified_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/modified_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
- name: Upload decoding results for librispeech pruned_transducer_stateless7
|
||||||
|
uses: actions/upload-artifact@v2
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
with:
|
||||||
|
name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-pruned_transducer_stateless7-2022-11-11
|
||||||
|
path: egs/librispeech/ASR/pruned_transducer_stateless7/exp/
|
159
.github/workflows/run-librispeech-2022-11-14-stateless8.yml
vendored
Normal file
@ -0,0 +1,159 @@
|
|||||||
|
# Copyright 2022 Fangjun Kuang (csukuangfj@gmail.com)
|
||||||
|
|
||||||
|
# See ../../LICENSE for clarification regarding multiple authors
|
||||||
|
#
|
||||||
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
# you may not use this file except in compliance with the License.
|
||||||
|
# You may obtain a copy of the License at
|
||||||
|
#
|
||||||
|
# http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
#
|
||||||
|
# Unless required by applicable law or agreed to in writing, software
|
||||||
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
# See the License for the specific language governing permissions and
|
||||||
|
# limitations under the License.
|
||||||
|
|
||||||
|
name: run-librispeech-2022-11-14-stateless8
|
||||||
|
# zipformer
|
||||||
|
|
||||||
|
on:
|
||||||
|
push:
|
||||||
|
branches:
|
||||||
|
- master
|
||||||
|
pull_request:
|
||||||
|
types: [labeled]
|
||||||
|
|
||||||
|
schedule:
|
||||||
|
# minute (0-59)
|
||||||
|
# hour (0-23)
|
||||||
|
# day of the month (1-31)
|
||||||
|
# month (1-12)
|
||||||
|
# day of the week (0-6)
|
||||||
|
# nightly build at 15:50 UTC time every day
|
||||||
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_2022_11_14_zipformer_stateless8-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
|
jobs:
|
||||||
|
run_librispeech_2022_11_14_zipformer_stateless8:
|
||||||
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
runs-on: ${{ matrix.os }}
|
||||||
|
strategy:
|
||||||
|
matrix:
|
||||||
|
os: [ubuntu-latest]
|
||||||
|
python-version: [3.8]
|
||||||
|
|
||||||
|
fail-fast: false
|
||||||
|
|
||||||
|
steps:
|
||||||
|
- uses: actions/checkout@v2
|
||||||
|
with:
|
||||||
|
fetch-depth: 0
|
||||||
|
|
||||||
|
- name: Setup Python ${{ matrix.python-version }}
|
||||||
|
uses: actions/setup-python@v2
|
||||||
|
with:
|
||||||
|
python-version: ${{ matrix.python-version }}
|
||||||
|
cache: 'pip'
|
||||||
|
cache-dependency-path: '**/requirements-ci.txt'
|
||||||
|
|
||||||
|
- name: Install Python dependencies
|
||||||
|
run: |
|
||||||
|
grep -v '^#' ./requirements-ci.txt | xargs -n 1 -L 1 pip install
|
||||||
|
pip uninstall -y protobuf
|
||||||
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
- name: Cache kaldifeat
|
||||||
|
id: my-cache
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/kaldifeat
|
||||||
|
key: cache-tmp-${{ matrix.python-version }}-2022-09-25
|
||||||
|
|
||||||
|
- name: Install kaldifeat
|
||||||
|
if: steps.my-cache.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/install-kaldifeat.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other datasets
|
||||||
|
id: libri-test-clean-and-test-other-data
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/download
|
||||||
|
key: cache-libri-test-clean-and-test-other
|
||||||
|
|
||||||
|
- name: Download LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-data.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/download-librispeech-test-clean-and-test-other-dataset.sh
|
||||||
|
|
||||||
|
- name: Prepare manifests for LibriSpeech test-clean and test-other
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/prepare-librispeech-test-clean-and-test-other-manifests.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other fbank features
|
||||||
|
id: libri-test-clean-and-test-other-fbank
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/fbank-libri
|
||||||
|
key: cache-libri-fbank-test-clean-and-test-other-v2
|
||||||
|
|
||||||
|
- name: Compute fbank for LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-fbank.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/compute-fbank-librispeech-test-clean-and-test-other.sh
|
||||||
|
|
||||||
|
- name: Inference with pre-trained model
|
||||||
|
shell: bash
|
||||||
|
env:
|
||||||
|
GITHUB_EVENT_NAME: ${{ github.event_name }}
|
||||||
|
GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name }}
|
||||||
|
run: |
|
||||||
|
mkdir -p egs/librispeech/ASR/data
|
||||||
|
ln -sfv ~/tmp/fbank-libri egs/librispeech/ASR/data/fbank
|
||||||
|
ls -lh egs/librispeech/ASR/data/*
|
||||||
|
|
||||||
|
sudo apt-get -qq install git-lfs tree sox
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
||||||
|
|
||||||
|
.github/scripts/run-librispeech-pruned-transducer-stateless8-2022-11-14.sh
|
||||||
|
|
||||||
|
- name: Display decoding results for librispeech pruned_transducer_stateless8
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
cd egs/librispeech/ASR/
|
||||||
|
tree ./pruned_transducer_stateless8/exp
|
||||||
|
|
||||||
|
cd pruned_transducer_stateless8
|
||||||
|
echo "results for pruned_transducer_stateless8"
|
||||||
|
echo "===greedy search==="
|
||||||
|
find exp/greedy_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/greedy_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===fast_beam_search==="
|
||||||
|
find exp/fast_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/fast_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===modified beam search==="
|
||||||
|
find exp/modified_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/modified_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
- name: Upload decoding results for librispeech pruned_transducer_stateless8
|
||||||
|
uses: actions/upload-artifact@v2
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
with:
|
||||||
|
name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-pruned_transducer_stateless8-2022-11-14
|
||||||
|
path: egs/librispeech/ASR/pruned_transducer_stateless8/exp/
|
163
.github/workflows/run-librispeech-2022-12-01-stateless7-ctc.yml
vendored
Normal file
@ -0,0 +1,163 @@
|
|||||||
|
# Copyright 2022 Fangjun Kuang (csukuangfj@gmail.com)
|
||||||
|
|
||||||
|
# See ../../LICENSE for clarification regarding multiple authors
|
||||||
|
#
|
||||||
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
# you may not use this file except in compliance with the License.
|
||||||
|
# You may obtain a copy of the License at
|
||||||
|
#
|
||||||
|
# http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
#
|
||||||
|
# Unless required by applicable law or agreed to in writing, software
|
||||||
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
# See the License for the specific language governing permissions and
|
||||||
|
# limitations under the License.
|
||||||
|
|
||||||
|
name: run-librispeech-2022-12-01-stateless7-ctc
|
||||||
|
# zipformer
|
||||||
|
|
||||||
|
on:
|
||||||
|
push:
|
||||||
|
branches:
|
||||||
|
- master
|
||||||
|
pull_request:
|
||||||
|
types: [labeled]
|
||||||
|
|
||||||
|
schedule:
|
||||||
|
# minute (0-59)
|
||||||
|
# hour (0-23)
|
||||||
|
# day of the month (1-31)
|
||||||
|
# month (1-12)
|
||||||
|
# day of the week (0-6)
|
||||||
|
# nightly build at 15:50 UTC time every day
|
||||||
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
jobs:
|
||||||
|
run_librispeech_2022_11_11_zipformer:
|
||||||
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
runs-on: ${{ matrix.os }}
|
||||||
|
strategy:
|
||||||
|
matrix:
|
||||||
|
os: [ubuntu-latest]
|
||||||
|
python-version: [3.8]
|
||||||
|
|
||||||
|
fail-fast: false
|
||||||
|
|
||||||
|
steps:
|
||||||
|
- uses: actions/checkout@v2
|
||||||
|
with:
|
||||||
|
fetch-depth: 0
|
||||||
|
|
||||||
|
- name: Setup Python ${{ matrix.python-version }}
|
||||||
|
uses: actions/setup-python@v2
|
||||||
|
with:
|
||||||
|
python-version: ${{ matrix.python-version }}
|
||||||
|
cache: 'pip'
|
||||||
|
cache-dependency-path: '**/requirements-ci.txt'
|
||||||
|
|
||||||
|
- name: Install Python dependencies
|
||||||
|
run: |
|
||||||
|
grep -v '^#' ./requirements-ci.txt | xargs -n 1 -L 1 pip install
|
||||||
|
pip uninstall -y protobuf
|
||||||
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
- name: Cache kaldifeat
|
||||||
|
id: my-cache
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/kaldifeat
|
||||||
|
key: cache-tmp-${{ matrix.python-version }}-2022-09-25
|
||||||
|
|
||||||
|
- name: Install kaldifeat
|
||||||
|
if: steps.my-cache.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/install-kaldifeat.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other datasets
|
||||||
|
id: libri-test-clean-and-test-other-data
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/download
|
||||||
|
key: cache-libri-test-clean-and-test-other
|
||||||
|
|
||||||
|
- name: Download LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-data.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/download-librispeech-test-clean-and-test-other-dataset.sh
|
||||||
|
|
||||||
|
- name: Prepare manifests for LibriSpeech test-clean and test-other
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/prepare-librispeech-test-clean-and-test-other-manifests.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other fbank features
|
||||||
|
id: libri-test-clean-and-test-other-fbank
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/fbank-libri
|
||||||
|
key: cache-libri-fbank-test-clean-and-test-other-v2
|
||||||
|
|
||||||
|
- name: Compute fbank for LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-fbank.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/compute-fbank-librispeech-test-clean-and-test-other.sh
|
||||||
|
|
||||||
|
- name: Inference with pre-trained model
|
||||||
|
shell: bash
|
||||||
|
env:
|
||||||
|
GITHUB_EVENT_NAME: ${{ github.event_name }}
|
||||||
|
GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name }}
|
||||||
|
run: |
|
||||||
|
mkdir -p egs/librispeech/ASR/data
|
||||||
|
ln -sfv ~/tmp/fbank-libri egs/librispeech/ASR/data/fbank
|
||||||
|
ls -lh egs/librispeech/ASR/data/*
|
||||||
|
|
||||||
|
sudo apt-get -qq install git-lfs tree sox
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
||||||
|
|
||||||
|
.github/scripts/run-librispeech-pruned-transducer-stateless7-ctc-2022-12-01.sh
|
||||||
|
|
||||||
|
- name: Display decoding results for librispeech pruned_transducer_stateless7_ctc
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
cd egs/librispeech/ASR/
|
||||||
|
tree ./pruned_transducer_stateless7_ctc/exp
|
||||||
|
|
||||||
|
cd pruned_transducer_stateless7_ctc
|
||||||
|
echo "results for pruned_transducer_stateless7_ctc"
|
||||||
|
echo "===greedy search==="
|
||||||
|
find exp/greedy_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/greedy_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===fast_beam_search==="
|
||||||
|
find exp/fast_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/fast_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===modified beam search==="
|
||||||
|
find exp/modified_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/modified_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===ctc decoding==="
|
||||||
|
find exp/ctc-decoding -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/ctc-decoding -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===1best==="
|
||||||
|
find exp/1best -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/1best -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
- name: Upload decoding results for librispeech pruned_transducer_stateless7_ctc
|
||||||
|
uses: actions/upload-artifact@v2
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
with:
|
||||||
|
name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-pruned_transducer_stateless7-ctc-2022-12-01
|
||||||
|
path: egs/librispeech/ASR/pruned_transducer_stateless7_ctc/exp/
|
167
.github/workflows/run-librispeech-2022-12-08-zipformer-mmi.yml
vendored
Normal file
@ -0,0 +1,167 @@
|
|||||||
|
# Copyright 2022 Zengwei Yao
|
||||||
|
|
||||||
|
# See ../../LICENSE for clarification regarding multiple authors
|
||||||
|
#
|
||||||
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
# you may not use this file except in compliance with the License.
|
||||||
|
# You may obtain a copy of the License at
|
||||||
|
#
|
||||||
|
# http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
#
|
||||||
|
# Unless required by applicable law or agreed to in writing, software
|
||||||
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
# See the License for the specific language governing permissions and
|
||||||
|
# limitations under the License.
|
||||||
|
|
||||||
|
name: run-librispeech-2022-12-08-zipformer-mmi
|
||||||
|
# zipformer
|
||||||
|
|
||||||
|
on:
|
||||||
|
push:
|
||||||
|
branches:
|
||||||
|
- master
|
||||||
|
pull_request:
|
||||||
|
types: [labeled]
|
||||||
|
|
||||||
|
schedule:
|
||||||
|
# minute (0-59)
|
||||||
|
# hour (0-23)
|
||||||
|
# day of the month (1-31)
|
||||||
|
# month (1-12)
|
||||||
|
# day of the week (0-6)
|
||||||
|
# nightly build at 15:50 UTC time every day
|
||||||
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_2022_12_08_zipformer-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
|
jobs:
|
||||||
|
run_librispeech_2022_12_08_zipformer:
|
||||||
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
runs-on: ${{ matrix.os }}
|
||||||
|
strategy:
|
||||||
|
matrix:
|
||||||
|
os: [ubuntu-latest]
|
||||||
|
python-version: [3.8]
|
||||||
|
|
||||||
|
fail-fast: false
|
||||||
|
|
||||||
|
steps:
|
||||||
|
- uses: actions/checkout@v2
|
||||||
|
with:
|
||||||
|
fetch-depth: 0
|
||||||
|
|
||||||
|
- name: Setup Python ${{ matrix.python-version }}
|
||||||
|
uses: actions/setup-python@v2
|
||||||
|
with:
|
||||||
|
python-version: ${{ matrix.python-version }}
|
||||||
|
cache: 'pip'
|
||||||
|
cache-dependency-path: '**/requirements-ci.txt'
|
||||||
|
|
||||||
|
- name: Install Python dependencies
|
||||||
|
run: |
|
||||||
|
grep -v '^#' ./requirements-ci.txt | xargs -n 1 -L 1 pip install
|
||||||
|
pip uninstall -y protobuf
|
||||||
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
- name: Cache kaldifeat
|
||||||
|
id: my-cache
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/kaldifeat
|
||||||
|
key: cache-tmp-${{ matrix.python-version }}-2022-09-25
|
||||||
|
|
||||||
|
- name: Install kaldifeat
|
||||||
|
if: steps.my-cache.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/install-kaldifeat.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other datasets
|
||||||
|
id: libri-test-clean-and-test-other-data
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/download
|
||||||
|
key: cache-libri-test-clean-and-test-other
|
||||||
|
|
||||||
|
- name: Download LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-data.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/download-librispeech-test-clean-and-test-other-dataset.sh
|
||||||
|
|
||||||
|
- name: Prepare manifests for LibriSpeech test-clean and test-other
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/prepare-librispeech-test-clean-and-test-other-manifests.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other fbank features
|
||||||
|
id: libri-test-clean-and-test-other-fbank
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/fbank-libri
|
||||||
|
key: cache-libri-fbank-test-clean-and-test-other-v2
|
||||||
|
|
||||||
|
- name: Compute fbank for LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-fbank.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/compute-fbank-librispeech-test-clean-and-test-other.sh
|
||||||
|
|
||||||
|
- name: Inference with pre-trained model
|
||||||
|
shell: bash
|
||||||
|
env:
|
||||||
|
GITHUB_EVENT_NAME: ${{ github.event_name }}
|
||||||
|
GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name }}
|
||||||
|
run: |
|
||||||
|
mkdir -p egs/librispeech/ASR/data
|
||||||
|
ln -sfv ~/tmp/fbank-libri egs/librispeech/ASR/data/fbank
|
||||||
|
ls -lh egs/librispeech/ASR/data/*
|
||||||
|
|
||||||
|
sudo apt-get -qq install git-lfs tree sox
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
||||||
|
|
||||||
|
.github/scripts/run-librispeech-zipformer-mmi-2022-12-08.sh
|
||||||
|
|
||||||
|
- name: Display decoding results for librispeech zipformer-mmi
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
cd egs/librispeech/ASR/
|
||||||
|
tree ./zipformer-mmi/exp
|
||||||
|
|
||||||
|
cd zipformer-mmi
|
||||||
|
echo "results for zipformer-mmi"
|
||||||
|
echo "===1best==="
|
||||||
|
find exp/1best -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/1best -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===nbest==="
|
||||||
|
find exp/nbest -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/nbest -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===nbest-rescoring-LG==="
|
||||||
|
find exp/nbest-rescoring-LG -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/nbest-rescoring-LG -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===nbest-rescoring-3-gram==="
|
||||||
|
find exp/nbest-rescoring-3-gram -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/nbest-rescoring-3-gram -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===nbest-rescoring-4-gram==="
|
||||||
|
find exp/nbest-rescoring-4-gram -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/nbest-rescoring-4-gram -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
- name: Upload decoding results for librispeech zipformer-mmi
|
||||||
|
uses: actions/upload-artifact@v2
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
with:
|
||||||
|
name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-zipformer_mmi-2022-12-08
