Skip to content

GitLab

Commit 70a8a9e0 authored by wangwei990215's avatar wangwei990215
Browse files

initial commit

parents
Pipeline #1738 failed with stages
in 0 seconds
Hide whitespace changes
Inline Side-by-side
Showing with 1375 additions and 0 deletions
FunASR/examples/aishell/branchformer/run.sh 0 → 100755
View file @ 70a8a9e0
#!/usr/bin/env bash
CUDA_VISIBLE_DEVICES="0,1,2,3"
# general configuration
feats_dir="../DATA" #feature output dictionary
exp_dir=`pwd`
lang=zh
token_type=char
stage=0
stop_stage=5
# feature configuration
nj=32
inference_device="cuda" #"cpu"
inference_checkpoint="model.pt.avg10"
inference_scp="wav.scp"
inference_batch_size=1
# data
raw_data=../raw_data
data_url=www.openslr.org/resources/33
# exp tag
tag="exp1"
workspace=`pwd`
master_port=12345
. utils/parse_options.sh || exit 1;
# Set bash to 'debug' mode, it will exit on :
# -e 'error', -u 'undefined variable', -o ... 'error in pipeline', -x 'print commands',
set -e
set -u
set -o pipefail
train_set=train
valid_set=dev
test_sets="dev test"
config=branchformer_12e_6d_2048_256.yaml
model_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"
if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
echo "stage -1: Data Download"
mkdir -p ${raw_data}
local/download_and_untar.sh ${raw_data} ${data_url} data_aishell
local/download_and_untar.sh ${raw_data} ${data_url} resource_aishell
fi
if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
echo "stage 0: Data preparation"
# Data preparation
local/aishell_data_prep.sh ${raw_data}/data_aishell/wav ${raw_data}/data_aishell/transcript ${feats_dir}
for x in train dev test; do
cp ${feats_dir}/data/${x}/text ${feats_dir}/data/${x}/text.org
paste -d " " <(cut -f 1 -d" " ${feats_dir}/data/${x}/text.org) <(cut -f 2- -d" " ${feats_dir}/data/${x}/text.org | tr -d " ") \
> ${feats_dir}/data/${x}/text
utils/text2token.py -n 1 -s 1 ${feats_dir}/data/${x}/text > ${feats_dir}/data/${x}/text.org
mv ${feats_dir}/data/${x}/text.org ${feats_dir}/data/${x}/text
# convert wav.scp text to jsonl
scp_file_list_arg="++scp_file_list='[\"${feats_dir}/data/${x}/wav.scp\",\"${feats_dir}/data/${x}/text\"]'"
python ../../../funasr/datasets/audio_datasets/scp2jsonl.py \
++data_type_list='["source", "target"]' \
++jsonl_file_out=${feats_dir}/data/${x}/audio_datasets.jsonl \
${scp_file_list_arg}
done
fi
if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
echo "stage 1: Feature and CMVN Generation"
python ../../../funasr/bin/compute_audio_cmvn.py \
--config-path "${workspace}/conf" \
--config-name "${config}" \
++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
++cmvn_file="${feats_dir}/data/${train_set}/cmvn.json" \
fi
token_list=${feats_dir}/data/${lang}_token_list/$token_type/tokens.txt
echo "dictionary: ${token_list}"
if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
echo "stage 2: Dictionary Preparation"
mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/
echo "make a dictionary"
echo "<blank>" > ${token_list}
echo "<s>" >> ${token_list}
echo "</s>" >> ${token_list}
utils/text2token.py -s 1 -n 1 --space "" ${feats_dir}/data/$train_set/text | cut -f 2- -d" " | tr " " "\n" \
| sort | uniq | grep -a -v -e '^\s*$' | awk '{print $0}' >> ${token_list}
echo "<unk>" >> ${token_list}
fi
# LM Training Stage
if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
echo "stage 3: LM Training"
fi
# ASR Training Stage
if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
echo "stage 4: ASR Training"
mkdir -p ${exp_dir}/exp/${model_dir}
current_time=$(date "+%Y-%m-%d_%H-%M")
log_file="${exp_dir}/exp/${model_dir}/train.log.txt.${current_time}"
echo "log_file: ${log_file}"
export CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES
gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
torchrun \
--nnodes 1 \
--nproc_per_node ${gpu_num} \
--master_port ${master_port} \
../../../funasr/bin/train.py \
--config-path "${workspace}/conf" \
--config-name "${config}" \
++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
++valid_data_set_list="${feats_dir}/data/${valid_set}/audio_datasets.jsonl" \
++tokenizer_conf.token_list="${token_list}" \
++frontend_conf.cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
++output_dir="${exp_dir}/exp/${model_dir}" &> ${log_file}
fi
# Testing Stage
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
echo "stage 5: Inference"
if [ ${inference_device} == "cuda" ]; then
nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
else
inference_batch_size=1
CUDA_VISIBLE_DEVICES=""
for JOB in $(seq ${nj}); do
CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"-1,"
done
fi
for dset in ${test_sets}; do
inference_dir="${exp_dir}/exp/${model_dir}/inference-${inference_checkpoint}/${dset}"
_logdir="${inference_dir}/logdir"
echo "inference_dir: ${inference_dir}"
mkdir -p "${_logdir}"
data_dir="${feats_dir}/data/${dset}"
key_file=${data_dir}/${inference_scp}
split_scps=
for JOB in $(seq "${nj}"); do
split_scps+=" ${_logdir}/keys.${JOB}.scp"
done
utils/split_scp.pl "${key_file}" ${split_scps}
gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })
for JOB in $(seq ${nj}); do
{
id=$((JOB-1))
gpuid=${gpuid_list_array[$id]}
export CUDA_VISIBLE_DEVICES=${gpuid}
python ../../../funasr/bin/inference.py \
--config-path="${exp_dir}/exp/${model_dir}" \
--config-name="config.yaml" \
++init_param="${exp_dir}/exp/${model_dir}/${inference_checkpoint}" \
++tokenizer_conf.token_list="${token_list}" \
++frontend_conf.cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
++input="${_logdir}/keys.${JOB}.scp" \
++output_dir="${inference_dir}/${JOB}" \
++device="${inference_device}" \
++ncpu=1 \
++disable_log=true \
++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
}&
done
wait
mkdir -p ${inference_dir}/1best_recog
for f in token score text; do
if [ -f "${inference_dir}/${JOB}/1best_recog/${f}" ]; then
for JOB in $(seq "${nj}"); do
cat "${inference_dir}/${JOB}/1best_recog/${f}"
done | sort -k1 >"${inference_dir}/1best_recog/${f}"
fi
done
echo "Computing WER ..."
