msb-public
/
PaddleSpeech
mirror of https://github.com/PaddlePaddle/PaddleSpeech
1
0
Code Issues Projects Releases Wiki Activity

Merge pull request #745 from PaddlePaddle/s2

Browse Source 
kaldi style data for u2 model
pull/748/head
Hui Zhang 3 years ago committed by GitHub
parent 9c00c5f804 dc44f552a1
commit 20d6f9ae4b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
16 changed files with 904 additions and 0 deletions
Show Stats Download Patch File Download Diff File

3
examples/librispeech/README.md
Unescape View File

@ -1,3 +1,6 @@
# ASR # ASR
* s0 is for deepspeech2 offline * s0 is for deepspeech2 offline
* s1 is for transformer/conformer/U2 * s1 is for transformer/conformer/U2
* s2 is for transformer/conformer/U2 w/ kaldi feat
need install Kaldi

42
examples/librispeech/s2/README.md
Unescape View File

@ -0,0 +1,42 @@
# LibriSpeech
## Data
| Data Subset | Duration in Seconds |
| data/manifest.train | 0.83s ~ 29.735s |
| data/manifest.dev | 1.065 ~ 35.155s |
| data/manifest.test-clean | 1.285s ~ 34.955s |
## Conformer
| Model | Params | Config | Augmentation| Test set | Decode method | Loss | WER |
| --- | --- | --- | --- | --- | --- | --- | --- |
| conformer | 47.63 M | conf/conformer.yaml | spec_aug + shift | test-clean | attention | - | - |
| conformer | 47.63 M | conf/conformer.yaml | spec_aug + shift | test-clean | ctc_greedy_search | | |
| conformer | 47.63 M | conf/conformer.yaml | spec_aug + shift | test-clean | ctc_prefix_beam_search | | |
| conformer | 47.63 M | conf/conformer.yaml | spec_aug + shift | test-clean | attention_rescoring | | |
### Test w/o length filter
| Model | Params | Config | Augmentation| Test set | Decode method | Loss | WER |
| --- | --- | --- | --- | --- | --- | --- | --- |
| conformer | 47.63 M | conf/conformer.yaml | spec_aug + shift | test-clean-all | attention | | |
## Chunk Conformer
| Model | Params | Config | Augmentation| Test set | Decode method | Chunk Size & Left Chunks | Loss | WER |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| conformer | 47.63 M | conf/chunk_conformer.yaml | spec_aug + shift | test-clean | attention | 16, -1 | | |
| conformer | 47.63 M | conf/chunk_conformer.yaml | spec_aug + shift | test-clean | ctc_greedy_search | 16, -1 | | |
| conformer | 47.63 M | conf/chunk_conformer.yaml | spec_aug + shift | test-clean | ctc_prefix_beam_search | 16, -1 | | - |
| conformer | 47.63 M | conf/chunk_conformer.yaml | spec_aug + shift | test-clean | attention_rescoring | 16, -1 | | - |
## Transformer
| Model | Params | Config | Augmentation| Test set | Decode method | Loss | WER |
| --- | --- | --- | --- | --- | --- | --- | --- |
| transformer | 32.52 M | conf/transformer.yaml | spec_aug + shift | test-clean | attention | | |
### Test w/o length filter
| Model | Params | Config | Augmentation| Test set | Decode method | Loss | WER |
| --- | --- | --- | --- | --- | --- | --- | --- |
| transformer | 32.52 M | conf/transformer.yaml | spec_aug + shift | test-clean-all | attention | | |

34
examples/librispeech/s2/conf/augmentation.json
Unescape View File

@ -0,0 +1,34 @@
[
{
"type": "shift",
"params": {
"min_shift_ms": -5,
"max_shift_ms": 5
},
"prob": 1.0
},
{
"type": "speed",
"params": {
"min_speed_rate": 0.9,
"max_speed_rate": 1.1,
"num_rates": 3
},
"prob": 0.0
},
{
"type": "specaug",
"params": {
"F": 10,
"T": 50,
"n_freq_masks": 2,
"n_time_masks": 2,
"p": 1.0,
"W": 80,
"adaptive_number_ratio": 0,
"adaptive_size_ratio": 0,
"max_n_time_masks": 20
},
"prob": 1.0
}
]