|
||||||
|
path: egs/librispeech/ASR/zipformer_mmi/exp/
|
155
.github/workflows/run-librispeech-conformer-ctc3-2022-11-28.yml
vendored
Normal file
@ -0,0 +1,155 @@
|
|||||||
|
# Copyright 2022 Fangjun Kuang (csukuangfj@gmail.com)
|
||||||
|
|
||||||
|
# See ../../LICENSE for clarification regarding multiple authors
|
||||||
|
#
|
||||||
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
# you may not use this file except in compliance with the License.
|
||||||
|
# You may obtain a copy of the License at
|
||||||
|
#
|
||||||
|
# http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
#
|
||||||
|
# Unless required by applicable law or agreed to in writing, software
|
||||||
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
# See the License for the specific language governing permissions and
|
||||||
|
# limitations under the License.
|
||||||
|
|
||||||
|
name: run-librispeech-conformer-ctc3-2022-11-28
|
||||||
|
# zipformer
|
||||||
|
|
||||||
|
on:
|
||||||
|
push:
|
||||||
|
branches:
|
||||||
|
- master
|
||||||
|
pull_request:
|
||||||
|
types: [labeled]
|
||||||
|
|
||||||
|
schedule:
|
||||||
|
# minute (0-59)
|
||||||
|
# hour (0-23)
|
||||||
|
# day of the month (1-31)
|
||||||
|
# month (1-12)
|
||||||
|
# day of the week (0-6)
|
||||||
|
# nightly build at 15:50 UTC time every day
|
||||||
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_2022_11_28_conformer_ctc3-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
|
jobs:
|
||||||
|
run_librispeech_2022_11_28_conformer_ctc3:
|
||||||
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
runs-on: ${{ matrix.os }}
|
||||||
|
strategy:
|
||||||
|
matrix:
|
||||||
|
os: [ubuntu-latest]
|
||||||
|
python-version: [3.8]
|
||||||
|
|
||||||
|
fail-fast: false
|
||||||
|
|
||||||
|
steps:
|
||||||
|
- uses: actions/checkout@v2
|
||||||
|
with:
|
||||||
|
fetch-depth: 0
|
||||||
|
|
||||||
|
- name: Setup Python ${{ matrix.python-version }}
|
||||||
|
uses: actions/setup-python@v2
|
||||||
|
with:
|
||||||
|
python-version: ${{ matrix.python-version }}
|
||||||
|
cache: 'pip'
|
||||||
|
cache-dependency-path: '**/requirements-ci.txt'
|
||||||
|
|
||||||
|
- name: Install Python dependencies
|
||||||
|
run: |
|
||||||
|
grep -v '^#' ./requirements-ci.txt | xargs -n 1 -L 1 pip install
|
||||||
|
pip uninstall -y protobuf
|
||||||
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
- name: Cache kaldifeat
|
||||||
|
id: my-cache
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/kaldifeat
|
||||||
|
key: cache-tmp-${{ matrix.python-version }}-2022-09-25
|
||||||
|
|
||||||
|
- name: Install kaldifeat
|
||||||
|
if: steps.my-cache.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/install-kaldifeat.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other datasets
|
||||||
|
id: libri-test-clean-and-test-other-data
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/download
|
||||||
|
key: cache-libri-test-clean-and-test-other
|
||||||
|
|
||||||
|
- name: Download LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-data.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/download-librispeech-test-clean-and-test-other-dataset.sh
|
||||||
|
|
||||||
|
- name: Prepare manifests for LibriSpeech test-clean and test-other
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/prepare-librispeech-test-clean-and-test-other-manifests.sh
|
||||||
|
|
||||||
|
- name: Cache LibriSpeech test-clean and test-other fbank features
|
||||||
|
id: libri-test-clean-and-test-other-fbank
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/fbank-libri
|
||||||
|
key: cache-libri-fbank-test-clean-and-test-other-v2
|
||||||
|
|
||||||
|
- name: Compute fbank for LibriSpeech test-clean and test-other
|
||||||
|
if: steps.libri-test-clean-and-test-other-fbank.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/compute-fbank-librispeech-test-clean-and-test-other.sh
|
||||||
|
|
||||||
|
- name: Inference with pre-trained model
|
||||||
|
shell: bash
|
||||||
|
env:
|
||||||
|
GITHUB_EVENT_NAME: ${{ github.event_name }}
|
||||||
|
GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name }}
|
||||||
|
run: |
|
||||||
|
mkdir -p egs/librispeech/ASR/data
|
||||||
|
ln -sfv ~/tmp/fbank-libri egs/librispeech/ASR/data/fbank
|
||||||
|
ls -lh egs/librispeech/ASR/data/*
|
||||||
|
|
||||||
|
sudo apt-get -qq install git-lfs tree sox
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
||||||
|
|
||||||
|
.github/scripts/run-librispeech-conformer-ctc3-2022-11-28.sh
|
||||||
|
|
||||||
|
- name: Display decoding results for librispeech conformer_ctc3
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
cd egs/librispeech/ASR/
|
||||||
|
tree ./conformer_ctc3/exp
|
||||||
|
|
||||||
|
cd conformer_ctc3
|
||||||
|
echo "results for conformer_ctc3"
|
||||||
|
echo "===ctc-decoding==="
|
||||||
|
find exp/ctc-decoding -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/ctc-decoding -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
echo "===1best==="
|
||||||
|
find exp/1best -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find exp/1best -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
- name: Upload decoding results for librispeech conformer_ctc3
|
||||||
|
uses: actions/upload-artifact@v2
|
||||||
|
if: github.event_name == 'schedule' || github.event.label.name == 'run-decode'
|
||||||
|
with:
|
||||||
|
name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-conformer_ctc3-2022-11-28
|
||||||
|
path: egs/librispeech/ASR/conformer_ctc3/exp/
|
77
.github/workflows/run-librispeech-conv-emformer-transducer-stateless2-2022-12-05.yml
vendored
Normal file
@ -0,0 +1,77 @@
|
|||||||
|
name: run-librispeech-conv-emformer-transducer-stateless2-2022-12-05
|
||||||
|
|
||||||
|
on:
|
||||||
|
push:
|
||||||
|
branches:
|
||||||
|
- master
|
||||||
|
pull_request:
|
||||||
|
types: [labeled]
|
||||||
|
|
||||||
|
schedule:
|
||||||
|
# minute (0-59)
|
||||||
|
# hour (0-23)
|
||||||
|
# day of the month (1-31)
|
||||||
|
# month (1-12)
|
||||||
|
# day of the week (0-6)
|
||||||
|
# nightly build at 15:50 UTC time every day
|
||||||
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
jobs:
|
||||||
|
run_librispeech_conv_emformer_transducer_stateless2_2022_12_05:
|
||||||
|
if: github.event.label.name == 'ready' || github.event.label.name == 'ncnn' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
runs-on: ${{ matrix.os }}
|
||||||
|
strategy:
|
||||||
|
matrix:
|
||||||
|
os: [ubuntu-latest]
|
||||||
|
python-version: [3.8]
|
||||||
|
|
||||||
|
fail-fast: false
|
||||||
|
|
||||||
|
steps:
|
||||||
|
- uses: actions/checkout@v2
|
||||||
|
with:
|
||||||
|
fetch-depth: 0
|
||||||
|
|
||||||
|
- name: Setup Python ${{ matrix.python-version }}
|
||||||
|
uses: actions/setup-python@v2
|
||||||
|
with:
|
||||||
|
python-version: ${{ matrix.python-version }}
|
||||||
|
cache: 'pip'
|
||||||
|
cache-dependency-path: '**/requirements-ci.txt'
|
||||||
|
|
||||||
|
- name: Install Python dependencies
|
||||||
|
run: |
|
||||||
|
grep -v '^#' ./requirements-ci.txt | grep -v kaldifst | xargs -n 1 -L 1 pip install
|
||||||
|
pip uninstall -y protobuf
|
||||||
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
- name: Cache kaldifeat
|
||||||
|
id: my-cache
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/kaldifeat
|
||||||
|
key: cache-tmp-${{ matrix.python-version }}-2022-09-25
|
||||||
|
|
||||||
|
- name: Install kaldifeat
|
||||||
|
if: steps.my-cache.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/install-kaldifeat.sh
|
||||||
|
|
||||||
|
- name: Inference with pre-trained model
|
||||||
|
shell: bash
|
||||||
|
env:
|
||||||
|
GITHUB_EVENT_NAME: ${{ github.event_name }}
|
||||||
|
GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name }}
|
||||||
|
run: |
|
||||||
|
mkdir -p egs/librispeech/ASR/data
|
||||||
|
ln -sfv ~/tmp/fbank-libri egs/librispeech/ASR/data/fbank
|
||||||
|
ls -lh egs/librispeech/ASR/data/*
|
||||||
|
|
||||||
|
sudo apt-get -qq install git-lfs tree sox
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
||||||
|
|
||||||
|
.github/scripts/run-librispeech-conv-emformer-transducer-stateless2-2022-12-05.sh
|
@ -1,4 +1,4 @@
|
|||||||
name: run-librispeech-lstm-transducer-2022-09-03
|
name: run-librispeech-lstm-transducer2-2022-09-03
|
||||||
|
|
||||||
on:
|
on:
|
||||||
push:
|
push:
|
||||||
@ -16,9 +16,13 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_lstm_transducer_stateless2_2022_09_03-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_librispeech_pruned_transducer_stateless3_2022_05_13:
|
run_librispeech_lstm_transducer_stateless2_2022_09_03:
|
||||||
if: github.event.label.name == 'ncnn' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'LODR' || github.event.label.name == 'shallow-fusion' || github.event.label.name == 'ncnn' || github.event.label.name == 'onnx' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
runs-on: ${{ matrix.os }}
|
runs-on: ${{ matrix.os }}
|
||||||
strategy:
|
strategy:
|
||||||
matrix:
|
matrix:
|
||||||
@ -107,10 +111,10 @@ jobs:
|
|||||||
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
||||||
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
||||||
|
|
||||||
.github/scripts/run-librispeech-lstm-transducer-stateless2-2022-09-03.yml
|
.github/scripts/run-librispeech-lstm-transducer-stateless2-2022-09-03.sh
|
||||||
|
|
||||||
- name: Display decoding results for lstm_transducer_stateless2
|
- name: Display decoding results for lstm_transducer_stateless2
|
||||||
if: github.event_name == 'schedule' || github.event.label.name == 'ncnn'
|
if: github.event_name == 'schedule'
|
||||||
shell: bash
|
shell: bash
|
||||||
run: |
|
run: |
|
||||||
cd egs/librispeech/ASR
|
cd egs/librispeech/ASR
|
||||||
@ -128,9 +132,31 @@ jobs:
|
|||||||
find modified_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
find modified_beam_search -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
find modified_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
find modified_beam_search -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
- name: Display decoding results for lstm_transducer_stateless2
|
||||||
|
if: github.event.label.name == 'shallow-fusion'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
tree lstm_transducer_stateless2/exp
|
||||||
|
cd lstm_transducer_stateless2/exp
|
||||||
|
echo "===modified_beam_search_rnnlm_shallow_fusion==="
|
||||||
|
find modified_beam_search_rnnlm_shallow_fusion -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find modified_beam_search_rnnlm_shallow_fusion -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
|
- name: Display decoding results for lstm_transducer_stateless2
|
||||||
|
if: github.event.label.name == 'LODR'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
tree lstm_transducer_stateless2/exp
|
||||||
|
cd lstm_transducer_stateless2/exp
|
||||||
|
echo "===modified_beam_search_rnnlm_LODR==="
|
||||||
|
find modified_beam_search_rnnlm_LODR -name "log-*" -exec grep -n --color "best for test-clean" {} + | sort -n -k2
|
||||||
|
find modified_beam_search_rnnlm_LODR -name "log-*" -exec grep -n --color "best for test-other" {} + | sort -n -k2
|
||||||
|
|
||||||
- name: Upload decoding results for lstm_transducer_stateless2
|
- name: Upload decoding results for lstm_transducer_stateless2
|
||||||
uses: actions/upload-artifact@v2
|
uses: actions/upload-artifact@v2
|
||||||
if: github.event_name == 'schedule' || github.event.label.name == 'ncnn'
|
if: github.event_name == 'schedule' || github.event.label.name == 'shallow-fusion' || github.event.label.name == 'LODR'
|
||||||
with:
|
with:
|
||||||
name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-lstm_transducer_stateless2-2022-09-03
|
name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-ubuntu-18.04-cpu-lstm_transducer_stateless2-2022-09-03
|
||||||
path: egs/librispeech/ASR/lstm_transducer_stateless2/exp/
|
path: egs/librispeech/ASR/lstm_transducer_stateless2/exp/
|
||||||
|
@ -33,6 +33,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_pruned_transducer_stateless3_2022_05_13-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_librispeech_pruned_transducer_stateless3_2022_05_13:
|
run_librispeech_pruned_transducer_stateless3_2022_05_13:
|
||||||
if: github.event.label.name == 'onnx' || github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'onnx' || github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -33,6 +33,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_streaming_2022_06_26-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_librispeech_streaming_2022_06_26:
|
run_librispeech_streaming_2022_06_26:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -33,6 +33,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_librispeech_2022_04_19-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_librispeech_2022_04_19:
|
run_librispeech_2022_04_19:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -23,6 +23,10 @@ on:
|
|||||||
pull_request:
|
pull_request:
|
||||||
types: [labeled]
|
types: [labeled]
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_pre_trained_conformer_ctc-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_pre_trained_conformer_ctc:
|
run_pre_trained_conformer_ctc:
|
||||||
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
||||||
|
@ -32,6 +32,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_pre_trained_transducer_stateless_multi_datasets_librispeech_100h-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_pre_trained_transducer_stateless_multi_datasets_librispeech_100h:
|
run_pre_trained_transducer_stateless_multi_datasets_librispeech_100h:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -32,6 +32,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_pre_trained_transducer_stateless_multi_datasets_librispeech_960h-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_pre_trained_transducer_stateless_multi_datasets_librispeech_960h:
|
run_pre_trained_transducer_stateless_multi_datasets_librispeech_960h:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -23,6 +23,10 @@ on:
|
|||||||
pull_request:
|
pull_request:
|
||||||
types: [labeled]
|
types: [labeled]
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_pre_trained_transducer_stateless_modified_2_aishell-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_pre_trained_transducer_stateless_modified_2_aishell:
|
run_pre_trained_transducer_stateless_modified_2_aishell:
|
||||||
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
||||||
|
@ -23,6 +23,10 @@ on:
|
|||||||
pull_request:
|
pull_request:
|
||||||
types: [labeled]
|
types: [labeled]
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_pre_trained_transducer_stateless_modified_aishell-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_pre_trained_transducer_stateless_modified_aishell:
|
run_pre_trained_transducer_stateless_modified_aishell:
|
||||||
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
||||||
|
@ -32,6 +32,10 @@ on:
|
|||||||
# nightly build at 15:50 UTC time every day
|
# nightly build at 15:50 UTC time every day
|
||||||
- cron: "50 15 * * *"
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_pre_trained_transducer_stateless-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_pre_trained_transducer_stateless:
|
run_pre_trained_transducer_stateless:
|
||||||
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
if: github.event.label.name == 'ready' || github.event.label.name == 'run-decode' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
@ -23,6 +23,10 @@ on:
|
|||||||
pull_request:
|
pull_request:
|
||||||
types: [labeled]
|
types: [labeled]
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_pre_trained_transducer-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run_pre_trained_transducer:
|
run_pre_trained_transducer:
|
||||||
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
||||||
|
71
.github/workflows/run-ptb-rnn-lm.yml
vendored
Normal file
@ -0,0 +1,71 @@
|
|||||||
|
name: run-ptb-rnn-lm-training
|
||||||
|
|
||||||
|
on:
|
||||||
|
push:
|
||||||
|
branches:
|
||||||
|
- master
|
||||||
|
pull_request:
|
||||||
|
types: [labeled]
|
||||||
|
|
||||||
|
schedule:
|
||||||
|
# minute (0-59)
|
||||||
|
# hour (0-23)
|
||||||
|
# day of the month (1-31)
|
||||||
|
# month (1-12)
|
||||||
|
# day of the week (0-6)
|
||||||
|
# nightly build at 15:50 UTC time every day
|
||||||
|
- cron: "50 15 * * *"
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_ptb_rnn_lm_training-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
|
jobs:
|
||||||
|
run_ptb_rnn_lm_training:
|
||||||
|
if: github.event.label.name == 'ready' || github.event.label.name == 'rnnlm' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
runs-on: ${{ matrix.os }}
|
||||||
|
strategy:
|
||||||
|
matrix:
|
||||||
|
os: [ubuntu-latest]
|
||||||
|
python-version: ["3.8"]
|
||||||
|
|
||||||
|
fail-fast: false
|
||||||
|
|
||||||
|
steps:
|
||||||
|
- uses: actions/checkout@v2
|
||||||
|
with:
|
||||||
|
fetch-depth: 0
|
||||||
|
|
||||||
|
- name: Setup Python ${{ matrix.python-version }}
|
||||||
|
uses: actions/setup-python@v2
|
||||||
|
with:
|
||||||
|
python-version: ${{ matrix.python-version }}
|
||||||
|
cache: 'pip'
|
||||||
|
cache-dependency-path: '**/requirements-ci.txt'
|
||||||
|
|
||||||
|
- name: Install Python dependencies
|
||||||
|
run: |
|
||||||
|
grep -v '^#' ./requirements-ci.txt | grep -v kaldifst | xargs -n 1 -L 1 pip install
|
||||||
|
pip uninstall -y protobuf
|
||||||
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
- name: Prepare data
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
cd egs/ptb/LM
|
||||||
|
./prepare.sh
|
||||||
|
|
||||||
|
- name: Run training
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
cd egs/ptb/LM
|
||||||
|
./train-rnn-lm.sh --world-size 1 --num-epochs 5 --use-epoch 4 --use-avg 2
|
||||||
|
|
||||||
|
- name: Upload pretrained models
|
||||||
|
uses: actions/upload-artifact@v2
|
||||||
|
if: github.event.label.name == 'ready' || github.event.label.name == 'rnnlm' || github.event_name == 'push' || github.event_name == 'schedule'
|
||||||
|
with:
