th.zhang
2 years ago
parent
958bfe5403
commit
559627c8de
@ -0,0 +1,9 @@
|
||||
# LibriSpeech
|
||||
|
||||
## hubertASR
|
||||
Fintuning on train-clean-100
|
||||
train: Epoch 3, 1*V100-32G, batchsize: 4, accum_grad: 8
|
||||
|
||||
| Model | Params | Config | Augmentation| Test set | Decode method | WER |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| hubertASR | 326.16M | conf/hubertASR.yaml | spec_aug | test-clean | greedy search | 0.05868 |
|
||||
@ -0,0 +1,142 @@
|
||||
############################################
|
||||
# Network Architecture #
|
||||
############################################
|
||||
freeze_hubert: False
|
||||
normalize_wav: True
|
||||
output_norm: True
|
||||
init_type: kaiming_uniform # !Warning: need to convergence
|
||||
enc:
|
||||
input_shape: 1024
|
||||
dnn_blocks: 2
|
||||
dnn_neurons: 1024
|
||||
activation: True
|
||||
ctc:
|
||||
enc_n_units: 1024
|
||||
blank_id: 0
|
||||
dropout_rate: 0.0
|
||||
hubert_params_path: "exp/hubert/pd_hubert_no_fintune.pdparams"
|
||||
|
||||
|
||||
task_cfg:
|
||||
label_rate: 50.0
|
||||
sample_rate: 16000
|
||||
normalize: True
|
||||
enable_padding: False
|
||||
max_keep_size: None
|
||||
max_sample_size: 250000
|
||||
min_sample_size: 32000
|
||||
single_target: False
|
||||
random_crop: True
|
||||
pad_audio: False
|
||||
|
||||
model_cfg:
|
||||
dropout_input: 0.0
|
||||
final_dropout: 0.0
|
||||
dropout: 0.0
|
||||
attention_dropout: 0.0
|
||||
activation_dropout: 0.1
|
||||
apply_mask: True
|
||||
mask_length: 10
|
||||
mask_prob: 0.5
|
||||
mask_selection: static
|
||||
mask_other: 0.0
|
||||
no_mask_overlap: False
|
||||
mask_channel_length: 64
|
||||
mask_channel_prob: 0.25
|
||||
mask_channel_selection: static
|
||||
mask_channel_other: 0.0
|
||||
no_mask_channel_overlap: False
|
||||
feature_grad_mult: 0.0
|
||||
layerdrop: 0.1
|
||||
normalize: True
|
||||
fp16: True
|
||||
label_rate: 50
|
||||
extractor_mode: layer_norm
|
||||
encoder_layers: 24
|
||||
encoder_embed_dim: 1024
|
||||
encoder_ffn_embed_dim: 4096
|
||||
encoder_attention_heads: 16
|
||||
activation_fn: gelu
|
||||
encoder_layerdrop: 0.1
|
||||
dropout_features: 0.0
|
||||
final_dim: 768
|
||||
untie_final_proj: True
|
||||
layer_norm_first: True
|
||||
conv_feature_layers: "[(512,10,5)] + [(512,3,2)] * 4 + [(512,2,2)] * 2"
|
||||
conv_bias: False
|
||||
logit_temp: 0.1
|
||||
target_glu: False
|
||||
mask_min_space: 1
|
||||
mask_channel_min_space: 1
|
||||
conv_pos: 128
|
||||
conv_pos_groups: 16
|
||||
latent_temp: [2.0, 0.5, 0.999995]
|
||||
skip_masked: False
|
||||
skip_nomask: True
|
||||
|
||||
###########################################
|
||||
# Data #
|
||||
###########################################
|
||||
train_manifest: data/manifest.train-clean-100
|
||||
dev_manifest: data/manifest.dev
|
||||
test_manifest: data/manifest.test-clean
|
||||
|
||||
###########################################
|
||||
# Dataloader #
|
||||
###########################################
|
||||
vocab_filepath: data/lang_char/vocab.txt
|
||||
unit_type: char
|
||||
mean_std_filepath: ""
|
||||
preprocess_config: conf/preprocess.yaml
|
||||
sortagrad: -1 # Feed samples from shortest to longest ; -1: enabled for all epochs 0: disabled other: enabled for other epochs
|
||||