python utils/postprocess_text_zh.py ${inference_dir}/1best_recog/text ${inference_dir}/1best_recog/text.proc
python utils/postprocess_text_zh.py ${data_dir}/text ${inference_dir}/1best_recog/text.ref
python utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.cer
tail -n 3 ${inference_dir}/1best_recog/text.cer
done
fi
\ No newline at end of file
FunASR/examples/aishell/branchformer/utils 0 → 120000
View file @ 70a8a9e0
../paraformer/utils
\ No newline at end of file
FunASR/examples/aishell/conformer/README.md 0 → 100644
View file @ 70a8a9e0
# Conformer Result
## Training Config
- Feature info: using 80 dims fbank, global cmvn, speed perturb(0.9, 1.0, 1.1), specaugment
- Train info: lr 5e-4, batch_size 25000, 2 gpu(Tesla V100), acc_grad 1, 50 epochs
- Train config: conf/train_asr_transformer.yaml
- LM config: LM was not used
- Model size: 46M
## Results (CER)
| testset | CER(%) |
|:-----------:|:-------:|
| dev | 4.42 |
| test | 4.87 |
\ No newline at end of file
FunASR/examples/aishell/conformer/conf/conformer_12e_6d_2048_256.yaml 0 → 100644
View file @ 70a8a9e0
# This is an example that demonstrates how to configure a model file.
# You can modify the configuration according to your own requirements.
# to print the register_table:
# from funasr.register import tables
# tables.print()
# network architecture
model: Conformer
model_conf:
ctc_weight: 0.3
lsm_weight: 0.1 # label smoothing option
length_normalized_loss: false
# encoder
encoder: ConformerEncoder
encoder_conf:
output_size: 256 # dimension of attention
attention_heads: 4
linear_units: 2048 # the number of units of position-wise feed forward
num_blocks: 12 # the number of encoder blocks
dropout_rate: 0.1
positional_dropout_rate: 0.1
attention_dropout_rate: 0.0
input_layer: conv2d # encoder architecture type
normalize_before: true
pos_enc_layer_type: rel_pos
selfattention_layer_type: rel_selfattn
activation_type: swish
macaron_style: true
use_cnn_module: true
cnn_module_kernel: 15
# decoder
decoder: TransformerDecoder
decoder_conf:
attention_heads: 4
linear_units: 2048
num_blocks: 6
dropout_rate: 0.1
positional_dropout_rate: 0.1
self_attention_dropout_rate: 0.0
src_attention_dropout_rate: 0.0
# frontend related
frontend: WavFrontend
frontend_conf:
fs: 16000
window: hamming
n_mels: 80
frame_length: 25
frame_shift: 10
lfr_m: 1
lfr_n: 1
specaug: SpecAug
specaug_conf:
apply_time_warp: true
time_warp_window: 5
time_warp_mode: bicubic
apply_freq_mask: true
freq_mask_width_range:
- 0
- 30
num_freq_mask: 2
apply_time_mask: true
time_mask_width_range:
- 0
- 40
num_time_mask: 2
train_conf:
accum_grad: 1
grad_clip: 5
max_epoch: 150
keep_nbest_models: 10
log_interval: 50
optim: adam
optim_conf:
lr: 0.0005
scheduler: warmuplr
scheduler_conf:
warmup_steps: 30000
dataset: AudioDataset
dataset_conf:
index_ds: IndexDSJsonl
batch_sampler: EspnetStyleBatchSampler
batch_type: length # example or length
batch_size: 25000 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
buffer_size: 1024
shuffle: True
num_workers: 4
preprocessor_speech: SpeechPreprocessSpeedPerturb
preprocessor_speech_conf:
speed_perturb: [0.9, 1.0, 1.1]
tokenizer: CharTokenizer
tokenizer_conf:
unk_symbol: <unk>
ctc_conf:
dropout_rate: 0.0
ctc_type: builtin
reduce: true
ignore_nan_grad: true
normalize: null
FunASR/examples/aishell/conformer/conf/conformer_rwkv.yaml 0 → 100644
View file @ 70a8a9e0
# This is an example that demonstrates how to configure a model file.
# You can modify the configuration according to your own requirements.
# to print the register_table:
# from funasr.register import tables
# tables.print()
# network architecture
model: Conformer
model_conf:
ctc_weight: 0.3
lsm_weight: 0.1 # label smoothing option
length_normalized_loss: false
# encoder
encoder: ConformerEncoder
encoder_conf:
output_size: 256 # dimension of attention
attention_heads: 4
linear_units: 2048 # the number of units of position-wise feed forward
num_blocks: 12 # the number of encoder blocks
dropout_rate: 0.1
positional_dropout_rate: 0.1
attention_dropout_rate: 0.0
input_layer: conv2d # encoder architecture type
normalize_before: true
pos_enc_layer_type: rel_pos
selfattention_layer_type: rel_selfattn
activation_type: swish
macaron_style: true
use_cnn_module: true
cnn_module_kernel: 15
# decoder
decoder: TransformerRWKVDecoder
decoder_conf:
attention_heads: 4
linear_units: 2048
num_blocks: 6
dropout_rate: 0.1
positional_dropout_rate: 0.1
self_attention_dropout_rate: 0.0
src_attention_dropout_rate: 0.0
input_layer: embed
rwkv_cfg:
n_embd: 256
dropout: 0
head_size_a: 64
ctx_len: 512
dim_att: 256 #${model_conf.rwkv_cfg.n_embd}
dim_ffn: null
head_size_divisor: 4
n_layer: 6
pre_ffn: 0
ln0: false
ln1: false
init_rwkv: true
# frontend related
frontend: WavFrontend
frontend_conf:
fs: 16000
window: hamming
n_mels: 80
frame_length: 25
frame_shift: 10
lfr_m: 1
lfr_n: 1
specaug: SpecAug
specaug_conf:
apply_time_warp: true
time_warp_window: 5
time_warp_mode: bicubic
apply_freq_mask: true
freq_mask_width_range:
- 0
- 30
num_freq_mask: 2
apply_time_mask: true
time_mask_width_range:
- 0
- 40
num_time_mask: 2
train_conf:
accum_grad: 1
grad_clip: 5
max_epoch: 150
keep_nbest_models: 10
log_interval: 50
optim: adam
optim_conf:
lr: 0.0005
scheduler: warmuplr
scheduler_conf:
warmup_steps: 30000
dataset: AudioDataset
dataset_conf:
index_ds: IndexDSJsonl
batch_sampler: EspnetStyleBatchSampler
batch_type: length # example or length
batch_size: 25000 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
buffer_size: 1024
shuffle: True
num_workers: 4
preprocessor_speech: SpeechPreprocessSpeedPerturb
preprocessor_speech_conf:
speed_perturb: [0.9, 1.0, 1.1]
tokenizer: CharTokenizer
tokenizer_conf:
unk_symbol: <unk>
ctc_conf:
dropout_rate: 0.0
ctc_type: builtin
reduce: true
ignore_nan_grad: true
normalize: null
FunASR/examples/aishell/conformer/demo_infer.sh 0 → 120000
View file @ 70a8a9e0
../paraformer/demo_infer.sh
\ No newline at end of file
FunASR/examples/aishell/conformer/demo_train_or_finetune.sh 0 → 120000
View file @ 70a8a9e0
../paraformer/demo_train_or_finetune.sh
\ No newline at end of file
FunASR/examples/aishell/conformer/local/aishell_data_prep.sh 0 → 100755
View file @ 70a8a9e0
#!/bin/bash
# Copyright 2017 Xingyu Na
# Apache 2.0
#. ./path.sh || exit 1;
if [ $# != 3 ]; then
echo "Usage: $0 <audio-path> <text-path> <output-path>"
echo " $0 /export/a05/xna/data/data_aishell/wav /export/a05/xna/data/data_aishell/transcript data"
exit 1;
fi
aishell_audio_dir=$1
aishell_text=$2/aishell_transcript_v0.8.txt
output_dir=$3
train_dir=$output_dir/data/local/train
dev_dir=$output_dir/data/local/dev
test_dir=$output_dir/data/local/test
tmp_dir=$output_dir/data/local/tmp
mkdir -p $train_dir
mkdir -p $dev_dir
mkdir -p $test_dir
mkdir -p $tmp_dir
# data directory check
if [ ! -d $aishell_audio_dir ] || [ ! -f $aishell_text ]; then
echo "Error: $0 requires two directory arguments"
exit 1;
fi
# find wav audio file for train, dev and test resp.
find $aishell_audio_dir -iname "*.wav" > $tmp_dir/wav.flist
n=`cat $tmp_dir/wav.flist | wc -l`
[ $n -ne 141925 ] && \
echo Warning: expected 141925 data data files, found $n
grep -i "wav/train" $tmp_dir/wav.flist > $train_dir/wav.flist || exit 1;
grep -i "wav/dev" $tmp_dir/wav.flist > $dev_dir/wav.flist || exit 1;
grep -i "wav/test" $tmp_dir/wav.flist > $test_dir/wav.flist || exit 1;
rm -r $tmp_dir
# Transcriptions preparation
for dir in $train_dir $dev_dir $test_dir; do
echo Preparing $dir transcriptions
sed -e 's/\.wav//' $dir/wav.flist | awk -F '/' '{print $NF}' > $dir/utt.list
paste -d' ' $dir/utt.list $dir/wav.flist > $dir/wav.scp_all
utils/filter_scp.pl -f 1 $dir/utt.list $aishell_text > $dir/transcripts.txt
awk '{print $1}' $dir/transcripts.txt > $dir/utt.list
utils/filter_scp.pl -f 1 $dir/utt.list $dir/wav.scp_all | sort -u > $dir/wav.scp
sort -u $dir/transcripts.txt > $dir/text
done
mkdir -p $output_dir/data/train $output_dir/data/dev $output_dir/data/test
for f in wav.scp text; do
cp $train_dir/$f $output_dir/data/train/$f || exit 1;
cp $dev_dir/$f $output_dir/data/dev/$f || exit 1;
cp $test_dir/$f $output_dir/data/test/$f || exit 1;
done
echo "$0: AISHELL data preparation succeeded"
exit 0;
FunASR/examples/aishell/conformer/local/download_and_untar.sh 0 → 100755
View file @ 70a8a9e0
#!/usr/bin/env bash
# Copyright 2014 Johns Hopkins University (author: Daniel Povey)
# 2017 Xingyu Na
# Apache 2.0
remove_archive=false
if [ "$1" == --remove-archive ]; then
remove_archive=true
shift
fi
if [ $# -ne 3 ]; then
echo "Usage: $0 [--remove-archive] <data-base> <url-base> <corpus-part>"
echo "e.g.: $0 /export/a05/xna/data www.openslr.org/resources/33 data_aishell"
echo "With --remove-archive it will remove the archive after successfully un-tarring it."