|
||||||
|
name: python-${{ matrix.python-version }}-ubuntu-rnn-lm-ptb
|
||||||
|
path: egs/ptb/LM/my-rnnlm-exp/
|
84
.github/workflows/run-wenetspeech-pruned-transducer-stateless2.yml
vendored
Normal file
@ -0,0 +1,84 @@
|
|||||||
|
# Copyright 2021 Fangjun Kuang (csukuangfj@gmail.com)
|
||||||
|
|
||||||
|
# See ../../LICENSE for clarification regarding multiple authors
|
||||||
|
#
|
||||||
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||||
|
# you may not use this file except in compliance with the License.
|
||||||
|
# You may obtain a copy of the License at
|
||||||
|
#
|
||||||
|
# http://www.apache.org/licenses/LICENSE-2.0
|
||||||
|
#
|
||||||
|
# Unless required by applicable law or agreed to in writing, software
|
||||||
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||||
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||||
|
# See the License for the specific language governing permissions and
|
||||||
|
# limitations under the License.
|
||||||
|
|
||||||
|
name: run-wenetspeech-pruned-transducer-stateless2
|
||||||
|
|
||||||
|
on:
|
||||||
|
push:
|
||||||
|
branches:
|
||||||
|
- master
|
||||||
|
pull_request:
|
||||||
|
types: [labeled]
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run_wenetspeech_pruned_transducer_stateless2-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
|
jobs:
|
||||||
|
run_wenetspeech_pruned_transducer_stateless2:
|
||||||
|
if: github.event.label.name == 'onnx' || github.event.label.name == 'ready' || github.event_name == 'push' || github.event.label.name == 'wenetspeech'
|
||||||
|
runs-on: ${{ matrix.os }}
|
||||||
|
strategy:
|
||||||
|
matrix:
|
||||||
|
os: [ubuntu-18.04]
|
||||||
|
python-version: [3.8]
|
||||||
|
|
||||||
|
fail-fast: false
|
||||||
|
|
||||||
|
steps:
|
||||||
|
- uses: actions/checkout@v2
|
||||||
|
with:
|
||||||
|
fetch-depth: 0
|
||||||
|
|
||||||
|
- name: Setup Python ${{ matrix.python-version }}
|
||||||
|
uses: actions/setup-python@v2
|
||||||
|
with:
|
||||||
|
python-version: ${{ matrix.python-version }}
|
||||||
|
cache: 'pip'
|
||||||
|
cache-dependency-path: '**/requirements-ci.txt'
|
||||||
|
|
||||||
|
- name: Install Python dependencies
|
||||||
|
run: |
|
||||||
|
grep -v '^#' ./requirements-ci.txt | xargs -n 1 -L 1 pip install
|
||||||
|
pip uninstall -y protobuf
|
||||||
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
- name: Cache kaldifeat
|
||||||
|
id: my-cache
|
||||||
|
uses: actions/cache@v2
|
||||||
|
with:
|
||||||
|
path: |
|
||||||
|
~/tmp/kaldifeat
|
||||||
|
key: cache-tmp-${{ matrix.python-version }}-2022-09-25
|
||||||
|
|
||||||
|
- name: Install kaldifeat
|
||||||
|
if: steps.my-cache.outputs.cache-hit != 'true'
|
||||||
|
shell: bash
|
||||||
|
run: |
|
||||||
|
.github/scripts/install-kaldifeat.sh
|
||||||
|
|
||||||
|
- name: Inference with pre-trained model
|
||||||
|
shell: bash
|
||||||
|
env:
|
||||||
|
GITHUB_EVENT_NAME: ${{ github.event_name }}
|
||||||
|
GITHUB_EVENT_LABEL_NAME: ${{ github.event.label.name }}
|
||||||
|
run: |
|
||||||
|
sudo apt-get -qq install git-lfs tree sox
|
||||||
|
export PYTHONPATH=$PWD:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/kaldifeat/python:$PYTHONPATH
|
||||||
|
export PYTHONPATH=~/tmp/kaldifeat/build/lib:$PYTHONPATH
|
||||||
|
|
||||||
|
.github/scripts/run-wenetspeech-pruned-transducer-stateless2.sh
|
10
.github/workflows/run-yesno-recipe.yml
vendored
@ -21,11 +21,15 @@ on:
|
|||||||
branches:
|
branches:
|
||||||
- master
|
- master
|
||||||
pull_request:
|
pull_request:
|
||||||
types: [labeled]
|
branches:
|
||||||
|
- master
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: run-yesno-recipe-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
run-yesno-recipe:
|
run-yesno-recipe:
|
||||||
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
|
||||||
runs-on: ${{ matrix.os }}
|
runs-on: ${{ matrix.os }}
|
||||||
strategy:
|
strategy:
|
||||||
matrix:
|
matrix:
|
||||||
@ -61,7 +65,7 @@ jobs:
|
|||||||
|
|
||||||
- name: Install Python dependencies
|
- name: Install Python dependencies
|
||||||
run: |
|
run: |
|
||||||
grep -v '^#' ./requirements-ci.txt | xargs -n 1 -L 1 pip install
|
grep -v '^#' ./requirements-ci.txt | grep -v kaldifst | xargs -n 1 -L 1 pip install
|
||||||
pip uninstall -y protobuf
|
pip uninstall -y protobuf
|
||||||
pip install --no-binary protobuf protobuf
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
19
.github/workflows/style_check.yml
vendored
@ -24,13 +24,17 @@ on:
|
|||||||
branches:
|
branches:
|
||||||
- master
|
- master
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: style_check-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
style_check:
|
style_check:
|
||||||
runs-on: ${{ matrix.os }}
|
runs-on: ${{ matrix.os }}
|
||||||
strategy:
|
strategy:
|
||||||
matrix:
|
matrix:
|
||||||
os: [ubuntu-18.04, macos-latest]
|
os: [ubuntu-latest]
|
||||||
python-version: [3.7, 3.9]
|
python-version: [3.8]
|
||||||
fail-fast: false
|
fail-fast: false
|
||||||
|
|
||||||
steps:
|
steps:
|
||||||
@ -45,17 +49,18 @@ jobs:
|
|||||||
|
|
||||||
- name: Install Python dependencies
|
- name: Install Python dependencies
|
||||||
run: |
|
run: |
|
||||||
python3 -m pip install --upgrade pip black==21.6b0 flake8==3.9.2 click==8.0.4
|
python3 -m pip install --upgrade pip black==22.3.0 flake8==5.0.4 click==8.1.0
|
||||||
# See https://github.com/psf/black/issues/2964
|
# Click issue fixed in https://github.com/psf/black/pull/2966
|
||||||
# The version of click should be selected from 8.0.0, 8.0.1, 8.0.2, 8.0.3, and 8.0.4
|
|
||||||
|
|
||||||
- name: Run flake8
|
- name: Run flake8
|
||||||
shell: bash
|
shell: bash
|
||||||
working-directory: ${{github.workspace}}
|
working-directory: ${{github.workspace}}
|
||||||
run: |
|
run: |
|
||||||
# stop the build if there are Python syntax errors or undefined names
|
# stop the build if there are Python syntax errors or undefined names
|
||||||
flake8 . --count --show-source --statistics
|
flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics
|
||||||
flake8 .
|
# exit-zero treats all errors as warnings. The GitHub editor is 127 chars wide
|
||||||
|
flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 \
|
||||||
|
--statistics --extend-ignore=E203,E266,E501,F401,E402,F403,F841,W503
|
||||||
|
|
||||||
- name: Run black
|
- name: Run black
|
||||||
shell: bash
|
shell: bash
|
||||||
|
64
.github/workflows/test.yml
vendored
@ -21,26 +21,23 @@ on:
|
|||||||
branches:
|
branches:
|
||||||
- master
|
- master
|
||||||
pull_request:
|
pull_request:
|
||||||
types: [labeled]
|
branches:
|
||||||
|
- master
|
||||||
|
|
||||||
|
concurrency:
|
||||||
|
group: test-${{ github.ref }}
|
||||||
|
cancel-in-progress: true
|
||||||
|
|
||||||
jobs:
|
jobs:
|
||||||
test:
|
test:
|
||||||
if: github.event.label.name == 'ready' || github.event_name == 'push'
|
|
||||||
runs-on: ${{ matrix.os }}
|
runs-on: ${{ matrix.os }}
|
||||||
strategy:
|
strategy:
|
||||||
matrix:
|
matrix:
|
||||||
# os: [ubuntu-18.04, macos-10.15]
|
os: [ubuntu-latest]
|
||||||
# disable macOS test for now.
|
python-version: ["3.8"]
|
||||||
os: [ubuntu-18.04]
|
torch: ["1.10.0"]
|
||||||
python-version: [3.7, 3.8]
|
torchaudio: ["0.10.0"]
|
||||||
torch: ["1.8.0", "1.11.0"]
|
k2-version: ["1.23.2.dev20221201"]
|
||||||
torchaudio: ["0.8.0", "0.11.0"]
|
|
||||||
k2-version: ["1.15.1.dev20220427"]
|
|
||||||
exclude:
|
|
||||||
- torch: "1.8.0"
|
|
||||||
torchaudio: "0.11.0"
|
|
||||||
- torch: "1.11.0"
|
|
||||||
torchaudio: "0.8.0"
|
|
||||||
|
|
||||||
fail-fast: false
|
fail-fast: false
|
||||||
|
|
||||||
@ -67,11 +64,7 @@ jobs:
|
|||||||
# numpy 1.20.x does not support python 3.6
|
# numpy 1.20.x does not support python 3.6
|
||||||
pip install numpy==1.19
|
pip install numpy==1.19
|
||||||
pip install torch==${{ matrix.torch }}+cpu -f https://download.pytorch.org/whl/cpu/torch_stable.html
|
pip install torch==${{ matrix.torch }}+cpu -f https://download.pytorch.org/whl/cpu/torch_stable.html
|
||||||
if [[ ${{ matrix.torchaudio }} == "0.11.0" ]]; then
|
pip install torchaudio==${{ matrix.torchaudio }}+cpu -f https://download.pytorch.org/whl/cpu/torch_stable.html
|
||||||
pip install torchaudio==${{ matrix.torchaudio }}+cpu -f https://download.pytorch.org/whl/cpu/torch_stable.html
|
|
||||||
else
|
|
||||||
pip install torchaudio==${{ matrix.torchaudio }}
|
|
||||||
fi
|
|
||||||
|
|
||||||
pip install k2==${{ matrix.k2-version }}+cpu.torch${{ matrix.torch }} -f https://k2-fsa.org/nightly/
|
pip install k2==${{ matrix.k2-version }}+cpu.torch${{ matrix.torch }} -f https://k2-fsa.org/nightly/
|
||||||
pip install git+https://github.com/lhotse-speech/lhotse
|
pip install git+https://github.com/lhotse-speech/lhotse
|
||||||
@ -79,6 +72,8 @@ jobs:
|
|||||||
pip uninstall -y protobuf
|
pip uninstall -y protobuf
|
||||||
pip install --no-binary protobuf protobuf
|
pip install --no-binary protobuf protobuf
|
||||||
|
|
||||||
|
pip install kaldifst
|
||||||
|
pip install onnxruntime
|
||||||
pip install -r requirements.txt
|
pip install -r requirements.txt
|
||||||
|
|
||||||
- name: Install graphviz
|
- name: Install graphviz
|
||||||
@ -118,19 +113,20 @@ jobs:
|
|||||||
cd ../pruned_transducer_stateless4
|
cd ../pruned_transducer_stateless4
|
||||||
pytest -v -s
|
pytest -v -s
|
||||||
|
|
||||||
|
cd ../pruned_transducer_stateless7
|
||||||
|
pytest -v -s
|
||||||
|
|
||||||
cd ../transducer_stateless
|
cd ../transducer_stateless
|
||||||
pytest -v -s
|
pytest -v -s
|
||||||
|
|
||||||
if [[ ${{ matrix.torchaudio }} == "0.10.0" ]]; then
|
cd ../transducer
|
||||||
cd ../transducer
|
pytest -v -s
|
||||||
pytest -v -s
|
|
||||||
|
|
||||||
cd ../transducer_stateless2
|
cd ../transducer_stateless2
|
||||||
pytest -v -s
|
pytest -v -s
|
||||||
|
|
||||||
cd ../transducer_lstm
|
cd ../transducer_lstm
|
||||||
pytest -v -s
|
pytest -v -s
|
||||||
fi
|
|
||||||
|
|
||||||
- name: Run tests
|
- name: Run tests
|
||||||
if: startsWith(matrix.os, 'macos')
|
if: startsWith(matrix.os, 'macos')
|
||||||
@ -161,13 +157,11 @@ jobs:
|
|||||||
cd ../transducer_stateless
|
cd ../transducer_stateless
|
||||||
pytest -v -s
|
pytest -v -s
|
||||||
|
|
||||||
if [[ ${{ matrix.torchaudio }} == "0.10.0" ]]; then
|
cd ../transducer
|
||||||
cd ../transducer
|
pytest -v -s
|
||||||
pytest -v -s
|
|
||||||
|
|
||||||
cd ../transducer_stateless2
|
cd ../transducer_stateless2
|
||||||
pytest -v -s
|
pytest -v -s
|
||||||
|
|
||||||
cd ../transducer_lstm
|
cd ../transducer_lstm
|
||||||
pytest -v -s
|
pytest -v -s
|
||||||
fi
|
|
||||||
|
20
.gitignore
vendored
@ -11,5 +11,25 @@ log
|
|||||||
*.bak
|
*.bak
|
||||||
*-bak
|
*-bak
|
||||||
*bak.py
|
*bak.py
|
||||||
|
|
||||||
|
# Ignore Mac system files
|
||||||
|
.DS_store
|
||||||
|
|
||||||
|
# Ignore node_modules folder
|
||||||
|
node_modules
|
||||||
|
|
||||||
|
# ignore .nfs
|
||||||
|
|
||||||
|
.nfs*
|
||||||
|
|
||||||
|
# Ignore all text files
|
||||||
|
*.txt
|
||||||
|
|
||||||
|
# Ignore files related to API keys
|
||||||
|
.env
|
||||||
|
|
||||||
|
# Ignore SASS config files
|
||||||
|
.sass-cache
|
||||||
|
|
||||||
*.param
|
*.param
|
||||||
*.bin
|
*.bin
|
||||||
|
@ -1,26 +1,38 @@
|
|||||||
repos:
|
repos:
|
||||||
- repo: https://github.com/psf/black
|
- repo: https://github.com/psf/black
|
||||||
rev: 21.6b0
|
rev: 22.3.0
|
||||||
hooks:
|
hooks:
|
||||||
- id: black
|
- id: black
|
||||||
args: [--line-length=80]
|
args: ["--line-length=88"]
|
||||||
additional_dependencies: ['click==8.0.1']
|
additional_dependencies: ['click==8.1.0']
|
||||||
exclude: icefall\/__init__\.py
|
exclude: icefall\/__init__\.py
|
||||||
|
|
||||||
- repo: https://github.com/PyCQA/flake8
|
- repo: https://github.com/PyCQA/flake8
|
||||||
rev: 3.9.2
|
rev: 5.0.4
|
||||||
hooks:
|
hooks:
|
||||||
- id: flake8
|
- id: flake8
|
||||||
args: [--max-line-length=80]
|
args: ["--max-line-length=88", "--extend-ignore=E203,E266,E501,F401,E402,F403,F841,W503"]