120
examples/librispeech/s2/conf/chunk_conformer.yaml
Unescape View File

@ -0,0 +1,120 @@
# https://yaml.org/type/float.html
data:
train_manifest: data/manifest.train
dev_manifest: data/manifest.dev
test_manifest: data/manifest.test
min_input_len: 0.5
max_input_len: 20.0
min_output_len: 0.0
max_output_len: 400.0
min_output_input_ratio: 0.05
max_output_input_ratio: 10.0
collator:
vocab_filepath: data/vocab.txt
unit_type: 'spm'
spm_model_prefix: 'data/bpe_unigram_5000'
mean_std_filepath: ""
augmentation_config: conf/augmentation.json
batch_size: 16
raw_wav: True # use raw_wav or kaldi feature
specgram_type: fbank #linear, mfcc, fbank
feat_dim: 80
delta_delta: False
dither: 1.0
target_sample_rate: 16000
max_freq: None
n_fft: None
stride_ms: 10.0
window_ms: 25.0
use_dB_normalization: True
target_dB: -20
random_seed: 0
keep_transcription_text: False
sortagrad: True
shuffle_method: batch_shuffle
num_workers: 2
# network architecture
model:
cmvn_file: "data/mean_std.json"
cmvn_file_type: "json"
# encoder related
encoder: conformer
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 input type, you can chose conv2d, conv2d6 and conv2d8
normalize_before: True
use_cnn_module: True
cnn_module_kernel: 15
activation_type: 'swish'
pos_enc_layer_type: 'rel_pos'
selfattention_layer_type: 'rel_selfattn'
causal: True
use_dynamic_chunk: true
cnn_module_norm: 'layer_norm' # using nn.LayerNorm makes model converge faster
use_dynamic_left_chunk: false
# decoder related
decoder: transformer
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
# hybrid CTC/attention
model_conf:
ctc_weight: 0.3
lsm_weight: 0.1 # label smoothing option
length_normalized_loss: false
training:
n_epoch: 240
accum_grad: 8
global_grad_clip: 5.0
optim: adam
optim_conf:
lr: 0.001
weight_decay: 1e-06
scheduler: warmuplr # pytorch v1.1.0+ required
scheduler_conf:
warmup_steps: 25000
lr_decay: 1.0
log_interval: 100
checkpoint:
kbest_n: 50
latest_n: 5
decoding:
batch_size: 128
error_rate_type: wer
decoding_method: attention # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
lang_model_path: data/lm/common_crawl_00.prune01111.trie.klm
alpha: 2.5
beta: 0.3
beam_size: 10
cutoff_prob: 1.0
cutoff_top_n: 0
num_proc_bsearch: 8
ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
# <0: for decoding, use full chunk.
# >0: for decoding, use fixed chunk size as set.
# 0: used for training, it's prohibited here.
num_decoding_left_chunks: -1 # number of left chunks for decoding. Defaults to -1.
simulate_streaming: true # simulate streaming inference. Defaults to False.

113
examples/librispeech/s2/conf/chunk_transformer.yaml
Unescape View File

@ -0,0 +1,113 @@
# https://yaml.org/type/float.html
data:
train_manifest: data/manifest.train
dev_manifest: data/manifest.dev
test_manifest: data/manifest.test
min_input_len: 0.5 # second
max_input_len: 20.0 # second
min_output_len: 0.0 # tokens
max_output_len: 400.0 # tokens
min_output_input_ratio: 0.05
max_output_input_ratio: 10.0
collator:
vocab_filepath: data/vocab.txt
unit_type: 'spm'
spm_model_prefix: 'data/bpe_unigram_5000'
mean_std_filepath: ""
augmentation_config: conf/augmentation.json
batch_size: 64
raw_wav: True # use raw_wav or kaldi feature
specgram_type: fbank #linear, mfcc, fbank
feat_dim: 80
delta_delta: False
dither: 1.0
target_sample_rate: 16000
max_freq: None
n_fft: None
stride_ms: 10.0
window_ms: 25.0
use_dB_normalization: True
target_dB: -20
random_seed: 0
keep_transcription_text: False
sortagrad: True
shuffle_method: batch_shuffle
num_workers: 2
# network architecture
model:
cmvn_file: "data/mean_std.json"
cmvn_file_type: "json"
# encoder related
encoder: transformer
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 input type, you can chose conv2d, conv2d6 and conv2d8
normalize_before: true
use_dynamic_chunk: true
use_dynamic_left_chunk: false
# decoder related
decoder: transformer
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
# hybrid CTC/attention
model_conf:
ctc_weight: 0.3
lsm_weight: 0.1 # label smoothing option
length_normalized_loss: false
training:
n_epoch: 120
accum_grad: 1
global_grad_clip: 5.0
optim: adam
optim_conf:
lr: 0.001
weight_decay: 1e-06
scheduler: warmuplr # pytorch v1.1.0+ required
scheduler_conf:
warmup_steps: 25000
lr_decay: 1.0
log_interval: 100
checkpoint:
kbest_n: 50
latest_n: 5
decoding:
batch_size: 64
error_rate_type: wer
decoding_method: attention # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
lang_model_path: data/lm/common_crawl_00.prune01111.trie.klm
alpha: 2.5
beta: 0.3
beam_size: 10
cutoff_prob: 1.0
cutoff_top_n: 0
num_proc_bsearch: 8
ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
# <0: for decoding, use full chunk.
# >0: for decoding, use fixed chunk size as set.
# 0: used for training, it's prohibited here.
num_decoding_left_chunks: -1 # number of left chunks for decoding. Defaults to -1.
simulate_streaming: true # simulate streaming inference. Defaults to False.