echo "<corpus-part> can be one of: data_aishell, resource_aishell."
fi
data=$1
url=$2
part=$3
if [ ! -d "$data" ]; then
echo "$0: no such directory $data"
exit 1;
fi
part_ok=false
list="data_aishell resource_aishell"
for x in $list; do
if [ "$part" == $x ]; then part_ok=true; fi
done
if ! $part_ok; then
echo "$0: expected <corpus-part> to be one of $list, but got '$part'"
exit 1;
fi
if [ -z "$url" ]; then
echo "$0: empty URL base."
exit 1;
fi
if [ -f $data/$part/.complete ]; then
echo "$0: data part $part was already successfully extracted, nothing to do."
exit 0;
fi
# sizes of the archive files in bytes.
sizes="15582913665 1246920"
if [ -f $data/$part.tgz ]; then
size=$(/bin/ls -l $data/$part.tgz | awk '{print $5}')
size_ok=false
for s in $sizes; do if [ $s == $size ]; then size_ok=true; fi; done
if ! $size_ok; then
echo "$0: removing existing file $data/$part.tgz because its size in bytes $size"
echo "does not equal the size of one of the archives."
rm $data/$part.tgz
else
echo "$data/$part.tgz exists and appears to be complete."
fi
fi
if [ ! -f $data/$part.tgz ]; then
if ! command -v wget >/dev/null; then
echo "$0: wget is not installed."
exit 1;
fi
full_url=$url/$part.tgz
echo "$0: downloading data from $full_url. This may take some time, please be patient."
cd $data || exit 1
if ! wget --no-check-certificate $full_url; then
echo "$0: error executing wget $full_url"
exit 1;
fi
fi
cd $data || exit 1
if ! tar -xvzf $part.tgz; then
echo "$0: error un-tarring archive $data/$part.tgz"
exit 1;
fi
touch $data/$part/.complete
if [ $part == "data_aishell" ]; then
cd $data/$part/wav || exit 1
for wav in ./*.tar.gz; do
echo "Extracting wav from $wav"
tar -zxf $wav && rm $wav
done
fi
echo "$0: Successfully downloaded and un-tarred $data/$part.tgz"
if $remove_archive; then
echo "$0: removing $data/$part.tgz file since --remove-archive option was supplied."
rm $data/$part.tgz
fi
exit 0;
FunASR/examples/aishell/conformer/run.sh 0 → 100755
View file @ 70a8a9e0
#!/usr/bin/env bash
CUDA_VISIBLE_DEVICES="0,1"
# general configuration
feats_dir="../DATA" #feature output dictionary
exp_dir=`pwd`
lang=zh
token_type=char
stage=0
stop_stage=5
# feature configuration
nj=32
inference_device="cuda" #"cpu", "cuda:0", "cuda:1"
inference_checkpoint="model.pt.avg10"
inference_scp="wav.scp"
inference_batch_size=1
# data
raw_data=../raw_data
data_url=www.openslr.org/resources/33
# exp tag
tag="exp1"
workspace=`pwd`
master_port=12345
. utils/parse_options.sh || exit 1;
# Set bash to 'debug' mode, it will exit on :
# -e 'error', -u 'undefined variable', -o ... 'error in pipeline', -x 'print commands',
set -e
set -u
set -o pipefail
train_set=train
valid_set=dev
test_sets="dev test"
config=conformer_12e_6d_2048_256.yaml
model_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"
if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
echo "stage -1: Data Download"
mkdir -p ${raw_data}
local/download_and_untar.sh ${raw_data} ${data_url} data_aishell
local/download_and_untar.sh ${raw_data} ${data_url} resource_aishell
fi
if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
echo "stage 0: Data preparation"
# Data preparation
local/aishell_data_prep.sh ${raw_data}/data_aishell/wav ${raw_data}/data_aishell/transcript ${feats_dir}
for x in train dev test; do
cp ${feats_dir}/data/${x}/text ${feats_dir}/data/${x}/text.org
paste -d " " <(cut -f 1 -d" " ${feats_dir}/data/${x}/text.org) <(cut -f 2- -d" " ${feats_dir}/data/${x}/text.org | tr -d " ") \
> ${feats_dir}/data/${x}/text
utils/text2token.py -n 1 -s 1 ${feats_dir}/data/${x}/text > ${feats_dir}/data/${x}/text.org
mv ${feats_dir}/data/${x}/text.org ${feats_dir}/data/${x}/text
# convert wav.scp text to jsonl
scp_file_list_arg="++scp_file_list='[\"${feats_dir}/data/${x}/wav.scp\",\"${feats_dir}/data/${x}/text\"]'"
python ../../../funasr/datasets/audio_datasets/scp2jsonl.py \
++data_type_list='["source", "target"]' \
++jsonl_file_out=${feats_dir}/data/${x}/audio_datasets.jsonl \
${scp_file_list_arg}
done
fi
if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
echo "stage 1: Feature and CMVN Generation"
python ../../../funasr/bin/compute_audio_cmvn.py \
--config-path "${workspace}/conf" \
--config-name "${config}" \
++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
++cmvn_file="${feats_dir}/data/${train_set}/cmvn.json" \
fi
token_list=${feats_dir}/data/${lang}_token_list/$token_type/tokens.txt
echo "dictionary: ${token_list}"
if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
echo "stage 2: Dictionary Preparation"
mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/
echo "make a dictionary"
echo "<blank>" > ${token_list}
echo "<s>" >> ${token_list}
echo "</s>" >> ${token_list}
utils/text2token.py -s 1 -n 1 --space "" ${feats_dir}/data/$train_set/text | cut -f 2- -d" " | tr " " "\n" \
| sort | uniq | grep -a -v -e '^\s*$' | awk '{print $0}' >> ${token_list}
echo "<unk>" >> ${token_list}
fi