|
||||||
|
|
||||||
|
# What are we ignoring here?
|
||||||
|
# E203: whitespace before ':'
|
||||||
|
# E266: too many leading '#' for block comment
|
||||||
|
# E501: line too long
|
||||||
|
# F401: module imported but unused
|
||||||
|
# E402: module level import not at top of file
|
||||||
|
# F403: 'from module import *' used; unable to detect undefined names
|
||||||
|
# F841: local variable is assigned to but never used
|
||||||
|
# W503: line break before binary operator
|
||||||
|
# In addition, the default ignore list is:
|
||||||
|
# E121,E123,E126,E226,E24,E704,W503,W504
|
||||||
|
|
||||||
- repo: https://github.com/pycqa/isort
|
- repo: https://github.com/pycqa/isort
|
||||||
rev: 5.9.2
|
rev: 5.10.1
|
||||||
hooks:
|
hooks:
|
||||||
- id: isort
|
- id: isort
|
||||||
args: [--profile=black, --line-length=80]
|
args: ["--profile=black"]
|
||||||
|
|
||||||
- repo: https://github.com/pre-commit/pre-commit-hooks
|
- repo: https://github.com/pre-commit/pre-commit-hooks
|
||||||
rev: v4.0.1
|
rev: v4.2.0
|
||||||
hooks:
|
hooks:
|
||||||
- id: check-executables-have-shebangs
|
- id: check-executables-have-shebangs
|
||||||
- id: end-of-file-fixer
|
- id: end-of-file-fixer
|
||||||
|
@ -82,7 +82,7 @@ The WER for this model is:
|
|||||||
|-----|------------|------------|
|
|-----|------------|------------|
|
||||||
| WER | 6.59 | 17.69 |
|
| WER | 6.59 | 17.69 |
|
||||||
|
|
||||||
We provide a Colab notebook to run a pre-trained TDNN LSTM CTC model: [](https://colab.research.google.com/drive/1kNmDXNMwREi0rZGAOIAOJo93REBuOTcd?usp=sharing)
|
We provide a Colab notebook to run a pre-trained TDNN LSTM CTC model: [](https://colab.research.google.com/drive/1-iSfQMp2So-We_Uu49N4AAcMInB72u9z?usp=sharing)
|
||||||
|
|
||||||
|
|
||||||
#### Transducer: Conformer encoder + LSTM decoder
|
#### Transducer: Conformer encoder + LSTM decoder
|
||||||
@ -162,7 +162,7 @@ The CER for this model is:
|
|||||||
|-----|-------|
|
|-----|-------|
|
||||||
| CER | 10.16 |
|
| CER | 10.16 |
|
||||||
|
|
||||||
We provide a Colab notebook to run a pre-trained TDNN LSTM CTC model: [](https://colab.research.google.com/drive/1qULaGvXq7PCu_P61oubfz9b53JzY4H3z?usp=sharing)
|
We provide a Colab notebook to run a pre-trained TDNN LSTM CTC model: [](https://colab.research.google.com/drive/1jbyzYq3ytm6j2nlEt-diQm-6QVWyDDEa?usp=sharing)
|
||||||
|
|
||||||
### TIMIT
|
### TIMIT
|
||||||
|
|
||||||
|
@ -72,14 +72,14 @@ docker run -it --runtime=nvidia --shm-size=2gb --name=icefall --gpus all icefall
|
|||||||
```
|
```
|
||||||
|
|
||||||
### Tips:
|
### Tips:
|
||||||
1. Since your data and models most probably won't be in the docker, you must use the -v flag to access the host machine. Do this by specifying `-v {/path/in/docker}:{/path/in/host/machine}`.
|
1. Since your data and models most probably won't be in the docker, you must use the -v flag to access the host machine. Do this by specifying `-v {/path/in/host/machine}:{/path/in/docker}`.
|
||||||
|
|
||||||
2. Also, if your environment requires a proxy, this would be a good time to add it in too: `-e http_proxy=http://aaa.bb.cc.net:8080 -e https_proxy=http://aaa.bb.cc.net:8080`.
|
2. Also, if your environment requires a proxy, this would be a good time to add it in too: `-e http_proxy=http://aaa.bb.cc.net:8080 -e https_proxy=http://aaa.bb.cc.net:8080`.
|
||||||
|
|
||||||
Overall, your docker run command should look like this.
|
Overall, your docker run command should look like this.
|
||||||
|
|
||||||
```bash
|
```bash
|
||||||
docker run -it --runtime=nvidia --shm-size=2gb --name=icefall --gpus all -v {/path/in/docker}:{/path/in/host/machine} -e http_proxy=http://aaa.bb.cc.net:8080 -e https_proxy=http://aaa.bb.cc.net:8080 icefall/pytorch1.12.1
|
docker run -it --runtime=nvidia --shm-size=2gb --name=icefall --gpus all -v {/path/in/host/machine}:{/path/in/docker} -e http_proxy=http://aaa.bb.cc.net:8080 -e https_proxy=http://aaa.bb.cc.net:8080 icefall/pytorch1.12.1
|
||||||
```
|
```
|
||||||
|
|
||||||
You can explore more docker run options [here](https://docs.docker.com/engine/reference/commandline/run/) to suit your environment.
|
You can explore more docker run options [here](https://docs.docker.com/engine/reference/commandline/run/) to suit your environment.
|
||||||
|
@ -51,8 +51,9 @@ RUN wget -P /opt https://downloads.xiph.org/releases/flac/flac-1.3.2.tar.xz &&
|
|||||||
find /opt/flac-1.3.2 -type f \( -name "*.o" -o -name "*.la" -o -name "*.a" \) -exec rm {} \; && \
|
find /opt/flac-1.3.2 -type f \( -name "*.o" -o -name "*.la" -o -name "*.a" \) -exec rm {} \; && \
|
||||||
cd -
|
cd -
|
||||||
|
|
||||||
RUN pip install kaldiio graphviz && \
|
RUN conda install -y -c pytorch torchaudio=0.12 && \
|
||||||
conda install -y -c pytorch torchaudio
|
pip install graphviz
|
||||||
|
|
||||||
|
|
||||||
#install k2 from source
|
#install k2 from source
|
||||||
RUN git clone https://github.com/k2-fsa/k2.git /opt/k2 && \
|
RUN git clone https://github.com/k2-fsa/k2.git /opt/k2 && \
|
||||||
@ -67,6 +68,7 @@ RUN git clone https://github.com/k2-fsa/icefall /workspace/icefall && \
|
|||||||
cd /workspace/icefall && \
|
cd /workspace/icefall && \
|
||||||
pip install -r requirements.txt
|
pip install -r requirements.txt
|
||||||
|
|
||||||
|
RUN pip install kaldifeat
|
||||||
ENV PYTHONPATH /workspace/icefall:$PYTHONPATH
|
ENV PYTHONPATH /workspace/icefall:$PYTHONPATH
|
||||||
|
|
||||||
WORKDIR /workspace/icefall
|
WORKDIR /workspace/icefall
|
@ -69,8 +69,8 @@ RUN wget -P /opt https://downloads.xiph.org/releases/flac/flac-1.3.2.tar.xz &&
|
|||||||
find /opt/flac-1.3.2 -type f \( -name "*.o" -o -name "*.la" -o -name "*.a" \) -exec rm {} \; && \
|
find /opt/flac-1.3.2 -type f \( -name "*.o" -o -name "*.la" -o -name "*.a" \) -exec rm {} \; && \
|
||||||
cd -
|
cd -
|
||||||
|
|
||||||
RUN pip install kaldiio graphviz && \
|
RUN conda install -y -c pytorch torchaudio=0.7.1 && \
|
||||||
conda install -y -c pytorch torchaudio=0.7.1
|
pip install graphviz
|
||||||
|
|
||||||
#install k2 from source
|
#install k2 from source
|
||||||
RUN git clone https://github.com/k2-fsa/k2.git /opt/k2 && \
|
RUN git clone https://github.com/k2-fsa/k2.git /opt/k2 && \
|
||||||
@ -88,4 +88,3 @@ RUN git clone https://github.com/k2-fsa/icefall /workspace/icefall && \
|
|||||||
ENV PYTHONPATH /workspace/icefall:$PYTHONPATH
|
ENV PYTHONPATH /workspace/icefall:$PYTHONPATH
|
||||||
|
|
||||||
WORKDIR /workspace/icefall
|
WORKDIR /workspace/icefall
|
||||||
|
|
||||||
|
@ -74,7 +74,7 @@ html_context = {
|
|||||||
"github_user": "k2-fsa",
|
"github_user": "k2-fsa",
|
||||||
"github_repo": "icefall",
|
"github_repo": "icefall",
|
||||||
"github_version": "master",
|
"github_version": "master",
|
||||||
"conf_py_path": "/icefall/docs/source/",
|
"conf_py_path": "/docs/source/",
|
||||||
}
|
}
|
||||||
|
|
||||||
todo_include_todos = True
|
todo_include_todos = True
|
||||||
|
@ -11,9 +11,9 @@ We use the following tools to make the code style to be as consistent as possibl
|
|||||||
|
|
||||||
The following versions of the above tools are used:
|
The following versions of the above tools are used:
|
||||||
|
|
||||||
- ``black == 12.6b0``
|
- ``black == 22.3.0``
|
||||||
- ``flake8 == 3.9.2``
|
- ``flake8 == 5.0.4``
|
||||||
- ``isort == 5.9.2``
|
- ``isort == 5.10.1``
|
||||||
|
|
||||||
After running the following commands:
|
After running the following commands:
|
||||||
|
|
||||||
@ -54,10 +54,17 @@ it should succeed this time:
|
|||||||
If you want to check the style of your code before ``git commit``, you
|
If you want to check the style of your code before ``git commit``, you
|
||||||
can do the following:
|
can do the following:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ pre-commit install
|
||||||
|
$ pre-commit run
|
||||||
|
|
||||||
|
Or without installing the pre-commit hooks:
|
||||||
|
|
||||||
.. code-block:: bash
|
.. code-block:: bash
|
||||||
|
|
||||||
$ cd icefall
|
$ cd icefall
|
||||||
$ pip install black==21.6b0 flake8==3.9.2 isort==5.9.2
|
$ pip install black==22.3.0 flake8==5.0.4 isort==5.10.1
|
||||||
$ black --check your_changed_file.py
|
$ black --check your_changed_file.py
|
||||||
$ black your_changed_file.py # modify it in-place
|
$ black your_changed_file.py # modify it in-place
|
||||||
$
|
$
|
||||||
|
@ -21,6 +21,15 @@ speech recognition recipes using `k2 <https://github.com/k2-fsa/k2>`_.
|
|||||||
:caption: Contents:
|
:caption: Contents:
|
||||||
|
|
||||||
installation/index
|
installation/index
|
||||||
|
model-export/index
|
||||||
|
|
||||||
|
.. toctree::
|
||||||
|
:maxdepth: 3
|
||||||
|
|
||||||
recipes/index
|
recipes/index
|
||||||
|
|
||||||
|
.. toctree::
|
||||||
|
:maxdepth: 2
|
||||||
|
|
||||||
contributing/index
|
contributing/index
|
||||||
huggingface/index
|
huggingface/index
|
||||||
|
@ -393,6 +393,17 @@ Now let us run the training part:
|
|||||||
We use ``export CUDA_VISIBLE_DEVICES=""`` so that ``icefall`` uses CPU
|
We use ``export CUDA_VISIBLE_DEVICES=""`` so that ``icefall`` uses CPU
|
||||||
even if there are GPUs available.
|
even if there are GPUs available.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
In case you get a ``Segmentation fault (core dump)`` error, please use:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python
|
||||||
|
|
||||||
|
See more at `<https://github.com/k2-fsa/icefall/issues/674>` if you are
|
||||||
|
interested.
|
||||||
|
|
||||||
The training log is given below:
|
The training log is given below:
|
||||||
|
|
||||||
.. code-block::
|
.. code-block::
|
||||||
|
@ -0,0 +1,21 @@
|
|||||||
|
2022-10-13 19:09:02,233 INFO [pretrained.py:265] {'best_train_loss': inf, 'best_valid_loss': inf, 'best_train_epoch': -1, 'best_valid_epoch': -1, 'batch_idx_train': 0, 'log_interval': 50, 'reset_interval': 200, 'valid_interval': 3000, 'feature_dim': 80, 'subsampling_factor': 4, 'encoder_dim': 512, 'nhead': 8, 'dim_feedforward': 2048, 'num_encoder_layers': 12, 'decoder_dim': 512, 'joiner_dim': 512, 'model_warm_step': 3000, 'env_info': {'k2-version': '1.21', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '4810e00d8738f1a21278b0156a42ff396a2d40ac', 'k2-git-date': 'Fri Oct 7 19:35:03 2022', 'lhotse-version': '1.3.0.dev+missing.version.file', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': False, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'onnx-doc-1013', 'icefall-git-sha1': 'c39cba5-dirty', 'icefall-git-date': 'Thu Oct 13 15:17:20 2022', 'icefall-path': '/k2-dev/fangjun/open-source/icefall-master', 'k2-path': '/k2-dev/fangjun/open-source/k2-master/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-fj/fangjun/open-source-2/lhotse-jsonl/lhotse/__init__.py', 'hostname': 'de-74279-k2-test-4-0324160024-65bfd8b584-jjlbn', 'IP address': '10.177.74.203'}, 'checkpoint': './icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/exp/pretrained-iter-1224000-avg-14.pt', 'bpe_model': './icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/data/lang_bpe_500/bpe.model', 'method': 'greedy_search', 'sound_files': ['./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1089-134686-0001.wav', './icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1221-135766-0001.wav', './icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1221-135766-0002.wav'], 'sample_rate': 16000, 'beam_size': 4, 'beam': 4, 'max_contexts': 4, 'max_states': 8, 'context_size': 2, 'max_sym_per_frame': 1, 'simulate_streaming': False, 'decode_chunk_size': 16, 'left_context': 64, 'dynamic_chunk_training': False, 'causal_convolution': False, 'short_chunk_size': 25, 'num_left_chunks': 4, 'blank_id': 0, 'unk_id': 2, 'vocab_size': 500}
|
||||||
|
2022-10-13 19:09:02,233 INFO [pretrained.py:271] device: cpu
|
||||||
|
2022-10-13 19:09:02,233 INFO [pretrained.py:273] Creating model
|
||||||
|
2022-10-13 19:09:02,612 INFO [train.py:458] Disable giga
|
||||||
|
2022-10-13 19:09:02,623 INFO [pretrained.py:277] Number of model parameters: 78648040
|
||||||
|
2022-10-13 19:09:02,951 INFO [pretrained.py:285] Constructing Fbank computer
|
||||||
|
2022-10-13 19:09:02,952 INFO [pretrained.py:295] Reading sound files: ['./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1089-134686-0001.wav', './icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1221-135766-0001.wav', './icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1221-135766-0002.wav']
|
||||||
|
2022-10-13 19:09:02,957 INFO [pretrained.py:301] Decoding started
|
||||||
|
2022-10-13 19:09:06,700 INFO [pretrained.py:329] Using greedy_search
|
||||||
|
2022-10-13 19:09:06,912 INFO [pretrained.py:388]
|
||||||
|
./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1089-134686-0001.wav:
|
||||||
|
AFTER EARLY NIGHTFALL THE YELLOW LAMPS WOULD LIGHT UP HERE AND THERE THE SQUALID QUARTER OF THE BROTHELS
|
||||||
|
|
||||||
|
./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1221-135766-0001.wav:
|
||||||
|
GOD AS A DIRECT CONSEQUENCE OF THE SIN WHICH MAN THUS PUNISHED HAD GIVEN HER A LOVELY CHILD WHOSE PLACE WAS ON THAT SAME DISHONORED BOSOM TO CONNECT HER PARENT FOREVER WITH THE RACE AND DESCENT OF MORTALS AND TO BE FINALLY A BLESSED SOUL IN HEAVEN
|
||||||
|
|
||||||
|
./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1221-135766-0002.wav:
|
||||||
|
YET THESE THOUGHTS AFFECTED HESTER PRYNNE LESS WITH HOPE THAN APPREHENSION
|
||||||
|
|
||||||
|
|
||||||
|
2022-10-13 19:09:06,912 INFO [pretrained.py:390] Decoding Done
|
135
docs/source/model-export/export-model-state-dict.rst
Normal file
@ -0,0 +1,135 @@
|
|||||||
|
Export model.state_dict()
|
||||||
|
=========================
|
||||||
|
|
||||||
|
When to use it
|
||||||
|
--------------
|
||||||
|
|
||||||
|
During model training, we save checkpoints periodically to disk.
|
||||||
|
|
||||||
|
A checkpoint contains the following information:
|
||||||
|
|
||||||
|
- ``model.state_dict()``
|
||||||
|
- ``optimizer.state_dict()``
|
||||||
|
- and some other information related to training
|
||||||
|
|
||||||
|
When we need to resume the training process from some point, we need a checkpoint.
|
||||||
|
However, if we want to publish the model for inference, then only
|
||||||
|
``model.state_dict()`` is needed. In this case, we need to strip all other information
|
||||||
|
except ``model.state_dict()`` to reduce the file size of the published model.
|
||||||
|
|
||||||
|
How to export
|
||||||
|
-------------
|
||||||
|
|
||||||
|
Every recipe contains a file ``export.py`` that you can use to
|
||||||
|
export ``model.state_dict()`` by taking some checkpoints as inputs.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
Each ``export.py`` contains well-documented usage information.
|
||||||
|
|
||||||
|
In the following, we use
|
||||||
|
`<https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/pruned_transducer_stateless3/export.py>`_
|
||||||
|
as an example.
|
||||||
|
|
||||||
|
.. note::
|
||||||
|
|
||||||
|
The steps for other recipes are almost the same.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
./pruned_transducer_stateless3/export.py \
|
||||||
|
--exp-dir ./pruned_transducer_stateless3/exp \
|
||||||
|
--bpe-model data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 20 \
|
||||||
|
--avg 10
|
||||||
|
|
||||||
|
will generate a file ``pruned_transducer_stateless3/exp/pretrained.pt``, which
|
||||||
|
is a dict containing ``{"model": model.state_dict()}`` saved by ``torch.save()``.