116
examples/librispeech/s2/conf/conformer.yaml
Unescape View File

@ -0,0 +1,116 @@
# https://yaml.org/type/float.html
data:
train_manifest: data/manifest.train
dev_manifest: data/manifest.dev
test_manifest: data/manifest.test-clean
min_input_len: 0.5 # seconds
max_input_len: 20.0 # seconds
min_output_len: 0.0 # tokens
max_output_len: 400.0 # tokens
min_output_input_ratio: 0.05
max_output_input_ratio: 10.0
collator:
vocab_filepath: data/vocab.txt
unit_type: 'spm'
spm_model_prefix: 'data/bpe_unigram_5000'
mean_std_filepath: ""
augmentation_config: conf/augmentation.json
batch_size: 16
raw_wav: True # use raw_wav or kaldi feature
specgram_type: fbank #linear, mfcc, fbank
feat_dim: 80
delta_delta: False
dither: 1.0
target_sample_rate: 16000
max_freq: None
n_fft: None
stride_ms: 10.0
window_ms: 25.0
use_dB_normalization: True
target_dB: -20
random_seed: 0
keep_transcription_text: False
sortagrad: True
shuffle_method: batch_shuffle
num_workers: 2
# network architecture
model:
cmvn_file: "data/mean_std.json"
cmvn_file_type: "json"
# encoder related
encoder: conformer
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 input type, you can chose conv2d, conv2d6 and conv2d8
normalize_before: True
use_cnn_module: True
cnn_module_kernel: 15
activation_type: 'swish'
pos_enc_layer_type: 'rel_pos'
selfattention_layer_type: 'rel_selfattn'
# decoder related
decoder: transformer
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
# hybrid CTC/attention
model_conf:
ctc_weight: 0.3
lsm_weight: 0.1 # label smoothing option
length_normalized_loss: false
training:
n_epoch: 120
accum_grad: 8
global_grad_clip: 3.0
optim: adam
optim_conf:
lr: 0.004
weight_decay: 1e-06
scheduler: warmuplr # pytorch v1.1.0+ required
scheduler_conf:
warmup_steps: 25000
lr_decay: 1.0
log_interval: 100
checkpoint:
kbest_n: 50
latest_n: 5
decoding:
batch_size: 64
error_rate_type: wer
decoding_method: attention # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
lang_model_path: data/lm/common_crawl_00.prune01111.trie.klm
alpha: 2.5
beta: 0.3
beam_size: 10
cutoff_prob: 1.0
cutoff_top_n: 0
num_proc_bsearch: 8
ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
# <0: for decoding, use full chunk.
# >0: for decoding, use fixed chunk size as set.
# 0: used for training, it's prohibited here.
num_decoding_left_chunks: -1 # number of left chunks for decoding. Defaults to -1.
simulate_streaming: False # simulate streaming inference. Defaults to False.