batch_size: 2 # Different batch_size may cause large differences in results
|
||||
maxlen_in: 51200000000 # if input length > maxlen-in batchsize is automatically reduced
|
||||
maxlen_out: 1500000 # if output length > maxlen-out batchsize is automatically reduced
|
||||
minibatches: 0 # for debug
|
||||
batch_count: auto
|
||||
batch_bins: 0
|
||||
batch_frames_in: 0
|
||||
batch_frames_out: 0
|
||||
batch_frames_inout: 0
|
||||
num_workers: 0
|
||||
subsampling_factor: 1
|
||||
num_encs: 1
|
||||
dist_sampler: True
|
||||
shortest_first: True
|
||||
return_lens_rate: True
|
||||
|
||||
############################################
|
||||
# Data Augmentation #
|
||||
############################################
|
||||
audio_augment: # for raw audio
|
||||
sample_rate: 16000
|
||||
speeds: [95, 100, 105]
|
||||
|
||||
###########################################
|
||||
# Training #
|
||||
###########################################
|
||||
n_epoch: 3
|
||||
accum_grad: 8
|
||||
global_grad_clip: 5.0
|
||||
model_optim: adadelta
|
||||
model_optim_conf:
|
||||
lr: 1.0
|
||||
epsilon: 1.0e-6
|
||||
rho: 0.95
|
||||
model_scheduler: constantlr
|
||||
model_scheduler_conf:
|
||||
warmup_steps: 25000
|
||||
lr_decay: 1.0
|
||||
hubert_optim: adadelta
|
||||
hubert_optim_conf:
|
||||
lr: 1.0
|
||||
epsilon: 1.0e-6
|
||||
rho: 0.95
|
||||
hubert_scheduler: constantlr
|
||||
hubert_scheduler_conf:
|
||||
warmup_steps: 25000
|
||||
lr_decay: 1.0
|
||||
log_interval: 1
|
||||
checkpoint:
|
||||
kbest_n: 50
|
||||
latest_n: 5
|
||||
@ -0,0 +1,110 @@
|
||||
#!/bin/bash
|
||||
|
||||
stage=-1
|
||||
stop_stage=100
|
||||
|
||||
unit_type=char
|
||||
dict_dir=data/lang_char
|
||||
|
||||
source ${MAIN_ROOT}/utils/parse_options.sh
|
||||
|
||||
mkdir -p data
|
||||
mkdir -p ${dict_dir}
|
||||
TARGET_DIR=${MAIN_ROOT}/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 ] && [ ${stop_stage} -ge 0 ]; then
|
||||
# compute mean and stddev for normalizer
|
||||
num_workers=$(nproc)
|
||||
python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
|
||||
--manifest_path="data/manifest.train.raw" \
|
||||
--num_samples=2000 \
|
||||
--spectrum_type="fbank" \
|
||||
--feat_dim=161 \
|
||||
--delta_delta=false \
|
||||
--sample_rate=16000 \
|
||||
--stride_ms=10 \
|
||||
--window_ms=25 \
|
||||
--use_dB_normalization=False \
|
||||
--num_workers=${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 1 ] && [ ${stop_stage} -ge 1 ]; then
|
||||
# build vocabulary
|
||||
python3 ${MAIN_ROOT}/utils/build_vocab.py \
|
||||
--unit_type ${unit_type} \
|
||||
--count_threshold=0 \
|
||||
--vocab_path="${dict_dir}/vocab.txt" \
|
||||
--manifest_paths="data/manifest.train.raw"
|
||||
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "Build vocabulary 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 \
|
||||
--cmvn_path "data/mean_std.json" \
|
||||
--unit_type ${unit_type} \
|
||||
--vocab_path="${dict_dir}/vocab.txt" \
|
||||
--manifest_path="data/manifest.${set}.raw" \
|
||||
--output_path="data/manifest.${set}"
|
||||
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "Formt mnaifest.${set} failed. Terminated."
|
||||
exit 1
|
||||
fi
|
||||
}&
|
||||
done
|
||||
wait
|
||||
fi
|
||||
|
||||
echo "LibriSpeech Data preparation done."