|
||||||
|
|
||||||
|
How to use the exported model
|
||||||
|
-----------------------------
|
||||||
|
|
||||||
|
For each recipe, we provide pretrained models hosted on huggingface.
|
||||||
|
You can find links to pretrained models in ``RESULTS.md`` of each dataset.
|
||||||
|
|
||||||
|
In the following, we demonstrate how to use the pretrained model from
|
||||||
|
`<https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13>`_.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
git lfs install
|
||||||
|
git clone https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13
|
||||||
|
|
||||||
|
After cloning the repo with ``git lfs``, you will find several files in the folder
|
||||||
|
``icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/exp``
|
||||||
|
that have a prefix ``pretrained-``. Those files contain ``model.state_dict()``
|
||||||
|
exported by the above ``export.py``.
|
||||||
|
|
||||||
|
In each recipe, there is also a file ``pretrained.py``, which can use
|
||||||
|
``pretrained-xxx.pt`` to decode waves. The following is an example:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
./pruned_transducer_stateless3/pretrained.py \
|
||||||
|
--checkpoint ./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/exp/pretrained-iter-1224000-avg-14.pt \
|
||||||
|
--bpe-model ./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/data/lang_bpe_500/bpe.model \
|
||||||
|
--method greedy_search \
|
||||||
|
./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1089-134686-0001.wav \
|
||||||
|
./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1221-135766-0001.wav \
|
||||||
|
./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/test_wavs/1221-135766-0002.wav
|
||||||
|
|
||||||
|
The above commands show how to use the exported model with ``pretrained.py`` to
|
||||||
|
decode multiple sound files. Its output is given as follows for reference:
|
||||||
|
|
||||||
|
.. literalinclude:: ./code/export-model-state-dict-pretrained-out.txt
|
||||||
|
|
||||||
|
Use the exported model to run decode.py
|
||||||
|
---------------------------------------
|
||||||
|
|
||||||
|
When we publish the model, we always note down its WERs on some test
|
||||||
|
dataset in ``RESULTS.md``. This section describes how to use the
|
||||||
|
pretrained model to reproduce the WER.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
git lfs install
|
||||||
|
git clone https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13
|
||||||
|
|
||||||
|
cd icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/exp
|
||||||
|
ln -s pretrained-iter-1224000-avg-14.pt epoch-9999.pt
|
||||||
|
cd ../..
|
||||||
|
|
||||||
|
We create a symlink with name ``epoch-9999.pt`` to ``pretrained-iter-1224000-avg-14.pt``,
|
||||||
|
so that we can pass ``--epoch 9999 --avg 1`` to ``decode.py`` in the following
|
||||||
|
command:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
./pruned_transducer_stateless3/decode.py \
|
||||||
|
--epoch 9999 \
|
||||||
|
--avg 1 \
|
||||||
|
--exp-dir ./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/exp \
|
||||||
|
--lang-dir ./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/data/lang_bpe_500 \
|
||||||
|
--max-duration 600 \
|
||||||
|
--decoding-method greedy_search
|
||||||
|
|
||||||
|
You will find the decoding results in
|
||||||
|
``./icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13/exp/greedy_search``.
|
||||||
|
|
||||||
|
.. caution::
|
||||||
|
|
||||||
|
For some recipes, you also need to pass ``--use-averaged-model False``
|
||||||
|
to ``decode.py``. The reason is that the exported pretrained model is already
|
||||||
|
the averaged one.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
Before running ``decode.py``, we assume that you have already run
|
||||||
|
``prepare.sh`` to prepare the test dataset.
|
12
docs/source/model-export/export-ncnn.rst
Normal file
@ -0,0 +1,12 @@
|
|||||||
|
Export to ncnn
|
||||||
|
==============
|
||||||
|
|
||||||
|
We support exporting LSTM transducer models to `ncnn <https://github.com/tencent/ncnn>`_.
|
||||||
|
|
||||||
|
Please refer to :ref:`export-model-for-ncnn` for details.
|
||||||
|
|
||||||
|
We also provide `<https://github.com/k2-fsa/sherpa-ncnn>`_
|
||||||
|
performing speech recognition using ``ncnn`` with exported models.
|
||||||
|
It has been tested on Linux, macOS, Windows, and Raspberry Pi. The project is
|
||||||
|
self-contained and can be statically linked to produce a binary containing
|
||||||
|
everything needed.
|
69
docs/source/model-export/export-onnx.rst
Normal file
@ -0,0 +1,69 @@
|
|||||||
|
Export to ONNX
|
||||||
|
==============
|
||||||
|
|
||||||
|
In this section, we describe how to export models to ONNX.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
Only non-streaming conformer transducer models are tested.
|
||||||
|
|
||||||
|
|
||||||
|
When to use it
|
||||||
|
--------------
|
||||||
|
|
||||||
|
It you want to use an inference framework that supports ONNX
|
||||||
|
to run the pretrained model.
|
||||||
|
|
||||||
|
|
||||||
|
How to export
|
||||||
|
-------------
|
||||||
|
|
||||||
|
We use
|
||||||
|
`<https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless3>`_
|
||||||
|
as an example in the following.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
epoch=14
|
||||||
|
avg=2
|
||||||
|
|
||||||
|
./pruned_transducer_stateless3/export.py \
|
||||||
|
--exp-dir ./pruned_transducer_stateless3/exp \
|
||||||
|
--bpe-model data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch $epoch \
|
||||||
|
--avg $avg \
|
||||||
|
--onnx 1
|
||||||
|
|
||||||
|
It will generate the following files inside ``pruned_transducer_stateless3/exp``:
|
||||||
|
|
||||||
|
- ``encoder.onnx``
|
||||||
|
- ``decoder.onnx``
|
||||||
|
- ``joiner.onnx``
|
||||||
|
- ``joiner_encoder_proj.onnx``
|
||||||
|
- ``joiner_decoder_proj.onnx``
|
||||||
|
|
||||||
|
You can use ``./pruned_transducer_stateless3/exp/onnx_pretrained.py`` to decode
|
||||||
|
waves with the generated files:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
./pruned_transducer_stateless3/onnx_pretrained.py \
|
||||||
|
--bpe-model ./data/lang_bpe_500/bpe.model \
|
||||||
|
--encoder-model-filename ./pruned_transducer_stateless3/exp/encoder.onnx \
|
||||||
|
--decoder-model-filename ./pruned_transducer_stateless3/exp/decoder.onnx \
|
||||||
|
--joiner-model-filename ./pruned_transducer_stateless3/exp/joiner.onnx \
|
||||||
|
--joiner-encoder-proj-model-filename ./pruned_transducer_stateless3/exp/joiner_encoder_proj.onnx \
|
||||||
|
--joiner-decoder-proj-model-filename ./pruned_transducer_stateless3/exp/joiner_decoder_proj.onnx \
|
||||||
|
/path/to/foo.wav \
|
||||||
|
/path/to/bar.wav \
|
||||||
|
/path/to/baz.wav
|
||||||
|
|
||||||
|
|
||||||
|
How to use the exported model
|
||||||
|
-----------------------------
|
||||||
|
|
||||||
|
We also provide `<https://github.com/k2-fsa/sherpa-onnx>`_
|
||||||
|
performing speech recognition using `onnxruntime <https://github.com/microsoft/onnxruntime>`_
|
||||||
|
with exported models.
|
||||||
|
It has been tested on Linux, macOS, and Windows.
|
58
docs/source/model-export/export-with-torch-jit-script.rst
Normal file
@ -0,0 +1,58 @@
|
|||||||
|
.. _export-model-with-torch-jit-script:
|
||||||
|
|
||||||
|
Export model with torch.jit.script()
|
||||||
|
===================================
|
||||||
|
|
||||||
|
In this section, we describe how to export a model via
|
||||||
|
``torch.jit.script()``.
|
||||||
|
|
||||||
|
When to use it
|
||||||
|
--------------
|
||||||
|
|
||||||
|
If we want to use our trained model with torchscript,
|
||||||
|
we can use ``torch.jit.script()``.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
See :ref:`export-model-with-torch-jit-trace`
|
||||||
|
if you want to use ``torch.jit.trace()``.
|
||||||
|
|
||||||
|
How to export
|
||||||
|
-------------
|
||||||
|
|
||||||
|
We use
|
||||||
|
`<https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless3>`_
|
||||||
|
as an example in the following.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
epoch=14
|
||||||
|
avg=1
|
||||||
|
|
||||||
|
./pruned_transducer_stateless3/export.py \
|
||||||
|
--exp-dir ./pruned_transducer_stateless3/exp \
|
||||||
|
--bpe-model data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch $epoch \
|
||||||
|
--avg $avg \
|
||||||
|
--jit 1
|
||||||
|
|
||||||
|
It will generate a file ``cpu_jit.pt`` in ``pruned_transducer_stateless3/exp``.
|
||||||
|
|
||||||
|
.. caution::
|
||||||
|
|
||||||
|
Don't be confused by ``cpu`` in ``cpu_jit.pt``. We move all parameters
|
||||||
|
to CPU before saving it into a ``pt`` file; that's why we use ``cpu``
|
||||||
|
in the filename.
|
||||||
|
|
||||||
|
How to use the exported model
|
||||||
|
-----------------------------
|
||||||
|
|
||||||
|
Please refer to the following pages for usage:
|
||||||
|
|
||||||
|
- `<https://k2-fsa.github.io/sherpa/python/streaming_asr/emformer/index.html>`_
|
||||||
|
- `<https://k2-fsa.github.io/sherpa/python/streaming_asr/conv_emformer/index.html>`_
|
||||||
|
- `<https://k2-fsa.github.io/sherpa/python/streaming_asr/conformer/index.html>`_
|
||||||
|
- `<https://k2-fsa.github.io/sherpa/python/offline_asr/conformer/index.html>`_
|
||||||
|
- `<https://k2-fsa.github.io/sherpa/cpp/offline_asr/gigaspeech.html>`_
|
||||||
|
- `<https://k2-fsa.github.io/sherpa/cpp/offline_asr/wenetspeech.html>`_
|
69
docs/source/model-export/export-with-torch-jit-trace.rst
Normal file
@ -0,0 +1,69 @@
|
|||||||
|
.. _export-model-with-torch-jit-trace:
|
||||||
|
|
||||||
|
Export model with torch.jit.trace()
|
||||||
|
===================================
|
||||||
|
|
||||||
|
In this section, we describe how to export a model via
|
||||||
|
``torch.jit.trace()``.
|
||||||
|
|
||||||
|
When to use it
|
||||||
|
--------------
|
||||||
|
|
||||||
|
If we want to use our trained model with torchscript,
|
||||||
|
we can use ``torch.jit.trace()``.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
See :ref:`export-model-with-torch-jit-script`
|
||||||
|
if you want to use ``torch.jit.script()``.
|
||||||
|
|
||||||
|
How to export
|
||||||
|
-------------
|
||||||
|
|
||||||
|
We use
|
||||||
|
`<https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/lstm_transducer_stateless2>`_
|
||||||
|
as an example in the following.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
iter=468000
|
||||||
|
avg=16
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
|
||||||
|
./lstm_transducer_stateless2/export.py \
|
||||||
|
--exp-dir ./lstm_transducer_stateless2/exp \
|
||||||
|
--bpe-model data/lang_bpe_500/bpe.model \
|
||||||
|
--iter $iter \
|
||||||
|
--avg $avg \
|
||||||
|
--jit-trace 1
|
||||||
|
|
||||||
|
It will generate three files inside ``lstm_transducer_stateless2/exp``:
|
||||||
|
|
||||||
|
- ``encoder_jit_trace.pt``
|
||||||
|
- ``decoder_jit_trace.pt``
|
||||||
|
- ``joiner_jit_trace.pt``
|
||||||
|
|
||||||
|
You can use
|
||||||
|
`<https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/lstm_transducer_stateless2/jit_pretrained.py>`_
|
||||||
|
to decode sound files with the following commands:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd egs/librispeech/ASR
|
||||||
|
./lstm_transducer_stateless2/jit_pretrained.py \
|
||||||
|
--bpe-model ./data/lang_bpe_500/bpe.model \
|
||||||
|
--encoder-model-filename ./lstm_transducer_stateless2/exp/encoder_jit_trace.pt \
|
||||||
|
--decoder-model-filename ./lstm_transducer_stateless2/exp/decoder_jit_trace.pt \
|
||||||
|
--joiner-model-filename ./lstm_transducer_stateless2/exp/joiner_jit_trace.pt \
|
||||||
|
/path/to/foo.wav \
|
||||||
|
/path/to/bar.wav \
|
||||||
|
/path/to/baz.wav
|
||||||
|
|
||||||
|
How to use the exported models
|
||||||
|
------------------------------
|
||||||
|
|
||||||
|
Please refer to
|
||||||
|
`<https://k2-fsa.github.io/sherpa/python/streaming_asr/lstm/index.html>`_
|
||||||
|
for its usage in `sherpa <https://k2-fsa.github.io/sherpa/python/streaming_asr/lstm/index.html>`_.
|
||||||
|
You can also find pretrained models there.
|
14
docs/source/model-export/index.rst
Normal file
@ -0,0 +1,14 @@
|
|||||||
|
Model export
|
||||||
|
============
|
||||||
|
|
||||||
|
In this section, we describe various ways to export models.
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
.. toctree::
|
||||||
|
|
||||||
|
export-model-state-dict
|
||||||
|
export-with-torch-jit-trace
|
||||||
|
export-with-torch-jit-script
|
||||||
|
export-onnx
|
||||||
|
export-ncnn
|
@ -422,7 +422,7 @@ The information of the test sound files is listed below:
|
|||||||
|
|
||||||
.. code-block:: bash
|
.. code-block:: bash
|
||||||
|
|
||||||
$ soxi tmp/icefall_asr_aishell_conformer_ctc/test_wavs/*.wav
|
$ soxi tmp/icefall_asr_aishell_conformer_ctc/test_waves/*.wav
|
||||||
|
|
||||||
Input File : 'tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0121.wav'
|
Input File : 'tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0121.wav'
|
||||||
Channels : 1
|
Channels : 1
|
||||||
@ -485,9 +485,9 @@ The command to run CTC decoding is:
|
|||||||
--checkpoint ./tmp/icefall_asr_aishell_conformer_ctc/exp/pretrained.pt \
|
--checkpoint ./tmp/icefall_asr_aishell_conformer_ctc/exp/pretrained.pt \
|
||||||
--tokens-file ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/tokens.txt \
|
--tokens-file ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/tokens.txt \
|
||||||
--method ctc-decoding \
|
--method ctc-decoding \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0121.wav \
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0121.wav \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0122.wav \
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0122.wav \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0123.wav
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0123.wav
|
||||||
|
|
||||||
The output is given below:
|
The output is given below:
|
||||||
|
|
||||||
@ -529,9 +529,9 @@ The command to run HLG decoding is:
|
|||||||
--words-file ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/words.txt \
|
--words-file ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/words.txt \
|
||||||
--HLG ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/HLG.pt \
|
--HLG ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/HLG.pt \
|
||||||
--method 1best \
|
--method 1best \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0121.wav \
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0121.wav \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0122.wav \
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0122.wav \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0123.wav
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0123.wav
|
||||||
|
|
||||||
The output is given below:
|
The output is given below:
|
||||||
|
|
||||||
@ -575,9 +575,9 @@ The command to run HLG decoding + attention decoder rescoring is:
|
|||||||
--words-file ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/words.txt \
|
--words-file ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/words.txt \
|
||||||
--HLG ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/HLG.pt \
|
--HLG ./tmp/icefall_asr_aishell_conformer_ctc/data/lang_char/HLG.pt \
|
||||||
--method attention-decoder \
|
--method attention-decoder \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0121.wav \
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0121.wav \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0122.wav \
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0122.wav \
|
||||||
./tmp/icefall_asr_aishell_conformer_ctc/test_wavs/BAC009S0764W0123.wav
|
./tmp/icefall_asr_aishell_conformer_ctc/test_waves/BAC009S0764W0123.wav
|
||||||
|
|
||||||
The output is below:
|
The output is below:
|
||||||
|
|
Before Width: | Height: | Size: 334 KiB After Width: | Height: | Size: 334 KiB |
Before Width: | Height: | Size: 426 KiB After Width: | Height: | Size: 426 KiB |
Before Width: | Height: | Size: 441 KiB After Width: | Height: | Size: 441 KiB |
@ -19,4 +19,3 @@ It can be downloaded from `<https://www.openslr.org/33/>`_
|
|||||||
tdnn_lstm_ctc
|
tdnn_lstm_ctc
|
||||||
conformer_ctc
|
conformer_ctc
|
||||||
stateless_transducer
|
stateless_transducer
|
||||||
|
|
@ -402,7 +402,7 @@ The information of the test sound files is listed below:
|
|||||||
|
|
||||||
.. code-block:: bash
|
.. code-block:: bash
|
||||||
|
|
||||||
$ soxi tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_wavs/*.wav
|
$ soxi tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_waves/*.wav
|
||||||
|
|
||||||
Input File : 'tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_waves/BAC009S0764W0121.wav'
|
Input File : 'tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_waves/BAC009S0764W0121.wav'
|
||||||
Channels : 1
|
Channels : 1
|
||||||
@ -461,9 +461,9 @@ The command to run HLG decoding is:
|
|||||||
--words-file ./tmp/icefall_asr_aishell_tdnn_lstm_ctc/data/lang_phone/words.txt \
|
--words-file ./tmp/icefall_asr_aishell_tdnn_lstm_ctc/data/lang_phone/words.txt \
|
||||||
--HLG ./tmp/icefall_asr_aishell_tdnn_lstm_ctc/data/lang_phone/HLG.pt \
|
--HLG ./tmp/icefall_asr_aishell_tdnn_lstm_ctc/data/lang_phone/HLG.pt \
|
||||||
--method 1best \
|
--method 1best \
|
||||||
./tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_wavs/BAC009S0764W0121.wav \
|
./tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_waves/BAC009S0764W0121.wav \
|
||||||
./tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_wavs/BAC009S0764W0122.wav \
|
./tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_waves/BAC009S0764W0122.wav \
|
||||||
./tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_wavs/BAC009S0764W0123.wav
|
./tmp/icefall_asr_aishell_tdnn_lstm_ctc/test_waves/BAC009S0764W0123.wav
|
||||||
|
|
||||||
The output is given below:
|
The output is given below:
|
||||||
|
|
||||||
@ -498,7 +498,7 @@ We do provide a colab notebook for this recipe showing how to use a pre-trained
|
|||||||
|aishell asr conformer ctc colab notebook|
|
|aishell asr conformer ctc colab notebook|
|
||||||
|
|
||||||
.. |aishell asr conformer ctc colab notebook| image:: https://colab.research.google.com/assets/colab-badge.svg
|
.. |aishell asr conformer ctc colab notebook| image:: https://colab.research.google.com/assets/colab-badge.svg
|
||||||
:target: https://colab.research.google.com/drive/1qULaGvXq7PCu_P61oubfz9b53JzY4H3z
|
:target: https://colab.research.google.com/drive/1jbyzYq3ytm6j2nlEt-diQm-6QVWyDDEa?usp=sharing
|
||||||
|
|
||||||
**Congratulations!** You have finished the aishell ASR recipe with
|
**Congratulations!** You have finished the aishell ASR recipe with
|
||||||
TDNN-LSTM CTC models in ``icefall``.