111
examples/librispeech/s2/conf/transformer.yaml
Unescape View File

@ -0,0 +1,111 @@
# https://yaml.org/type/float.html
data:
train_manifest: data/manifest.train
dev_manifest: data/manifest.dev
test_manifest: data/manifest.test-clean
min_input_len: 0.5 # second
max_input_len: 20.0 # second
min_output_len: 0.0 # tokens
max_output_len: 400.0 # tokens
min_output_input_ratio: 0.05
max_output_input_ratio: 10.0
collator:
vocab_filepath: data/vocab.txt
unit_type: 'spm'
spm_model_prefix: 'data/bpe_unigram_5000'
mean_std_filepath: ""
augmentation_config: conf/augmentation.json
batch_size: 64
raw_wav: True # use raw_wav or kaldi feature
specgram_type: fbank #linear, mfcc, fbank
feat_dim: 80
delta_delta: False
dither: 1.0
target_sample_rate: 16000
max_freq: None
n_fft: None
stride_ms: 10.0
window_ms: 25.0
use_dB_normalization: True
target_dB: -20
random_seed: 0
keep_transcription_text: False
sortagrad: True
shuffle_method: batch_shuffle
num_workers: 2
# network architecture
model:
cmvn_file: "data/mean_std.json"
cmvn_file_type: "json"
# encoder related
encoder: transformer
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 input type, you can chose conv2d, conv2d6 and conv2d8
normalize_before: true
# decoder related
decoder: transformer
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
# hybrid CTC/attention
model_conf:
ctc_weight: 0.3
lsm_weight: 0.1 # label smoothing option
length_normalized_loss: false
training:
n_epoch: 120
accum_grad: 2
global_grad_clip: 5.0
optim: adam
optim_conf:
lr: 0.004
weight_decay: 1e-06
scheduler: warmuplr # pytorch v1.1.0+ required
scheduler_conf:
warmup_steps: 25000
lr_decay: 1.0
log_interval: 100
checkpoint:
kbest_n: 50
latest_n: 5
decoding:
batch_size: 64
error_rate_type: wer
decoding_method: attention # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
lang_model_path: data/lm/common_crawl_00.prune01111.trie.klm
alpha: 2.5
beta: 0.3
beam_size: 10
cutoff_prob: 1.0
cutoff_top_n: 0
num_proc_bsearch: 8
ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
# <0: for decoding, use full chunk.
# >0: for decoding, use fixed chunk size as set.
# 0: used for training, it's prohibited here.
num_decoding_left_chunks: -1 # number of left chunks for decoding. Defaults to -1.
simulate_streaming: False # simulate streaming inference. Defaults to False.

37
examples/librispeech/s2/local/align.sh
Unescape View File

@ -0,0 +1,37 @@
#!/bin/bash
if [ $# != 2 ];then
echo "usage: ${0} config_path ckpt_path_prefix"
exit -1
fi
ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
echo "using $ngpu gpus..."
device=gpu
if [ ${ngpu} == 0 ];then
device=cpu
fi
config_path=$1
ckpt_prefix=$2
batch_size=1
output_dir=${ckpt_prefix}
mkdir -p ${output_dir}
# align dump in `result_file`
# .tier, .TextGrid dump in `dir of result_file`
python3 -u ${BIN_DIR}/alignment.py \
--device ${device} \
--nproc 1 \
--config ${config_path} \
--result_file ${output_dir}/${type}.align \
--checkpoint_path ${ckpt_prefix} \
--opts decoding.batch_size ${batch_size}
if [ $? -ne 0 ]; then
echo "Failed in ctc alignment!"
exit 1
fi
exit 0