# LM Training Stage
if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
echo "stage 3: LM Training"
fi
# ASR Training Stage
if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
echo "stage 4: ASR Training"
mkdir -p ${exp_dir}/exp/${model_dir}
current_time=$(date "+%Y-%m-%d_%H-%M")
log_file="${exp_dir}/exp/${model_dir}/train.log.txt.${current_time}"
echo "log_file: ${log_file}"
export CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES
gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
torchrun \
--nnodes 1 \
--nproc_per_node ${gpu_num} \
--master_port ${master_port} \
../../../funasr/bin/train.py \
--config-path "${workspace}/conf" \
--config-name "${config}" \
++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
++valid_data_set_list="${feats_dir}/data/${valid_set}/audio_datasets.jsonl" \
++tokenizer_conf.token_list="${token_list}" \
++frontend_conf.cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
++output_dir="${exp_dir}/exp/${model_dir}" &> ${log_file}
fi
# Testing Stage
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
echo "stage 5: Inference"
if [ ${inference_device} == "cuda" ]; then
nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
else
inference_batch_size=1
CUDA_VISIBLE_DEVICES=""
for JOB in $(seq ${nj}); do
CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"-1,"
done
fi
for dset in ${test_sets}; do
inference_dir="${exp_dir}/exp/${model_dir}/inference-${inference_checkpoint}/${dset}"
_logdir="${inference_dir}/logdir"
echo "inference_dir: ${inference_dir}"
mkdir -p "${_logdir}"
data_dir="${feats_dir}/data/${dset}"
key_file=${data_dir}/${inference_scp}
split_scps=
for JOB in $(seq "${nj}"); do
split_scps+=" ${_logdir}/keys.${JOB}.scp"
done
utils/split_scp.pl "${key_file}" ${split_scps}
gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })
for JOB in $(seq ${nj}); do
{
id=$((JOB-1))
gpuid=${gpuid_list_array[$id]}
export CUDA_VISIBLE_DEVICES=${gpuid}
python ../../../funasr/bin/inference.py \
--config-path="${exp_dir}/exp/${model_dir}" \
--config-name="config.yaml" \
++init_param="${exp_dir}/exp/${model_dir}/${inference_checkpoint}" \
++tokenizer_conf.token_list="${token_list}" \
++frontend_conf.cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
++input="${_logdir}/keys.${JOB}.scp" \
++output_dir="${inference_dir}/${JOB}" \
++device="${inference_device}" \
++ncpu=1 \
++disable_log=true \
++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
}&
done
wait
mkdir -p ${inference_dir}/1best_recog
for f in token score text; do
if [ -f "${inference_dir}/${JOB}/1best_recog/${f}" ]; then
for JOB in $(seq "${nj}"); do
cat "${inference_dir}/${JOB}/1best_recog/${f}"
done | sort -k1 >"${inference_dir}/1best_recog/${f}"
fi
done
echo "Computing WER ..."
python utils/postprocess_text_zh.py ${inference_dir}/1best_recog/text ${inference_dir}/1best_recog/text.proc
python utils/postprocess_text_zh.py ${data_dir}/text ${inference_dir}/1best_recog/text.ref
python utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.cer
tail -n 3 ${inference_dir}/1best_recog/text.cer
done
fi
FunASR/examples/aishell/conformer/utils 0 → 120000
View file @ 70a8a9e0
../paraformer/utils
\ No newline at end of file
FunASR/examples/aishell/e_branchformer/README.md 0 → 100644
View file @ 70a8a9e0
# E-Branchformer Result
## Training Config
- Feature info: using raw speech, extracting 80 dims fbank online, global cmvn, speed perturb(0.9, 1.0, 1.1), specaugment
- Train info: lr 0.001, batch_size 10000, 4 gpu(Tesla V100), acc_grad 1, 180 epochs
- Train config: conf/train_asr_e_branchformer.yaml
- LM config: LM was not used
## Results (CER)
| testset | CER(%) |
|:-----------:|:-------:|
| dev | 4.10 |
| test | 4.52 |
\ No newline at end of file
FunASR/examples/aishell/e_branchformer/conf/e_branchformer_12e_6d_2048_256.yaml 0 → 100644
View file @ 70a8a9e0
# This is an example that demonstrates how to configure a model file.
# You can modify the configuration according to your own requirements.
# to print the register_table:
# from funasr.register import tables
# tables.print()
# network architecture
model: Branchformer
model_conf:
ctc_weight: 0.3
lsm_weight: 0.1 # label smoothing option
length_normalized_loss: false
# encoder
encoder: EBranchformerEncoder
encoder_conf:
output_size: 256
attention_heads: 4
attention_layer_type: rel_selfattn
pos_enc_layer_type: rel_pos
rel_pos_type: latest
cgmlp_linear_units: 1024
cgmlp_conv_kernel: 31
use_linear_after_conv: false
gate_activation: identity
num_blocks: 12
dropout_rate: 0.1
positional_dropout_rate: 0.1
attention_dropout_rate: 0.1
input_layer: conv2d
layer_drop_rate: 0.0
linear_units: 1024
positionwise_layer_type: linear
use_ffn: true
macaron_ffn: true
merge_conv_kernel: 31
# decoder
decoder: TransformerDecoder
decoder_conf:
attention_heads: 4
linear_units: 2048
num_blocks: 6
dropout_rate: 0.1
positional_dropout_rate: 0.1
self_attention_dropout_rate: 0.
src_attention_dropout_rate: 0.