|
TDNN-LSTM CTC models in ``icefall``.
|
10
docs/source/recipes/Non-streaming-ASR/index.rst
Normal file
@ -0,0 +1,10 @@
|
|||||||
|
Non Streaming ASR
|
||||||
|
=================
|
||||||
|
|
||||||
|
.. toctree::
|
||||||
|
:maxdepth: 2
|
||||||
|
|
||||||
|
aishell/index
|
||||||
|
librispeech/index
|
||||||
|
timit/index
|
||||||
|
yesno/index
|
Before Width: | Height: | Size: 422 KiB After Width: | Height: | Size: 422 KiB |
After Width: | Height: | Size: 554 KiB |
11
docs/source/recipes/Non-streaming-ASR/librispeech/index.rst
Normal file
@ -0,0 +1,11 @@
|
|||||||
|
LibriSpeech
|
||||||
|
===========
|
||||||
|
|
||||||
|
.. toctree::
|
||||||
|
:maxdepth: 1
|
||||||
|
|
||||||
|
tdnn_lstm_ctc
|
||||||
|
conformer_ctc
|
||||||
|
pruned_transducer_stateless
|
||||||
|
lstm_pruned_stateless_transducer
|
||||||
|
zipformer_mmi
|
@ -0,0 +1,545 @@
|
|||||||
|
Pruned transducer statelessX
|
||||||
|
============================
|
||||||
|
|
||||||
|
This tutorial shows you how to run a conformer transducer model
|
||||||
|
with the `LibriSpeech <https://www.openslr.org/12>`_ dataset.
|
||||||
|
|
||||||
|
.. Note::
|
||||||
|
|
||||||
|
The tutorial is suitable for `pruned_transducer_stateless <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless>`_,
|
||||||
|
`pruned_transducer_stateless2 <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless2>`_,
|
||||||
|
`pruned_transducer_stateless4 <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless4>`_,
|
||||||
|
`pruned_transducer_stateless5 <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless5>`_,
|
||||||
|
We will take pruned_transducer_stateless4 as an example in this tutorial.
|
||||||
|
|
||||||
|
.. HINT::
|
||||||
|
|
||||||
|
We assume you have read the page :ref:`install icefall` and have setup
|
||||||
|
the environment for ``icefall``.
|
||||||
|
|
||||||
|
.. HINT::
|
||||||
|
|
||||||
|
We recommend you to use a GPU or several GPUs to run this recipe.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
Please scroll down to the bottom of this page to find download links
|
||||||
|
for pretrained models if you don't want to train a model from scratch.
|
||||||
|
|
||||||
|
|
||||||
|
We use pruned RNN-T to compute the loss.
|
||||||
|
|
||||||
|
.. note::
|
||||||
|
|
||||||
|
You can find the paper about pruned RNN-T at the following address:
|
||||||
|
|
||||||
|
`<https://arxiv.org/abs/2206.13236>`_
|
||||||
|
|
||||||
|
The transducer model consists of 3 parts:
|
||||||
|
|
||||||
|
- Encoder, a.k.a, the transcription network. We use a Conformer model (the reworked version by Daniel Povey)
|
||||||
|
- Decoder, a.k.a, the prediction network. We use a stateless model consisting of
|
||||||
|
``nn.Embedding`` and ``nn.Conv1d``
|
||||||
|
- Joiner, a.k.a, the joint network.
|
||||||
|
|
||||||
|
.. caution::
|
||||||
|
|
||||||
|
Contrary to the conventional RNN-T models, we use a stateless decoder.
|
||||||
|
That is, it has no recurrent connections.
|
||||||
|
|
||||||
|
|
||||||
|
Data preparation
|
||||||
|
----------------
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
The data preparation is the same as other recipes on LibriSpeech dataset,
|
||||||
|
if you have finished this step, you can skip to ``Training`` directly.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./prepare.sh
|
||||||
|
|
||||||
|
The script ``./prepare.sh`` handles the data preparation for you, **automagically**.
|
||||||
|
All you need to do is to run it.
|
||||||
|
|
||||||
|
The data preparation contains several stages, you can use the following two
|
||||||
|
options:
|
||||||
|
|
||||||
|
- ``--stage``
|
||||||
|
- ``--stop-stage``
|
||||||
|
|
||||||
|
to control which stage(s) should be run. By default, all stages are executed.
|
||||||
|
|
||||||
|
|
||||||
|
For example,
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./prepare.sh --stage 0 --stop-stage 0
|
||||||
|
|
||||||
|
means to run only stage 0.
|
||||||
|
|
||||||
|
To run stage 2 to stage 5, use:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ ./prepare.sh --stage 2 --stop-stage 5
|
||||||
|
|
||||||
|
.. HINT::
|
||||||
|
|
||||||
|
If you have pre-downloaded the `LibriSpeech <https://www.openslr.org/12>`_
|
||||||
|
dataset and the `musan <http://www.openslr.org/17/>`_ dataset, say,
|
||||||
|
they are saved in ``/tmp/LibriSpeech`` and ``/tmp/musan``, you can modify
|
||||||
|
the ``dl_dir`` variable in ``./prepare.sh`` to point to ``/tmp`` so that
|
||||||
|
``./prepare.sh`` won't re-download them.
|
||||||
|
|
||||||
|
.. NOTE::
|
||||||
|
|
||||||
|
All generated files by ``./prepare.sh``, e.g., features, lexicon, etc,
|
||||||
|
are saved in ``./data`` directory.
|
||||||
|
|
||||||
|
We provide the following YouTube video showing how to run ``./prepare.sh``.
|
||||||
|
|
||||||
|
.. note::
|
||||||
|
|
||||||
|
To get the latest news of `next-gen Kaldi <https://github.com/k2-fsa>`_, please subscribe
|
||||||
|
the following YouTube channel by `Nadira Povey <https://www.youtube.com/channel/UC_VaumpkmINz1pNkFXAN9mw>`_:
|
||||||
|
|
||||||
|
`<https://www.youtube.com/channel/UC_VaumpkmINz1pNkFXAN9mw>`_
|
||||||
|
|
||||||
|
.. youtube:: ofEIoJL-mGM
|
||||||
|
|
||||||
|
|
||||||
|
Training
|
||||||
|
--------
|
||||||
|
|
||||||
|
Configurable options
|
||||||
|
~~~~~~~~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./pruned_transducer_stateless4/train.py --help
|
||||||
|
|
||||||
|
|
||||||
|
shows you the training options that can be passed from the commandline.
|
||||||
|
The following options are used quite often:
|
||||||
|
|
||||||
|
- ``--exp-dir``
|
||||||
|
|
||||||
|
The directory to save checkpoints, training logs and tensorboard.
|
||||||
|
|
||||||
|
- ``--full-libri``
|
||||||
|
|
||||||
|
If it's True, the training part uses all the training data, i.e.,
|
||||||
|
960 hours. Otherwise, the training part uses only the subset
|
||||||
|
``train-clean-100``, which has 100 hours of training data.
|
||||||
|
|
||||||
|
.. CAUTION::
|
||||||
|
The training set is perturbed by speed with two factors: 0.9 and 1.1.
|
||||||
|
If ``--full-libri`` is True, each epoch actually processes
|
||||||
|
``3x960 == 2880`` hours of data.
|
||||||
|
|
||||||
|
- ``--num-epochs``
|
||||||
|
|
||||||
|
It is the number of epochs to train. For instance,
|
||||||
|
``./pruned_transducer_stateless4/train.py --num-epochs 30`` trains for 30 epochs
|
||||||
|
and generates ``epoch-1.pt``, ``epoch-2.pt``, ..., ``epoch-30.pt``
|
||||||
|
in the folder ``./pruned_transducer_stateless4/exp``.
|
||||||
|
|
||||||
|
- ``--start-epoch``
|
||||||
|
|
||||||
|
It's used to resume training.
|
||||||
|
``./pruned_transducer_stateless4/train.py --start-epoch 10`` loads the
|
||||||
|
checkpoint ``./pruned_transducer_stateless4/exp/epoch-9.pt`` and starts
|
||||||
|
training from epoch 10, based on the state from epoch 9.
|
||||||
|
|
||||||
|
- ``--world-size``
|
||||||
|
|
||||||
|
It is used for multi-GPU single-machine DDP training.
|
||||||
|
|
||||||
|
- (a) If it is 1, then no DDP training is used.
|
||||||
|
|
||||||
|
- (b) If it is 2, then GPU 0 and GPU 1 are used for DDP training.
|
||||||
|
|
||||||
|
The following shows some use cases with it.
|
||||||
|
|
||||||
|
**Use case 1**: You have 4 GPUs, but you only want to use GPU 0 and
|
||||||
|
GPU 2 for training. You can do the following:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ export CUDA_VISIBLE_DEVICES="0,2"
|
||||||
|
$ ./pruned_transducer_stateless4/train.py --world-size 2
|
||||||
|
|
||||||
|
**Use case 2**: You have 4 GPUs and you want to use all of them
|
||||||
|
for training. You can do the following:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./pruned_transducer_stateless4/train.py --world-size 4
|
||||||
|
|
||||||
|
**Use case 3**: You have 4 GPUs but you only want to use GPU 3
|
||||||
|
for training. You can do the following:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ export CUDA_VISIBLE_DEVICES="3"
|
||||||
|
$ ./pruned_transducer_stateless4/train.py --world-size 1
|
||||||
|
|
||||||
|
.. caution::
|
||||||
|
|
||||||
|
Only multi-GPU single-machine DDP training is implemented at present.
|
||||||
|
Multi-GPU multi-machine DDP training will be added later.
|
||||||
|
|
||||||
|
- ``--max-duration``
|
||||||
|
|
||||||
|
It specifies the number of seconds over all utterances in a
|
||||||
|
batch, before **padding**.
|
||||||
|
If you encounter CUDA OOM, please reduce it.
|
||||||
|
|
||||||
|
.. HINT::
|
||||||
|
|
||||||
|
Due to padding, the number of seconds of all utterances in a
|
||||||
|
batch will usually be larger than ``--max-duration``.
|
||||||
|
|
||||||
|
A larger value for ``--max-duration`` may cause OOM during training,
|
||||||
|
while a smaller value may increase the training time. You have to
|
||||||
|
tune it.
|
||||||
|
|
||||||
|
- ``--use-fp16``
|
||||||
|
|
||||||
|
If it is True, the model will train with half precision, from our experiment
|
||||||
|
results, by using half precision you can train with two times larger ``--max-duration``
|
||||||
|
so as to get almost 2X speed up.
|
||||||
|
|
||||||
|
|
||||||
|
Pre-configured options
|
||||||
|
~~~~~~~~~~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
There are some training options, e.g., number of encoder layers,
|
||||||
|
encoder dimension, decoder dimension, number of warmup steps etc,
|
||||||
|
that are not passed from the commandline.
|
||||||
|
They are pre-configured by the function ``get_params()`` in
|
||||||
|
`pruned_transducer_stateless4/train.py <https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/pruned_transducer_stateless4/train.py>`_
|
||||||
|
|
||||||
|
You don't need to change these pre-configured parameters. If you really need to change
|
||||||
|
them, please modify ``./pruned_transducer_stateless4/train.py`` directly.
|
||||||
|
|
||||||
|
|
||||||
|
.. NOTE::
|
||||||
|
|
||||||
|
The options for `pruned_transducer_stateless5 <https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/pruned_transducer_stateless5/train.py>`_ are a little different from
|
||||||
|
other recipes. It allows you to configure ``--num-encoder-layers``, ``--dim-feedforward``, ``--nhead``, ``--encoder-dim``, ``--decoder-dim``, ``--joiner-dim`` from commandline, so that you can train models with different size with pruned_transducer_stateless5.
|
||||||
|
|
||||||
|
|
||||||
|
Training logs
|
||||||
|
~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
Training logs and checkpoints are saved in ``--exp-dir`` (e.g. ``pruned_transducer_stateless4/exp``.
|
||||||
|
You will find the following files in that directory:
|
||||||
|
|
||||||
|
- ``epoch-1.pt``, ``epoch-2.pt``, ...
|
||||||
|
|
||||||
|
These are checkpoint files saved at the end of each epoch, containing model
|
||||||
|
``state_dict`` and optimizer ``state_dict``.
|
||||||
|
To resume training from some checkpoint, say ``epoch-10.pt``, you can use:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ ./pruned_transducer_stateless4/train.py --start-epoch 11
|
||||||
|
|
||||||
|
- ``checkpoint-436000.pt``, ``checkpoint-438000.pt``, ...
|
||||||
|
|
||||||
|
These are checkpoint files saved every ``--save-every-n`` batches,
|
||||||
|
containing model ``state_dict`` and optimizer ``state_dict``.
|
||||||
|
To resume training from some checkpoint, say ``checkpoint-436000``, you can use:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ ./pruned_transducer_stateless4/train.py --start-batch 436000
|
||||||
|
|
||||||
|
- ``tensorboard/``
|
||||||
|
|
||||||
|
This folder contains tensorBoard logs. Training loss, validation loss, learning
|
||||||
|
rate, etc, are recorded in these logs. You can visualize them by:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd pruned_transducer_stateless4/exp/tensorboard
|
||||||
|
$ tensorboard dev upload --logdir . --description "pruned transducer training for LibriSpeech with icefall"
|
||||||
|
|
||||||
|
It will print something like below:
|
||||||
|
|
||||||
|
.. code-block::
|
||||||
|
|
||||||
|
TensorFlow installation not found - running with reduced feature set.
|
||||||
|
Upload started and will continue reading any new data as it's added to the logdir.
|
||||||
|
|
||||||
|
To stop uploading, press Ctrl-C.
|
||||||
|
|
||||||
|
New experiment created. View your TensorBoard at: https://tensorboard.dev/experiment/QOGSPBgsR8KzcRMmie9JGw/
|
||||||
|
|
||||||
|
[2022-11-20T15:50:50] Started scanning logdir.
|
||||||
|
Uploading 4468 scalars...
|
||||||
|
[2022-11-20T15:53:02] Total uploaded: 210171 scalars, 0 tensors, 0 binary objects
|
||||||
|
Listening for new data in logdir...