111
examples/librispeech/s2/local/data.sh
Unescape View File

@ -0,0 +1,111 @@
#!/bin/bash
stage=-1
stop_stage=100
# bpemode (unigram or bpe)
nbpe=5000
bpemode=unigram
bpeprefix="data/bpe_${bpemode}_${nbpe}"
source ${MAIN_ROOT}/utils/parse_options.sh
mkdir -p data
TARGET_DIR=${MAIN_ROOT}/examples/dataset
mkdir -p ${TARGET_DIR}
if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
# download data, generate manifests
python3 ${TARGET_DIR}/librispeech/librispeech.py \
--manifest_prefix="data/manifest" \
--target_dir="${TARGET_DIR}/librispeech" \
--full_download="True"
if [ $? -ne 0 ]; then
echo "Prepare LibriSpeech failed. Terminated."
exit 1
fi
for set in train-clean-100 train-clean-360 train-other-500 dev-clean dev-other test-clean test-other; do
mv data/manifest.${set} data/manifest.${set}.raw
done
rm -rf data/manifest.train.raw data/manifest.dev.raw data/manifest.test.raw
for set in train-clean-100 train-clean-360 train-other-500; do
cat data/manifest.${set}.raw >> data/manifest.train.raw
done
for set in dev-clean dev-other; do
cat data/manifest.${set}.raw >> data/manifest.dev.raw
done
for set in test-clean test-other; do
cat data/manifest.${set}.raw >> data/manifest.test.raw
done
fi
if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
# build vocabulary
python3 ${MAIN_ROOT}/utils/build_vocab.py \
--unit_type "spm" \
--spm_vocab_size=${nbpe} \
--spm_mode ${bpemode} \
--spm_model_prefix ${bpeprefix} \
--vocab_path="data/vocab.txt" \
--manifest_paths="data/manifest.train.raw"
if [ $? -ne 0 ]; then
echo "Build vocabulary failed. Terminated."
exit 1
fi
fi
if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
# compute mean and stddev for normalizer
num_workers=$(nproc)
python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
--manifest_path="data/manifest.train.raw" \
--num_samples=-1 \
--specgram_type="fbank" \
--feat_dim=80 \
--delta_delta=false \
--sample_rate=16000 \
--stride_ms=10.0 \
--window_ms=25.0 \
--use_dB_normalization=False \
--num_workers=${num_workers} \
--output_path="data/mean_std.json"
if [ $? -ne 0 ]; then
echo "Compute mean and stddev failed. Terminated."
exit 1
fi
fi
if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
# format manifest with tokenids, vocab size
for set in train dev test dev-clean dev-other test-clean test-other; do
{
python3 ${MAIN_ROOT}/utils/format_data.py \
--feat_type "raw" \
--cmvn_path "data/mean_std.json" \
--unit_type "spm" \
--spm_model_prefix ${bpeprefix} \
--vocab_path="data/vocab.txt" \
--manifest_path="data/manifest.${set}.raw" \
--output_path="data/manifest.${set}"
if [ $? -ne 0 ]; then
echo "Formt mnaifest failed. Terminated."
exit 1
fi
}&
done
wait
fi
echo "LibriSpeech Data preparation done."
exit 0

20
examples/librispeech/s2/local/download_lm_en.sh
Unescape View File

@ -0,0 +1,20 @@
#!/bin/bash
. ${MAIN_ROOT}/utils/utility.sh
DIR=data/lm
mkdir -p ${DIR}
URL=https://deepspeech.bj.bcebos.com/en_lm/common_crawl_00.prune01111.trie.klm
MD5="099a601759d467cd0a8523ff939819c5"
TARGET=${DIR}/common_crawl_00.prune01111.trie.klm
echo "Download language model ..."
download $URL $MD5 $TARGET
if [ $? -ne 0 ]; then
echo "Fail to download the language model!"
exit 1
fi
exit 0

34
examples/librispeech/s2/local/export.sh
Unescape View File

@ -0,0 +1,34 @@
#!/bin/bash
if [ $# != 3 ];then
echo "usage: $0 config_path ckpt_prefix jit_model_path"
exit -1
fi
ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
echo "using $ngpu gpus..."
config_path=$1
ckpt_path_prefix=$2
jit_model_export_path=$3
device=gpu
if [ ${ngpu} == 0 ];then
device=cpu
fi
python3 -u ${BIN_DIR}/export.py \
--device ${device} \
--nproc ${ngpu} \
--config ${config_path} \
--checkpoint_path ${ckpt_path_prefix} \
--export_path ${jit_model_export_path}
if [ $? -ne 0 ]; then
echo "Failed in export!"
exit 1
fi
exit 0