|
||||
|
||||
if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then
|
||||
mkdir -p exp/hubert
|
||||
echo "Pretrained hubert model download"
|
||||
wget -P exp/hubert https://paddlespeech.bj.bcebos.com/hubert/hubert-large-lv60.pdparams
|
||||
fi
|
||||
|
||||
exit 0
|
||||
@ -0,0 +1,83 @@
|
||||
#!/bin/bash
|
||||
|
||||
set -e
|
||||
|
||||
ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
|
||||
echo "using $ngpu gpus..."
|
||||
|
||||
expdir=exp
|
||||
datadir=data
|
||||
|
||||
recog_set="test-clean test-other dev-clean dev-other"
|
||||
recog_set="test-clean"
|
||||
|
||||
config_path=$1
|
||||
decode_config_path=$2
|
||||
ckpt_prefix=$3
|
||||
|
||||
source ${MAIN_ROOT}/utils/parse_options.sh || exit 1;
|
||||
|
||||
# download language model
|
||||
#bash local/download_lm_en.sh
|
||||
#if [ $? -ne 0 ]; then
|
||||
# exit 1
|
||||
#fi
|
||||
|
||||
python3 utils/format_rsl.py \
|
||||
--origin_ref data/manifest.test-clean.raw \
|
||||
--trans_ref data/manifest.test-clean.text
|
||||
|
||||
|
||||
for type in ctc_greedy_search; do
|
||||
echo "decoding ${type}"
|
||||
batch_size=16
|
||||
python3 -u ${BIN_DIR}/test.py \
|
||||
--ngpu ${ngpu} \
|
||||
--config ${config_path} \
|
||||
--decode_cfg ${decode_config_path} \
|
||||
--result_file ${ckpt_prefix}.${type}.rsl \
|
||||
--checkpoint_path ${ckpt_prefix} \
|
||||
--opts decode.decoding_method ${type} \
|
||||
--opts decode.decode_batch_size ${batch_size}
|
||||
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "Failed in evaluation!"
|
||||
exit 1
|
||||
fi
|
||||
python3 utils/format_rsl.py \
|
||||
--origin_hyp ${ckpt_prefix}.${type}.rsl \
|
||||
--trans_hyp ${ckpt_prefix}.${type}.rsl.text
|
||||
|
||||
python3 utils/compute-wer.py --char=1 --v=1 \
|
||||
data/manifest.test-clean.text ${ckpt_prefix}.${type}.rsl.text > ${ckpt_prefix}.${type}.error
|
||||
echo "decoding ${type} done."
|
||||
done
|
||||
|
||||
for type in ctc_prefix_beam_search; do
|
||||
echo "decoding ${type}"
|
||||
batch_size=1
|
||||
python3 -u ${BIN_DIR}/test.py \
|
||||
--ngpu ${ngpu} \
|
||||
--config ${config_path} \
|
||||
--decode_cfg ${decode_config_path} \
|
||||
--result_file ${ckpt_prefix}.${type}.rsl \
|
||||
--checkpoint_path ${ckpt_prefix} \
|
||||
--opts decode.decoding_method ${type} \
|
||||
--opts decode.decode_batch_size ${batch_size}
|
||||
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "Failed in evaluation!"
|
||||
exit 1
|
||||
fi
|
||||
python3 utils/format_rsl.py \
|
||||
--origin_hyp ${ckpt_prefix}.${type}.rsl \
|
||||
--trans_hyp ${ckpt_prefix}.${type}.rsl.text
|
||||
|
||||
python3 utils/compute-wer.py --char=1 --v=1 \
|
||||
data/manifest.test-clean.text ${ckpt_prefix}.${type}.rsl.text > ${ckpt_prefix}.${type}.error
|
||||
echo "decoding ${type} done."
|
||||
done
|
||||
|
||||
echo "Finished"
|
||||
|
||||
exit 0
|
||||
@ -0,0 +1,58 @@
|
||||
#!/bin/bash
|
||||
|
||||
if [ $# != 4 ];then
|
||||
echo "usage: ${0} config_path decode_config_path ckpt_path_prefix audio_file"
|
||||
exit -1
|
||||
fi
|
||||
|
||||
ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
|
||||
echo "using $ngpu gpus..."