|
||||||
|
|
||||||
|
Note there is a URL in the above output. Click it and you will see
|
||||||
|
the following screenshot:
|
||||||
|
|
||||||
|
.. figure:: images/librispeech-pruned-transducer-tensorboard-log.jpg
|
||||||
|
:width: 600
|
||||||
|
:alt: TensorBoard screenshot
|
||||||
|
:align: center
|
||||||
|
:target: https://tensorboard.dev/experiment/QOGSPBgsR8KzcRMmie9JGw/
|
||||||
|
|
||||||
|
TensorBoard screenshot.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
If you don't have access to google, you can use the following command
|
||||||
|
to view the tensorboard log locally:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd pruned_transducer_stateless4/exp/tensorboard
|
||||||
|
tensorboard --logdir . --port 6008
|
||||||
|
|
||||||
|
It will print the following message:
|
||||||
|
|
||||||
|
.. code-block::
|
||||||
|
|
||||||
|
Serving TensorBoard on localhost; to expose to the network, use a proxy or pass --bind_all
|
||||||
|
TensorBoard 2.8.0 at http://localhost:6008/ (Press CTRL+C to quit)
|
||||||
|
|
||||||
|
Now start your browser and go to `<http://localhost:6008>`_ to view the tensorboard
|
||||||
|
logs.
|
||||||
|
|
||||||
|
|
||||||
|
- ``log/log-train-xxxx``
|
||||||
|
|
||||||
|
It is the detailed training log in text format, same as the one
|
||||||
|
you saw printed to the console during training.
|
||||||
|
|
||||||
|
Usage example
|
||||||
|
~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
You can use the following command to start the training using 6 GPUs:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5"
|
||||||
|
./pruned_transducer_stateless4/train.py \
|
||||||
|
--world-size 6 \
|
||||||
|
--num-epochs 30 \
|
||||||
|
--start-epoch 1 \
|
||||||
|
--exp-dir pruned_transducer_stateless4/exp \
|
||||||
|
--full-libri 1 \
|
||||||
|
--max-duration 300
|
||||||
|
|
||||||
|
|
||||||
|
Decoding
|
||||||
|
--------
|
||||||
|
|
||||||
|
The decoding part uses checkpoints saved by the training part, so you have
|
||||||
|
to run the training part first.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
There are two kinds of checkpoints:
|
||||||
|
|
||||||
|
- (1) ``epoch-1.pt``, ``epoch-2.pt``, ..., which are saved at the end
|
||||||
|
of each epoch. You can pass ``--epoch`` to
|
||||||
|
``pruned_transducer_stateless4/decode.py`` to use them.
|
||||||
|
|
||||||
|
- (2) ``checkpoints-436000.pt``, ``epoch-438000.pt``, ..., which are saved
|
||||||
|
every ``--save-every-n`` batches. You can pass ``--iter`` to
|
||||||
|
``pruned_transducer_stateless4/decode.py`` to use them.
|
||||||
|
|
||||||
|
We suggest that you try both types of checkpoints and choose the one
|
||||||
|
that produces the lowest WERs.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./pruned_transducer_stateless4/decode.py --help
|
||||||
|
|
||||||
|
shows the options for decoding.
|
||||||
|
|
||||||
|
The following shows two examples (for two types of checkpoints):
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
for m in greedy_search fast_beam_search modified_beam_search; do
|
||||||
|
for epoch in 25 20; do
|
||||||
|
for avg in 7 5 3 1; do
|
||||||
|
./pruned_transducer_stateless4/decode.py \
|
||||||
|
--epoch $epoch \
|
||||||
|
--avg $avg \
|
||||||
|
--exp-dir pruned_transducer_stateless4/exp \
|
||||||
|
--max-duration 600 \
|
||||||
|
--decoding-method $m
|
||||||
|
done
|
||||||
|
done
|
||||||
|
done
|
||||||
|
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
for m in greedy_search fast_beam_search modified_beam_search; do
|
||||||
|
for iter in 474000; do
|
||||||
|
for avg in 8 10 12 14 16 18; do
|
||||||
|
./pruned_transducer_stateless4/decode.py \
|
||||||
|
--iter $iter \
|
||||||
|
--avg $avg \
|
||||||
|
--exp-dir pruned_transducer_stateless4/exp \
|
||||||
|
--max-duration 600 \
|
||||||
|
--decoding-method $m
|
||||||
|
done
|
||||||
|
done
|
||||||
|
done
|
||||||
|
|
||||||
|
|
||||||
|
.. Note::
|
||||||
|
|
||||||
|
Supporting decoding methods are as follows:
|
||||||
|
|
||||||
|
- ``greedy_search`` : It takes the symbol with largest posterior probability
|
||||||
|
of each frame as the decoding result.
|
||||||
|
|
||||||
|
- ``beam_search`` : It implements Algorithm 1 in https://arxiv.org/pdf/1211.3711.pdf and
|
||||||
|
`espnet/nets/beam_search_transducer.py <https://github.com/espnet/espnet/blob/master/espnet/nets/beam_search_transducer.py#L247>`_
|
||||||
|
is used as a reference. Basicly, it keeps topk states for each frame, and expands the kept states with their own contexts to
|
||||||
|
next frame.
|
||||||
|
|
||||||
|
- ``modified_beam_search`` : It implements the same algorithm as ``beam_search`` above, but it
|
||||||
|
runs in batch mode with ``--max-sym-per-frame=1`` being hardcoded.
|
||||||
|
|
||||||
|
- ``fast_beam_search`` : It implements graph composition between the output ``log_probs`` and
|
||||||
|
given ``FSAs``. It is hard to describe the details in several lines of texts, you can read
|
||||||
|
our paper in https://arxiv.org/pdf/2211.00484.pdf or our `rnnt decode code in k2 <https://github.com/k2-fsa/k2/blob/master/k2/csrc/rnnt_decode.h>`_. ``fast_beam_search`` can decode with ``FSAs`` on GPU efficiently.
|
||||||
|
|
||||||
|
- ``fast_beam_search_LG`` : The same as ``fast_beam_search`` above, ``fast_beam_search`` uses
|
||||||
|
an trivial graph that has only one state, while ``fast_beam_search_LG`` uses an LG graph
|
||||||
|
(with N-gram LM).
|
||||||
|
|
||||||
|
- ``fast_beam_search_nbest`` : It produces the decoding results as follows:
|
||||||
|
|
||||||
|
- (1) Use ``fast_beam_search`` to get a lattice
|
||||||
|
- (2) Select ``num_paths`` paths from the lattice using ``k2.random_paths()``
|
||||||
|
- (3) Unique the selected paths
|
||||||
|
- (4) Intersect the selected paths with the lattice and compute the
|
||||||
|
shortest path from the intersection result
|
||||||
|
- (5) The path with the largest score is used as the decoding output.
|
||||||
|
|
||||||
|
- ``fast_beam_search_nbest_LG`` : It implements same logic as ``fast_beam_search_nbest``, the
|
||||||
|
only difference is that it uses ``fast_beam_search_LG`` to generate the lattice.
|
||||||
|
|
||||||
|
|
||||||
|
Export Model
|
||||||
|
------------
|
||||||
|
|
||||||
|
`pruned_transducer_stateless4/export.py <https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/pruned_transducer_stateless4/export.py>`_ supports exporting checkpoints from ``pruned_transducer_stateless4/exp`` in the following ways.
|
||||||
|
|
||||||
|
Export ``model.state_dict()``
|
||||||
|
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
Checkpoints saved by ``pruned_transducer_stateless4/train.py`` also include
|
||||||
|
``optimizer.state_dict()``. It is useful for resuming training. But after training,
|
||||||
|
we are interested only in ``model.state_dict()``. You can use the following
|
||||||
|
command to extract ``model.state_dict()``.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
# Assume that --epoch 25 --avg 3 produces the smallest WER
|
||||||
|
# (You can get such information after running ./pruned_transducer_stateless4/decode.py)
|
||||||
|
|
||||||
|
epoch=25
|
||||||
|
avg=3
|
||||||
|
|
||||||
|
./pruned_transducer_stateless4/export.py \
|
||||||
|
--exp-dir ./pruned_transducer_stateless4/exp \
|
||||||
|
--bpe-model data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch $epoch \
|
||||||
|
--avg $avg
|
||||||
|
|
||||||
|
It will generate a file ``./pruned_transducer_stateless4/exp/pretrained.pt``.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
To use the generated ``pretrained.pt`` for ``pruned_transducer_stateless4/decode.py``,
|
||||||
|
you can run:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd pruned_transducer_stateless4/exp
|
||||||
|
ln -s pretrained.pt epoch-999.pt
|
||||||
|
|
||||||
|
And then pass ``--epoch 999 --avg 1 --use-averaged-model 0`` to
|
||||||
|
``./pruned_transducer_stateless4/decode.py``.
|
||||||
|
|
||||||
|
To use the exported model with ``./pruned_transducer_stateless4/pretrained.py``, you
|
||||||
|
can run:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
./pruned_transducer_stateless4/pretrained.py \
|
||||||
|
--checkpoint ./pruned_transducer_stateless4/exp/pretrained.pt \
|
||||||
|
--bpe-model ./data/lang_bpe_500/bpe.model \
|
||||||
|
--method greedy_search \
|
||||||
|
/path/to/foo.wav \
|
||||||
|
/path/to/bar.wav
|
||||||
|
|
||||||
|
|
||||||
|
Export model using ``torch.jit.script()``
|
||||||
|
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
./pruned_transducer_stateless4/export.py \
|
||||||
|
--exp-dir ./pruned_transducer_stateless4/exp \
|
||||||
|
--bpe-model data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 25 \
|
||||||
|
--avg 3 \
|
||||||
|
--jit 1
|
||||||
|
|
||||||
|
It will generate a file ``cpu_jit.pt`` in the given ``exp_dir``. You can later
|
||||||
|
load it by ``torch.jit.load("cpu_jit.pt")``.
|
||||||
|
|
||||||
|
Note ``cpu`` in the name ``cpu_jit.pt`` means the parameters when loaded into Python
|
||||||
|
are on CPU. You can use ``to("cuda")`` to move them to a CUDA device.
|
||||||
|
|
||||||
|
.. NOTE::
|
||||||
|
|
||||||
|
You will need this ``cpu_jit.pt`` when deploying with Sherpa framework.
|
||||||
|
|
||||||
|
|
||||||
|
Download pretrained models
|
||||||
|
--------------------------
|
||||||
|
|
||||||
|
If you don't want to train from scratch, you can download the pretrained models
|
||||||
|
by visiting the following links:
|
||||||
|
|
||||||
|
- `pruned_transducer_stateless <https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless-2022-03-12>`_
|
||||||
|
|
||||||
|
- `pruned_transducer_stateless2 <https://huggingface.co/csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless2-2022-04-29>`_
|
||||||
|
|
||||||
|
- `pruned_transducer_stateless4 <https://huggingface.co/Zengwei/icefall-asr-librispeech-pruned-transducer-stateless4-2022-06-03>`_
|
||||||
|
|
||||||
|
- `pruned_transducer_stateless5 <https://huggingface.co/Zengwei/icefall-asr-librispeech-pruned-transducer-stateless5-2022-07-07>`_
|
||||||
|
|
||||||
|
See `<https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/RESULTS.md>`_
|
||||||
|
for the details of the above pretrained models
|
||||||
|
|
||||||
|
|
||||||
|
Deploy with Sherpa
|
||||||
|
------------------
|
||||||
|
|
||||||
|
Please see `<https://k2-fsa.github.io/sherpa/python/offline_asr/conformer/librispeech.html#>`_
|
||||||
|
for how to deploy the models in ``sherpa``.
|
@ -398,7 +398,7 @@ We provide a colab notebook for decoding with pre-trained model.
|
|||||||
|librispeech tdnn_lstm_ctc colab notebook|
|
|librispeech tdnn_lstm_ctc colab notebook|
|
||||||
|
|
||||||
.. |librispeech tdnn_lstm_ctc colab notebook| image:: https://colab.research.google.com/assets/colab-badge.svg
|
.. |librispeech tdnn_lstm_ctc colab notebook| image:: https://colab.research.google.com/assets/colab-badge.svg
|
||||||
:target: https://colab.research.google.com/drive/1kNmDXNMwREi0rZGAOIAOJo93REBuOTcd
|
:target: https://colab.research.google.com/drive/1-iSfQMp2So-We_Uu49N4AAcMInB72u9z?usp=sharing
|
||||||
|
|
||||||
|
|
||||||
**Congratulations!** You have finished the TDNN-LSTM-CTC recipe on librispeech in ``icefall``.
|
**Congratulations!** You have finished the TDNN-LSTM-CTC recipe on librispeech in ``icefall``.
|
@ -0,0 +1,422 @@
|
|||||||
|
Zipformer MMI
|
||||||
|
===============
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
Please scroll down to the bottom of this page to find download links
|
||||||
|
for pretrained models if you don't want to train a model from scratch.
|
||||||
|
|
||||||
|
|
||||||
|
This tutorial shows you how to train an Zipformer MMI model
|
||||||
|
with the `LibriSpeech <https://www.openslr.org/12>`_ dataset.
|
||||||
|
|
||||||
|
We use LF-MMI to compute the loss.
|
||||||
|
|
||||||
|
.. note::
|
||||||
|
|
||||||
|
You can find the document about LF-MMI training at the following address:
|
||||||
|
|
||||||
|
`<https://github.com/k2-fsa/next-gen-kaldi-wechat/blob/master/pdf/LF-MMI-training-and-decoding-in-k2-Part-I.pdf>`_
|
||||||
|
|
||||||
|
|
||||||
|
Data preparation
|
||||||
|
----------------
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./prepare.sh
|
||||||
|
|
||||||
|
The script ``./prepare.sh`` handles the data preparation for you, **automagically**.
|
||||||
|
All you need to do is to run it.
|
||||||
|
|
||||||
|
.. note::
|
||||||
|
|
||||||
|
We encourage you to read ``./prepare.sh``.
|
||||||
|
|
||||||
|
The data preparation contains several stages. You can use the following two
|
||||||
|
options:
|
||||||
|
|
||||||
|
- ``--stage``
|
||||||
|
- ``--stop-stage``
|
||||||
|
|
||||||
|
to control which stage(s) should be run. By default, all stages are executed.
|
||||||
|
|
||||||
|
|
||||||
|
For example,
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./prepare.sh --stage 0 --stop-stage 0
|
||||||
|
|
||||||
|
means to run only stage 0.
|
||||||
|
|
||||||
|
To run stage 2 to stage 5, use:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ ./prepare.sh --stage 2 --stop-stage 5
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
If you have pre-downloaded the `LibriSpeech <https://www.openslr.org/12>`_
|
||||||
|
dataset and the `musan <http://www.openslr.org/17/>`_ dataset, say,
|
||||||
|
they are saved in ``/tmp/LibriSpeech`` and ``/tmp/musan``, you can modify
|
||||||
|
the ``dl_dir`` variable in ``./prepare.sh`` to point to ``/tmp`` so that
|
||||||
|
``./prepare.sh`` won't re-download them.
|
||||||
|
|
||||||
|
.. note::
|
||||||
|
|
||||||
|
All generated files by ``./prepare.sh``, e.g., features, lexicon, etc,
|
||||||
|
are saved in ``./data`` directory.
|
||||||
|
|
||||||
|
We provide the following YouTube video showing how to run ``./prepare.sh``.
|
||||||
|
|
||||||
|
.. note::
|
||||||
|
|
||||||
|
To get the latest news of `next-gen Kaldi <https://github.com/k2-fsa>`_, please subscribe
|
||||||
|
the following YouTube channel by `Nadira Povey <https://www.youtube.com/channel/UC_VaumpkmINz1pNkFXAN9mw>`_:
|
||||||
|
|
||||||
|
`<https://www.youtube.com/channel/UC_VaumpkmINz1pNkFXAN9mw>`_
|
||||||
|
|
||||||
|
.. youtube:: ofEIoJL-mGM
|
||||||
|
|
||||||
|
Training
|
||||||
|
--------
|
||||||
|
|
||||||
|
For stability, it uses CTC loss for model warm-up and then switches to MMI loss.
|
||||||
|
|
||||||
|
Configurable options
|
||||||
|
~~~~~~~~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./zipformer_mmi/train.py --help
|
||||||
|
|
||||||
|
shows you the training options that can be passed from the commandline.
|
||||||
|
The following options are used quite often:
|
||||||
|
|
||||||
|
- ``--full-libri``
|
||||||
|
|
||||||
|
If it's True, the training part uses all the training data, i.e.,
|
||||||
|
960 hours. Otherwise, the training part uses only the subset
|
||||||
|
``train-clean-100``, which has 100 hours of training data.
|
||||||
|
|
||||||
|
.. CAUTION::
|
||||||
|
|
||||||
|
The training set is perturbed by speed with two factors: 0.9 and 1.1.
|
||||||
|
If ``--full-libri`` is True, each epoch actually processes
|
||||||
|
``3x960 == 2880`` hours of data.
|
||||||
|
|
||||||
|
- ``--num-epochs``
|
||||||
|
|
||||||
|
It is the number of epochs to train. For instance,
|
||||||
|
``./zipformer_mmi/train.py --num-epochs 30`` trains for 30 epochs
|
||||||
|
and generates ``epoch-1.pt``, ``epoch-2.pt``, ..., ``epoch-30.pt``
|
||||||
|
in the folder ``./zipformer_mmi/exp``.
|
||||||
|
|
||||||
|
- ``--start-epoch``
|
||||||
|
|
||||||
|
It's used to resume training.
|
||||||
|
``./zipformer_mmi/train.py --start-epoch 10`` loads the
|
||||||
|
checkpoint ``./zipformer_mmi/exp/epoch-9.pt`` and starts
|
||||||
|
training from epoch 10, based on the state from epoch 9.