72
examples/librispeech/s2/local/test.sh
Unescape View File

@ -0,0 +1,72 @@
#!/bin/bash
if [ $# != 2 ];then
echo "usage: ${0} config_path ckpt_path_prefix"
exit -1
fi
ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
echo "using $ngpu gpus..."
device=gpu
if [ ${ngpu} == 0 ];then
device=cpu
fi
config_path=$1
ckpt_prefix=$2
chunk_mode=false
if [[ ${config_path} =~ ^.*chunk_.*yaml$ ]];then
chunk_mode=true
fi
echo "chunk mode ${chunk_mode}"
# download language model
#bash local/download_lm_en.sh
#if [ $? -ne 0 ]; then
# exit 1
#fi
for type in attention ctc_greedy_search; do
echo "decoding ${type}"
if [ ${chunk_mode} == true ];then
# stream decoding only support batchsize=1
batch_size=1
else
batch_size=64
fi
python3 -u ${BIN_DIR}/test.py \
--device ${device} \
--nproc 1 \
--config ${config_path} \
--result_file ${ckpt_prefix}.${type}.rsl \
--checkpoint_path ${ckpt_prefix} \
--opts decoding.decoding_method ${type} decoding.batch_size ${batch_size}
if [ $? -ne 0 ]; then
echo "Failed in evaluation!"
exit 1
fi
done
for type in ctc_prefix_beam_search attention_rescoring; do
echo "decoding ${type}"
batch_size=1
python3 -u ${BIN_DIR}/test.py \
--device ${device} \
--nproc 1 \
--config ${config_path} \
--result_file ${ckpt_prefix}.${type}.rsl \
--checkpoint_path ${ckpt_prefix} \
--opts decoding.decoding_method ${type} decoding.batch_size ${batch_size}
if [ $? -ne 0 ]; then
echo "Failed in evaluation!"
exit 1
fi
done
exit 0

33
examples/librispeech/s2/local/train.sh
Unescape View File

@ -0,0 +1,33 @@
#!/bin/bash
if [ $# != 2 ];then
echo "usage: CUDA_VISIBLE_DEVICES=0 ${0} config_path ckpt_name"
exit -1
fi
ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
echo "using $ngpu gpus..."
config_path=$1
ckpt_name=$2
device=gpu
if [ ${ngpu} == 0 ];then
device=cpu
fi
echo "using ${device}..."
mkdir -p exp
python3 -u ${BIN_DIR}/train.py \
--device ${device} \
--nproc ${ngpu} \
--config ${config_path} \
--output exp/${ckpt_name}
if [ $? -ne 0 ]; then
echo "Failed in training!"
exit 1
fi
exit 0

14
examples/librispeech/s2/path.sh
Unescape View File

@ -0,0 +1,14 @@
export MAIN_ROOT=${PWD}/../../../
export PATH=${MAIN_ROOT}:${PWD}/utils:${PATH}
export LC_ALL=C
# Use UTF-8 in Python to avoid UnicodeDecodeError when LC_ALL=C
export PYTHONIOENCODING=UTF-8
export PYTHONPATH=${MAIN_ROOT}:${PYTHONPATH}
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib/
MODEL=u2
export BIN_DIR=${MAIN_ROOT}/deepspeech/exps/${MODEL}/bin

43
examples/librispeech/s2/run.sh
Unescape View File

@ -0,0 +1,43 @@
#!/bin/bash
set -e
source path.sh
stage=0
stop_stage=100
conf_path=conf/transformer.yaml
avg_num=30
source ${MAIN_ROOT}/utils/parse_options.sh || exit 1;
avg_ckpt=avg_${avg_num}
ckpt=$(basename ${conf_path} | awk -F'.' '{print $1}')
echo "checkpoint name ${ckpt}"
if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
# prepare data
bash ./local/data.sh || exit -1
fi
if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
# train model, all `ckpt` under `exp` dir
CUDA_VISIBLE_DEVICES=4,5,6,7 ./local/train.sh ${conf_path} ${ckpt}
fi
if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
# avg n best model
avg.sh exp/${ckpt}/checkpoints ${avg_num}
fi
if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
# test ckpt avg_n
CUDA_VISIBLE_DEVICES=7 ./local/test.sh ${conf_path} exp/${ckpt}/checkpoints/${avg_ckpt} || exit -1
fi
if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then
# ctc alignment of test data
CUDA_VISIBLE_DEVICES=0 ./local/align.sh ${conf_path} exp/${ckpt}/checkpoints/${avg_ckpt} || exit -1
fi
if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then
# export ckpt avg_n
CUDA_VISIBLE_DEVICES= ./local/export.sh ${conf_path} exp/${ckpt}/checkpoints/${avg_ckpt} exp/${ckpt}/checkpoints/${avg_ckpt}.jit
fi

1
examples/librispeech/s2/utils
Unescape View File

@ -0,0 +1 @@
../../../utils/
Write Preview
Loading…
Cancel
Save