# frontend related
frontend: WavFrontend
frontend_conf:
fs: 16000
window: hamming
n_mels: 80
frame_length: 25
frame_shift: 10
dither: 0.0
lfr_m: 1
lfr_n: 1
specaug: SpecAug
specaug_conf:
apply_time_warp: true
time_warp_window: 5
time_warp_mode: bicubic
apply_freq_mask: true
freq_mask_width_range:
- 0
- 30
num_freq_mask: 2
apply_time_mask: true
time_mask_width_range:
- 0
- 40
num_time_mask: 2
train_conf:
accum_grad: 1
grad_clip: 5
max_epoch: 180
keep_nbest_models: 10
log_interval: 50
optim: adam
optim_conf:
lr: 0.001
weight_decay: 0.000001
scheduler: warmuplr
scheduler_conf:
warmup_steps: 35000
dataset: AudioDataset
dataset_conf:
index_ds: IndexDSJsonl
batch_sampler: EspnetStyleBatchSampler
batch_type: length # example or length
batch_size: 10000 # if batch_type is example, batch_size is the numbers of samples; if length, batch_size is source_token_len+target_token_len;
max_token_length: 2048 # filter samples if source_token_len+target_token_len > max_token_length,
buffer_size: 1024
shuffle: True
num_workers: 4
preprocessor_speech: SpeechPreprocessSpeedPerturb
preprocessor_speech_conf:
speed_perturb: [0.9, 1.0, 1.1]
tokenizer: CharTokenizer
tokenizer_conf:
unk_symbol: <unk>
ctc_conf:
dropout_rate: 0.0
ctc_type: builtin
reduce: true
ignore_nan_grad: true
normalize: null
FunASR/examples/aishell/e_branchformer/demo_infer.sh 0 → 120000
View file @ 70a8a9e0
../paraformer/demo_infer.sh
\ No newline at end of file
FunASR/examples/aishell/e_branchformer/demo_train_or_finetune.sh 0 → 120000
View file @ 70a8a9e0
../paraformer/demo_train_or_finetune.sh
\ No newline at end of file
FunASR/examples/aishell/e_branchformer/local/aishell_data_prep.sh 0 → 100755
View file @ 70a8a9e0
#!/bin/bash
# Copyright 2017 Xingyu Na
# Apache 2.0
#. ./path.sh || exit 1;
if [ $# != 3 ]; then
echo "Usage: $0 <audio-path> <text-path> <output-path>"
echo " $0 /export/a05/xna/data/data_aishell/wav /export/a05/xna/data/data_aishell/transcript data"
exit 1;
fi
aishell_audio_dir=$1
aishell_text=$2/aishell_transcript_v0.8.txt
output_dir=$3
train_dir=$output_dir/data/local/train
dev_dir=$output_dir/data/local/dev
test_dir=$output_dir/data/local/test
tmp_dir=$output_dir/data/local/tmp
mkdir -p $train_dir
mkdir -p $dev_dir
mkdir -p $test_dir
mkdir -p $tmp_dir
# data directory check
if [ ! -d $aishell_audio_dir ] || [ ! -f $aishell_text ]; then
echo "Error: $0 requires two directory arguments"
exit 1;
fi
# find wav audio file for train, dev and test resp.
find $aishell_audio_dir -iname "*.wav" > $tmp_dir/wav.flist
n=`cat $tmp_dir/wav.flist | wc -l`
[ $n -ne 141925 ] && \
echo Warning: expected 141925 data data files, found $n
grep -i "wav/train" $tmp_dir/wav.flist > $train_dir/wav.flist || exit 1;
grep -i "wav/dev" $tmp_dir/wav.flist > $dev_dir/wav.flist || exit 1;
grep -i "wav/test" $tmp_dir/wav.flist > $test_dir/wav.flist || exit 1;
rm -r $tmp_dir
# Transcriptions preparation
for dir in $train_dir $dev_dir $test_dir; do
echo Preparing $dir transcriptions
sed -e 's/\.wav//' $dir/wav.flist | awk -F '/' '{print $NF}' > $dir/utt.list
paste -d' ' $dir/utt.list $dir/wav.flist > $dir/wav.scp_all
utils/filter_scp.pl -f 1 $dir/utt.list $aishell_text > $dir/transcripts.txt
awk '{print $1}' $dir/transcripts.txt > $dir/utt.list
utils/filter_scp.pl -f 1 $dir/utt.list $dir/wav.scp_all | sort -u > $dir/wav.scp
sort -u $dir/transcripts.txt > $dir/text
done
mkdir -p $output_dir/data/train $output_dir/data/dev $output_dir/data/test
for f in wav.scp text; do
cp $train_dir/$f $output_dir/data/train/$f || exit 1;
cp $dev_dir/$f $output_dir/data/dev/$f || exit 1;
cp $test_dir/$f $output_dir/data/test/$f || exit 1;
done
echo "$0: AISHELL data preparation succeeded"
exit 0;
FunASR/examples/aishell/e_branchformer/local/download_and_untar.sh 0 → 100755
View file @ 70a8a9e0
#!/usr/bin/env bash
# Copyright 2014 Johns Hopkins University (author: Daniel Povey)
# 2017 Xingyu Na
# Apache 2.0
remove_archive=false
if [ "$1" == --remove-archive ]; then
remove_archive=true
shift
fi
if [ $# -ne 3 ]; then
echo "Usage: $0 [--remove-archive] <data-base> <url-base> <corpus-part>"
echo "e.g.: $0 /export/a05/xna/data www.openslr.org/resources/33 data_aishell"
echo "With --remove-archive it will remove the archive after successfully un-tarring it."