|
||||||
|
|
||||||
|
- ``--world-size``
|
||||||
|
|
||||||
|
It is used for multi-GPU single-machine DDP training.
|
||||||
|
|
||||||
|
- (a) If it is 1, then no DDP training is used.
|
||||||
|
|
||||||
|
- (b) If it is 2, then GPU 0 and GPU 1 are used for DDP training.
|
||||||
|
|
||||||
|
The following shows some use cases with it.
|
||||||
|
|
||||||
|
**Use case 1**: You have 4 GPUs, but you only want to use GPU 0 and
|
||||||
|
GPU 2 for training. You can do the following:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ export CUDA_VISIBLE_DEVICES="0,2"
|
||||||
|
$ ./zipformer_mmi/train.py --world-size 2
|
||||||
|
|
||||||
|
**Use case 2**: You have 4 GPUs and you want to use all of them
|
||||||
|
for training. You can do the following:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./zipformer_mmi/train.py --world-size 4
|
||||||
|
|
||||||
|
**Use case 3**: You have 4 GPUs but you only want to use GPU 3
|
||||||
|
for training. You can do the following:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ export CUDA_VISIBLE_DEVICES="3"
|
||||||
|
$ ./zipformer_mmi/train.py --world-size 1
|
||||||
|
|
||||||
|
.. caution::
|
||||||
|
|
||||||
|
Only multi-GPU single-machine DDP training is implemented at present.
|
||||||
|
Multi-GPU multi-machine DDP training will be added later.
|
||||||
|
|
||||||
|
- ``--max-duration``
|
||||||
|
|
||||||
|
It specifies the number of seconds over all utterances in a
|
||||||
|
batch, before **padding**.
|
||||||
|
If you encounter CUDA OOM, please reduce it.
|
||||||
|
|
||||||
|
.. HINT::
|
||||||
|
|
||||||
|
Due to padding, the number of seconds of all utterances in a
|
||||||
|
batch will usually be larger than ``--max-duration``.
|
||||||
|
|
||||||
|
A larger value for ``--max-duration`` may cause OOM during training,
|
||||||
|
while a smaller value may increase the training time. You have to
|
||||||
|
tune it.
|
||||||
|
|
||||||
|
|
||||||
|
Pre-configured options
|
||||||
|
~~~~~~~~~~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
There are some training options, e.g., weight decay,
|
||||||
|
number of warmup steps, results dir, etc,
|
||||||
|
that are not passed from the commandline.
|
||||||
|
They are pre-configured by the function ``get_params()`` in
|
||||||
|
`zipformer_mmi/train.py <https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/zipformer_mmi/train.py>`_
|
||||||
|
|
||||||
|
You don't need to change these pre-configured parameters. If you really need to change
|
||||||
|
them, please modify ``./zipformer_mmi/train.py`` directly.
|
||||||
|
|
||||||
|
Training logs
|
||||||
|
~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
Training logs and checkpoints are saved in ``zipformer_mmi/exp``.
|
||||||
|
You will find the following files in that directory:
|
||||||
|
|
||||||
|
- ``epoch-1.pt``, ``epoch-2.pt``, ...
|
||||||
|
|
||||||
|
These are checkpoint files saved at the end of each epoch, containing model
|
||||||
|
``state_dict`` and optimizer ``state_dict``.
|
||||||
|
To resume training from some checkpoint, say ``epoch-10.pt``, you can use:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ ./zipformer_mmi/train.py --start-epoch 11
|
||||||
|
|
||||||
|
- ``checkpoint-436000.pt``, ``checkpoint-438000.pt``, ...
|
||||||
|
|
||||||
|
These are checkpoint files saved every ``--save-every-n`` batches,
|
||||||
|
containing model ``state_dict`` and optimizer ``state_dict``.
|
||||||
|
To resume training from some checkpoint, say ``checkpoint-436000``, you can use:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ ./zipformer_mmi/train.py --start-batch 436000
|
||||||
|
|
||||||
|
- ``tensorboard/``
|
||||||
|
|
||||||
|
This folder contains tensorBoard logs. Training loss, validation loss, learning
|
||||||
|
rate, etc, are recorded in these logs. You can visualize them by:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd zipformer_mmi/exp/tensorboard
|
||||||
|
$ tensorboard dev upload --logdir . --description "Zipformer MMI training for LibriSpeech with icefall"
|
||||||
|
|
||||||
|
It will print something like below:
|
||||||
|
|
||||||
|
.. code-block::
|
||||||
|
|
||||||
|
TensorFlow installation not found - running with reduced feature set.
|
||||||
|
Upload started and will continue reading any new data as it's added to the logdir.
|
||||||
|
|
||||||
|
To stop uploading, press Ctrl-C.
|
||||||
|
|
||||||
|
New experiment created. View your TensorBoard at: https://tensorboard.dev/experiment/xyOZUKpEQm62HBIlUD4uPA/
|
||||||
|
|
||||||
|
Note there is a URL in the above output. Click it and you will see
|
||||||
|
tensorboard.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
If you don't have access to google, you can use the following command
|
||||||
|
to view the tensorboard log locally:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd zipformer_mmi/exp/tensorboard
|
||||||
|
tensorboard --logdir . --port 6008
|
||||||
|
|
||||||
|
It will print the following message:
|
||||||
|
|
||||||
|
.. code-block::
|
||||||
|
|
||||||
|
Serving TensorBoard on localhost; to expose to the network, use a proxy or pass --bind_all
|
||||||
|
TensorBoard 2.8.0 at http://localhost:6008/ (Press CTRL+C to quit)
|
||||||
|
|
||||||
|
Now start your browser and go to `<http://localhost:6008>`_ to view the tensorboard
|
||||||
|
logs.
|
||||||
|
|
||||||
|
|
||||||
|
- ``log/log-train-xxxx``
|
||||||
|
|
||||||
|
It is the detailed training log in text format, same as the one
|
||||||
|
you saw printed to the console during training.
|
||||||
|
|
||||||
|
Usage example
|
||||||
|
~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
You can use the following command to start the training using 8 GPUs:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
export CUDA_VISIBLE_DEVICES="0,1,2,3"
|
||||||
|
./zipformer_mmi/train.py \
|
||||||
|
--world-size 4 \
|
||||||
|
--num-epochs 30 \
|
||||||
|
--start-epoch 1 \
|
||||||
|
--full-libri 1 \
|
||||||
|
--exp-dir zipformer_mmi/exp \
|
||||||
|
--max-duration 500 \
|
||||||
|
--use-fp16 1 \
|
||||||
|
--num-workers 2
|
||||||
|
|
||||||
|
Decoding
|
||||||
|
--------
|
||||||
|
|
||||||
|
The decoding part uses checkpoints saved by the training part, so you have
|
||||||
|
to run the training part first.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
There are two kinds of checkpoints:
|
||||||
|
|
||||||
|
- (1) ``epoch-1.pt``, ``epoch-2.pt``, ..., which are saved at the end
|
||||||
|
of each epoch. You can pass ``--epoch`` to
|
||||||
|
``zipformer_mmi/decode.py`` to use them.
|
||||||
|
|
||||||
|
- (2) ``checkpoints-436000.pt``, ``epoch-438000.pt``, ..., which are saved
|
||||||
|
every ``--save-every-n`` batches. You can pass ``--iter`` to
|
||||||
|
``zipformer_mmi/decode.py`` to use them.
|
||||||
|
|
||||||
|
We suggest that you try both types of checkpoints and choose the one
|
||||||
|
that produces the lowest WERs.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
$ cd egs/librispeech/ASR
|
||||||
|
$ ./zipformer_mmi/decode.py --help
|
||||||
|
|
||||||
|
shows the options for decoding.
|
||||||
|
|
||||||
|
The following shows the example using ``epoch-*.pt``:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
for m in nbest nbest-rescoring-LG nbest-rescoring-3-gram nbest-rescoring-4-gram; do
|
||||||
|
./zipformer_mmi/decode.py \
|
||||||
|
--epoch 30 \
|
||||||
|
--avg 10 \
|
||||||
|
--exp-dir ./zipformer_mmi/exp/ \
|
||||||
|
--max-duration 100 \
|
||||||
|
--lang-dir data/lang_bpe_500 \
|
||||||
|
--nbest-scale 1.2 \
|
||||||
|
--hp-scale 1.0 \
|
||||||
|
--decoding-method $m
|
||||||
|
done
|
||||||
|
|
||||||
|
|
||||||
|
Export models
|
||||||
|
-------------
|
||||||
|
|
||||||
|
`zipformer_mmi/export.py <https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/zipformer_mmi/export.py>`_ supports exporting checkpoints from ``zipformer_mmi/exp`` in the following ways.
|
||||||
|
|
||||||
|
Export ``model.state_dict()``
|
||||||
|
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
Checkpoints saved by ``zipformer_mmi/train.py`` also include
|
||||||
|
``optimizer.state_dict()``. It is useful for resuming training. But after training,
|
||||||
|
we are interested only in ``model.state_dict()``. You can use the following
|
||||||
|
command to extract ``model.state_dict()``.
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
./zipformer_mmi/export.py \
|
||||||
|
--exp-dir ./zipformer_mmi/exp \
|
||||||
|
--bpe-model data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 30 \
|
||||||
|
--avg 9 \
|
||||||
|
--jit 0
|
||||||
|
|
||||||
|
It will generate a file ``./zipformer_mmi/exp/pretrained.pt``.
|
||||||
|
|
||||||
|
.. hint::
|
||||||
|
|
||||||
|
To use the generated ``pretrained.pt`` for ``zipformer_mmi/decode.py``,
|
||||||
|
you can run:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
cd zipformer_mmi/exp
|
||||||
|
ln -s pretrained epoch-9999.pt
|
||||||
|
|
||||||
|
And then pass ``--epoch 9999 --avg 1 --use-averaged-model 0`` to
|
||||||
|
``./zipformer_mmi/decode.py``.
|
||||||
|
|
||||||
|
To use the exported model with ``./zipformer_mmi/pretrained.py``, you
|
||||||
|
can run:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
./zipformer_mmi/pretrained.py \
|
||||||
|
--checkpoint ./zipformer_mmi/exp/pretrained.pt \
|
||||||
|
--bpe-model ./data/lang_bpe_500/bpe.model \
|
||||||
|
--method 1best \
|
||||||
|
/path/to/foo.wav \
|
||||||
|
/path/to/bar.wav
|
||||||
|
|
||||||
|
Export model using ``torch.jit.script()``
|
||||||
|
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
./zipformer_mmi/export.py \
|
||||||
|
--exp-dir ./zipformer_mmi/exp \
|
||||||
|
--bpe-model data/lang_bpe_500/bpe.model \
|
||||||
|
--epoch 30 \
|
||||||
|
--avg 9 \
|
||||||
|
--jit 1
|
||||||
|
|
||||||
|
It will generate a file ``cpu_jit.pt`` in the given ``exp_dir``. You can later
|
||||||
|
load it by ``torch.jit.load("cpu_jit.pt")``.
|
||||||
|
|
||||||
|
Note ``cpu`` in the name ``cpu_jit.pt`` means the parameters when loaded into Python
|
||||||
|
are on CPU. You can use ``to("cuda")`` to move them to a CUDA device.
|
||||||
|
|
||||||
|
To use the generated files with ``./zipformer_mmi/jit_pretrained.py``:
|
||||||
|
|
||||||
|
.. code-block:: bash
|
||||||
|
|
||||||
|
./zipformer_mmi/jit_pretrained.py \
|
||||||
|
--nn-model-filename ./zipformer_mmi/exp/cpu_jit.pt \
|
||||||
|
--bpe-model ./data/lang_bpe_500/bpe.model \
|
||||||
|
--method 1best \
|
||||||
|
/path/to/foo.wav \
|
||||||
|
/path/to/bar.wav
|
||||||
|
|
||||||
|
Download pretrained models
|
||||||
|
--------------------------
|
||||||
|
|
||||||
|
If you don't want to train from scratch, you can download the pretrained models
|
||||||
|
by visiting the following links:
|
||||||
|
|
||||||
|
- `<https://huggingface.co/Zengwei/icefall-asr-librispeech-zipformer-mmi-2022-12-08>`_
|
||||||
|
|
||||||
|
See `<https://github.com/k2-fsa/icefall/blob/master/egs/librispeech/ASR/RESULTS.md>`_
|
||||||
|
for the details of the above pretrained models
|
@ -6,4 +6,3 @@ TIMIT
|
|||||||
|
|
||||||
tdnn_ligru_ctc
|
tdnn_ligru_ctc
|
||||||
tdnn_lstm_ctc
|
tdnn_lstm_ctc
|
||||||
|
|
Before Width: | Height: | Size: 121 KiB After Width: | Height: | Size: 121 KiB |
12
docs/source/recipes/Streaming-ASR/index.rst
Normal file
@ -0,0 +1,12 @@
|
|||||||
|
Streaming ASR
|
||||||
|
=============
|
||||||
|
|
||||||
|
.. toctree::
|
||||||
|
:maxdepth: 1
|
||||||
|
|
||||||
|
introduction
|
||||||
|
|
||||||
|
.. toctree::
|
||||||
|
:maxdepth: 2
|
||||||
|
|
||||||
|
librispeech/index
|
52
docs/source/recipes/Streaming-ASR/introduction.rst
Normal file
@ -0,0 +1,52 @@
|
|||||||
|
Introduction
|
||||||
|
============
|
||||||
|
|
||||||
|
This page shows you how we implement streaming **X-former transducer** models for ASR.
|
||||||
|
|
||||||
|
.. HINT::
|
||||||
|
X-former transducer here means the encoder of the transducer model uses Multi-Head Attention,
|
||||||
|
like `Conformer <https://arxiv.org/pdf/2005.08100.pdf>`_, `EmFormer <https://arxiv.org/pdf/2010.10759.pdf>`_ etc.
|
||||||
|
|
||||||
|
Currently we have implemented two types of streaming models, one uses Conformer as encoder, the other uses Emformer as encoder.
|
||||||
|
|
||||||
|
Streaming Conformer
|
||||||
|
-------------------
|
||||||
|
|
||||||
|
The main idea of training a streaming model is to make the model see limited contexts
|
||||||
|
in training time, we can achieve this by applying a mask to the output of self-attention.
|
||||||
|
In icefall, we implement the streaming conformer the way just like what `WeNet <https://arxiv.org/pdf/2012.05481.pdf>`_ did.
|
||||||
|
|
||||||
|
.. NOTE::
|
||||||
|
The conformer-transducer recipes in LibriSpeech datasets, like, `pruned_transducer_stateless <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless>`_,
|
||||||
|
`pruned_transducer_stateless2 <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless2>`_,
|
||||||
|
`pruned_transducer_stateless3 <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless3>`_,
|
||||||
|
`pruned_transducer_stateless4 <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless4>`_,
|
||||||
|
`pruned_transducer_stateless5 <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_transducer_stateless5>`_
|
||||||
|
all support streaming.
|
||||||
|
|
||||||
|
.. NOTE::
|
||||||
|
Training a streaming conformer model in ``icefall`` is almost the same as training a
|
||||||
|
non-streaming model, all you need to do is passing several extra arguments.
|
||||||
|
See :doc:`Pruned transducer statelessX <librispeech/pruned_transducer_stateless>` for more details.
|
||||||
|
|
||||||
|
.. HINT::
|
||||||
|
If you want to adapt a non-streaming conformer model to be streaming, please refer
|
||||||
|
to `this pull request <https://github.com/k2-fsa/icefall/pull/454>`_.
|
||||||
|
|
||||||
|
|
||||||
|
Streaming Emformer
|
||||||
|
------------------
|
||||||
|
|
||||||
|
The Emformer model proposed `here <https://arxiv.org/pdf/2010.10759.pdf>`_ uses more
|
||||||
|
complicated techniques. It has a memory bank component to memorize history information,
|
||||||
|
what' more, it also introduces right context in training time by hard-copying part of
|
||||||
|
the input features.
|
||||||
|
|
||||||
|
We have three variants of Emformer models in ``icefall``.
|
||||||
|
|
||||||
|
- ``pruned_stateless_emformer_rnnt2`` using Emformer from torchaudio, see `LibriSpeech recipe <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/pruned_stateless_emformer_rnnt2>`_.
|
||||||
|
- ``conv_emformer_transducer_stateless`` using ConvEmformer implemented by ourself. Different from the Emformer in torchaudio,
|
||||||
|
ConvEmformer has a convolution in each layer and uses the mechanisms in our reworked conformer model.
|
||||||
|
See `LibriSpeech recipe <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/conv_emformer_transducer_stateless>`_.
|
||||||
|
- ``conv_emformer_transducer_stateless2`` using ConvEmformer implemented by ourself. The only difference from the above one is that
|
||||||
|
it uses a simplified memory bank. See `LibriSpeech recipe <https://github.com/k2-fsa/icefall/tree/master/egs/librispeech/ASR/conv_emformer_transducer_stateless2>`_.
|
Before Width: | Height: | Size: 413 KiB After Width: | Height: | Size: 413 KiB |