|
||||
|
||||
config_path=$1
|
||||
decode_config_path=$2
|
||||
ckpt_prefix=$3
|
||||
audio_file=$4
|
||||
|
||||
mkdir -p data
|
||||
wget -nc https://paddlespeech.bj.bcebos.com/datasets/single_wav/en/demo_002_en.wav -P data/
|
||||
if [ $? -ne 0 ]; then
|
||||
exit 1
|
||||
fi
|
||||
|
||||
if [ ! -f ${audio_file} ]; then
|
||||
echo "Plase input the right audio_file path"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
chunk_mode=false
|
||||
if [[ ${config_path} =~ ^.*chunk_.*yaml$ ]];then
|
||||
chunk_mode=true
|
||||
fi
|
||||
|
||||
# download language model
|
||||
#bash local/download_lm_ch.sh
|
||||
#if [ $? -ne 0 ]; then
|
||||
# exit 1
|
||||
#fi
|
||||
|
||||
for type in ctc_greedy_search; do
|
||||
echo "decoding ${type}"
|
||||
batch_size=1
|
||||
output_dir=${ckpt_prefix}
|
||||
mkdir -p ${output_dir}
|
||||
python3 -u ${BIN_DIR}/test_wav.py \
|
||||
--ngpu ${ngpu} \
|
||||
--config ${config_path} \
|
||||
--decode_cfg ${decode_config_path} \
|
||||
--result_file ${output_dir}/${type}.rsl \
|
||||
--checkpoint_path ${ckpt_prefix} \
|
||||
--opts decode.decoding_method ${type} \
|
||||
--opts decode.decode_batch_size ${batch_size} \
|
||||
--audio_file ${audio_file}
|
||||
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "Failed in evaluation!"
|
||||
exit 1
|
||||
fi
|
||||
done
|
||||
exit 0
|
||||
@ -0,0 +1,58 @@
|
||||
#!/bin/bash
|
||||
|
||||
if [ $# -lt 2 ] && [ $# -gt 3 ];then
|
||||
echo "usage: CUDA_VISIBLE_DEVICES=0 ${0} config_path ckpt_name ips(optional)"
|
||||
exit -1
|
||||
fi
|
||||
|
||||
ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
|
||||
echo "using $ngpu gpus..."
|
||||
|
||||
config_path=$1
|
||||
ckpt_name=$2
|
||||
resume=$3
|
||||
ips=$4
|
||||
|
||||
if [ ! $ips ];then
|
||||
ips_config=
|
||||
else
|
||||
ips_config="--ips="${ips}
|
||||
fi
|
||||
|
||||
mkdir -p exp
|
||||
|
||||
# seed may break model convergence
|
||||
seed=1988
|
||||
if [ ${seed} != 0 ]; then
|
||||
export FLAGS_cudnn_deterministic=True
|
||||
fi
|
||||
|
||||
# export FLAGS_cudnn_exhaustive_search=true
|
||||
# export FLAGS_conv_workspace_size_limit=4000
|
||||
export FLAGS_allocator_strategy=naive_best_fit
|
||||
if [ ${ngpu} == 0 ]; then
|
||||
python3 -u ${BIN_DIR}/train.py \
|
||||
--ngpu ${ngpu} \
|
||||
--config ${config_path} \
|
||||
--output exp/${ckpt_name} \
|
||||
--seed ${seed} \
|
||||
--resume ${resume}
|
||||
else
|
||||
python3 -m paddle.distributed.launch --gpus=${CUDA_VISIBLE_DEVICES} ${ips_config} ${BIN_DIR}/train.py \
|
||||
--ngpu ${ngpu} \
|
||||
--config ${config_path} \
|
||||
--output exp/${ckpt_name} \
|
||||
--seed ${seed} \
|
||||
--resume ${resume}
|
||||
fi
|
||||
|
||||
if [ ${seed} != 0 ]; then
|
||||
unset FLAGS_cudnn_deterministic
|
||||
fi
|
||||
|
||||
if [ $? -ne 0 ]; then
|
||||
echo "Failed in training!"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
exit 0
|
||||
@ -0,0 +1 @@
|
||||
../../../utils
|
||||
Loading…
Reference in new issue