echo "<corpus-part> can be one of: data_aishell, resource_aishell."
fi
data=$1
url=$2
part=$3
if [ ! -d "$data" ]; then
echo "$0: no such directory $data"
exit 1;
fi
part_ok=false
list="data_aishell resource_aishell"
for x in $list; do
if [ "$part" == $x ]; then part_ok=true; fi
done
if ! $part_ok; then
echo "$0: expected <corpus-part> to be one of $list, but got '$part'"
exit 1;
fi
if [ -z "$url" ]; then
echo "$0: empty URL base."
exit 1;
fi
if [ -f $data/$part/.complete ]; then
echo "$0: data part $part was already successfully extracted, nothing to do."
exit 0;
fi
# sizes of the archive files in bytes.
sizes="15582913665 1246920"
if [ -f $data/$part.tgz ]; then
size=$(/bin/ls -l $data/$part.tgz | awk '{print $5}')
size_ok=false
for s in $sizes; do if [ $s == $size ]; then size_ok=true; fi; done
if ! $size_ok; then
echo "$0: removing existing file $data/$part.tgz because its size in bytes $size"
echo "does not equal the size of one of the archives."
rm $data/$part.tgz
else
echo "$data/$part.tgz exists and appears to be complete."
fi
fi
if [ ! -f $data/$part.tgz ]; then
if ! command -v wget >/dev/null; then
echo "$0: wget is not installed."
exit 1;
fi
full_url=$url/$part.tgz
echo "$0: downloading data from $full_url. This may take some time, please be patient."
cd $data || exit 1
if ! wget --no-check-certificate $full_url; then
echo "$0: error executing wget $full_url"
exit 1;
fi
fi
cd $data || exit 1
if ! tar -xvzf $part.tgz; then
echo "$0: error un-tarring archive $data/$part.tgz"
exit 1;
fi
touch $data/$part/.complete
if [ $part == "data_aishell" ]; then
cd $data/$part/wav || exit 1
for wav in ./*.tar.gz; do
echo "Extracting wav from $wav"
tar -zxf $wav && rm $wav
done
fi
echo "$0: Successfully downloaded and un-tarred $data/$part.tgz"
if $remove_archive; then
echo "$0: removing $data/$part.tgz file since --remove-archive option was supplied."
rm $data/$part.tgz
fi
exit 0;
FunASR/examples/aishell/e_branchformer/run.sh 0 → 100755
View file @ 70a8a9e0
#!/usr/bin/env bash
CUDA_VISIBLE_DEVICES="0,1,2,3"
# general configuration
feats_dir="../DATA" #feature output dictionary
exp_dir=`pwd`
lang=zh
token_type=char
stage=0
stop_stage=5
# feature configuration
nj=32
inference_device="cuda" #"cpu"
inference_checkpoint="model.pt.avg10"
inference_scp="wav.scp"
inference_batch_size=1
# data
raw_data=../raw_data
data_url=www.openslr.org/resources/33
# exp tag
tag="exp1"
workspace=`pwd`
master_port=12345
. utils/parse_options.sh || exit 1;
# Set bash to 'debug' mode, it will exit on :
# -e 'error', -u 'undefined variable', -o ... 'error in pipeline', -x 'print commands',
set -e
set -u
set -o pipefail
train_set=train
valid_set=dev
test_sets="dev test"
config=e_branchformer_12e_6d_2048_256.yaml
model_dir="baseline_$(basename "${config}" .yaml)_${lang}_${token_type}_${tag}"
if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
echo "stage -1: Data Download"
mkdir -p ${raw_data}
local/download_and_untar.sh ${raw_data} ${data_url} data_aishell
local/download_and_untar.sh ${raw_data} ${data_url} resource_aishell
fi
if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
echo "stage 0: Data preparation"
# Data preparation
local/aishell_data_prep.sh ${raw_data}/data_aishell/wav ${raw_data}/data_aishell/transcript ${feats_dir}
for x in train dev test; do
cp ${feats_dir}/data/${x}/text ${feats_dir}/data/${x}/text.org
paste -d " " <(cut -f 1 -d" " ${feats_dir}/data/${x}/text.org) <(cut -f 2- -d" " ${feats_dir}/data/${x}/text.org | tr -d " ") \
> ${feats_dir}/data/${x}/text
utils/text2token.py -n 1 -s 1 ${feats_dir}/data/${x}/text > ${feats_dir}/data/${x}/text.org
mv ${feats_dir}/data/${x}/text.org ${feats_dir}/data/${x}/text
# convert wav.scp text to jsonl
scp_file_list_arg="++scp_file_list='[\"${feats_dir}/data/${x}/wav.scp\",\"${feats_dir}/data/${x}/text\"]'"
python ../../../funasr/datasets/audio_datasets/scp2jsonl.py \
++data_type_list='["source", "target"]' \
++jsonl_file_out=${feats_dir}/data/${x}/audio_datasets.jsonl \
${scp_file_list_arg}
done
fi
if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
echo "stage 1: Feature and CMVN Generation"
python ../../../funasr/bin/compute_audio_cmvn.py \
--config-path "${workspace}/conf" \
--config-name "${config}" \
++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
++cmvn_file="${feats_dir}/data/${train_set}/cmvn.json" \
fi
token_list=${feats_dir}/data/${lang}_token_list/$token_type/tokens.txt
echo "dictionary: ${token_list}"
if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
echo "stage 2: Dictionary Preparation"
mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/
echo "make a dictionary"
echo "<blank>" > ${token_list}
echo "<s>" >> ${token_list}
echo "</s>" >> ${token_list}
utils/text2token.py -s 1 -n 1 --space "" ${feats_dir}/data/$train_set/text | cut -f 2- -d" " | tr " " "\n" \
| sort | uniq | grep -a -v -e '^\s*$' | awk '{print $0}' >> ${token_list}
echo "<unk>" >> ${token_list}
fi
# LM Training Stage
if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
echo "stage 3: LM Training"
fi
# ASR Training Stage
if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
echo "stage 4: ASR Training"
mkdir -p ${exp_dir}/exp/${model_dir}
current_time=$(date "+%Y-%m-%d_%H-%M")
log_file="${exp_dir}/exp/${model_dir}/train.log.txt.${current_time}"
echo "log_file: ${log_file}"
export CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES
gpu_num=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
torchrun \
--nnodes 1 \
--nproc_per_node ${gpu_num} \
--master_port ${master_port} \
../../../funasr/bin/train.py \
--config-path "${workspace}/conf" \
--config-name "${config}" \
++train_data_set_list="${feats_dir}/data/${train_set}/audio_datasets.jsonl" \
++valid_data_set_list="${feats_dir}/data/${valid_set}/audio_datasets.jsonl" \
++tokenizer_conf.token_list="${token_list}" \
++frontend_conf.cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
++output_dir="${exp_dir}/exp/${model_dir}" &> ${log_file}
fi
# Testing Stage
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
echo "stage 5: Inference"
if [ ${inference_device} == "cuda" ]; then
nj=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
else
inference_batch_size=1
CUDA_VISIBLE_DEVICES=""
for JOB in $(seq ${nj}); do
CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"-1,"
done
fi
for dset in ${test_sets}; do
inference_dir="${exp_dir}/exp/${model_dir}/inference-${inference_checkpoint}/${dset}"
_logdir="${inference_dir}/logdir"
echo "inference_dir: ${inference_dir}"
mkdir -p "${_logdir}"
data_dir="${feats_dir}/data/${dset}"
key_file=${data_dir}/${inference_scp}
split_scps=
for JOB in $(seq "${nj}"); do
split_scps+=" ${_logdir}/keys.${JOB}.scp"
done
utils/split_scp.pl "${key_file}" ${split_scps}
gpuid_list_array=(${CUDA_VISIBLE_DEVICES//,/ })
for JOB in $(seq ${nj}); do
{
id=$((JOB-1))
gpuid=${gpuid_list_array[$id]}
export CUDA_VISIBLE_DEVICES=${gpuid}
python ../../../funasr/bin/inference.py \
--config-path="${exp_dir}/exp/${model_dir}" \
--config-name="config.yaml" \
++init_param="${exp_dir}/exp/${model_dir}/${inference_checkpoint}" \
++tokenizer_conf.token_list="${token_list}" \
++frontend_conf.cmvn_file="${feats_dir}/data/${train_set}/am.mvn" \
++input="${_logdir}/keys.${JOB}.scp" \
++output_dir="${inference_dir}/${JOB}" \
++device="${inference_device}" \
++ncpu=1 \
++disable_log=true \
++batch_size="${inference_batch_size}" &> ${_logdir}/log.${JOB}.txt
}&
done
wait
mkdir -p ${inference_dir}/1best_recog
for f in token score text; do
if [ -f "${inference_dir}/${JOB}/1best_recog/${f}" ]; then
for JOB in $(seq "${nj}"); do
cat "${inference_dir}/${JOB}/1best_recog/${f}"
done | sort -k1 >"${inference_dir}/1best_recog/${f}"
fi
done
echo "Computing WER ..."
python utils/postprocess_text_zh.py ${inference_dir}/1best_recog/text ${inference_dir}/1best_recog/text.proc
python utils/postprocess_text_zh.py ${data_dir}/text ${inference_dir}/1best_recog/text.ref
python utils/compute_wer.py ${inference_dir}/1best_recog/text.ref ${inference_dir}/1best_recog/text.proc ${inference_dir}/1best_recog/text.cer
tail -n 3 ${inference_dir}/1best_recog/text.cer
done
fi
\ No newline at end of file
FunASR/examples/aishell/e_branchformer/utils 0 → 120000
View file @ 70a8a9e0
../paraformer/utils
\ No newline at end of file
FunASR/examples/aishell/paraformer/README.md 0 → 100644
View file @ 70a8a9e0
# Paraformer
pretrained model in [ModelScope](https://www.modelscope.cn/home)：[speech_paraformer_asr_nat-aishell1-pytorch](https://www.modelscope.cn/models/damo/speech_paraformer_asr_nat-aishell1-pytorch/summary)
## Training Config
- Feature info: using 80 dims fbank, global cmvn, speed perturb(0.9, 1.0, 1.1), specaugment
- Train info: lr 5e-4, batch_size 25000, 2 gpu(Tesla V100), acc_grad 1, 50 epochs
- Train config: conf/train_asr_paraformer_conformer_12e_6d_2048_256.yaml
- LM config: LM was not used
## Results (CER)
- Decode config: conf/decode_asr_transformer_noctc_1best.yaml (ctc weight:0.0)
| testset | CER(%) |
|:-----------:|:-------:|
| dev | 4.66 |
| test | 5.11 |
- Decode config: conf/decode_asr_transformer.yaml (ctc weight:0.5)
| testset | CER(%) |
|:-----------:|:-------:|
| dev | 4.52 |
| test | 4.94 |
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment