docs/academic_recipe/asr_recipe.md
@@ -12,7 +12,7 @@ Then you can directly start the recipe as follows: ```sh conda activate funasr . ./run.sh . ./run.sh --CUDA_VISIBLE_DEVICES="0,1" --gpu_num=2 ``` The training log files are saved in `${exp_dir}/exp/${model_dir}/log/train.log.*`， which can be viewed using the following command: @@ -26,15 +26,18 @@ ... 1epoch:train:801-850batch:850num_updates: ... loss_ctc=107.890, loss_att=87.832, acc=0.029, loss_pre=1.702 ... ``` Also, users can use tensorboard to observe these training information by the following command: ```sh tensorboard --logdir ${exp_dir}/exp/${model_dir}/tensorboard/train ``` At the end of each epoch, the evaluation metrics are calculated on the validation set, like follows: ```text ... [valid] loss_ctc=99.914, cer_ctc=1.000, loss_att=80.512, acc=0.029, cer=0.971, wer=1.000, loss_pre=1.952, loss=88.285 ... ``` Also, users can use tensorboard to observe these training information by the following command: ```sh tensorboard --logdir ${exp_dir}/exp/${model_dir}/tensorboard/train ``` Here is an example of loss: <img src="images/loss.png" width="200"/> The inference results are saved in `${exp_dir}/exp/${model_dir}/decode_asr_*/$dset`. The main two files are `text.cer` and `text.cer.txt`. `text.cer` saves the comparison between the recognized text and the reference text, like follows: ```text @@ -177,7 +180,7 @@ encoder: conformer encoder_conf: output_size: 256 # dimension of attention attention_heads: 4 # number of heads in multi-head attention attention_heads: 4 # the number of heads in multi-head attention linear_units: 2048 # the number of units of position-wise feed forward num_blocks: 12 # the number of encoder blocks dropout_rate: 0.1 docs/academic_recipe/images/loss.png
@@ -85,14 +85,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -130,7 +130,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell/data2vec_paraformer_finetune/run.sh
@@ -88,14 +88,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -134,7 +134,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell/data2vec_transformer_finetune/run.sh
@@ -88,14 +88,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -134,7 +134,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell/paraformer/run.sh
@@ -85,14 +85,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -130,7 +130,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --dataset_type small \ --data_dir ${feats_dir}/data \ egs/aishell/paraformerbert/run.sh
@@ -89,14 +89,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -141,7 +141,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell/transformer/run.sh
@@ -85,14 +85,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -130,7 +130,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell/transformer/utils/compute_cmvn.py
@@ -27,7 +27,7 @@ help="the path of wav scps", ) parser.add_argument( "--config", "--config_file", type=str, help="the config file for computing cmvn", ) @@ -89,7 +89,7 @@ # var_stats += np.sum(np.square(mat), axis=0) # total_frames += mat.shape[0] with open(args.config) as f: with open(args.config_file) as f: configs = yaml.safe_load(f) frontend_configs = configs.get("frontend_conf", {}) num_mel_bins = frontend_configs.get("n_mels", 80) egs/aishell/transformer/utils/compute_cmvn.sh
@@ -2,11 +2,11 @@ . ./path.sh || exit 1; # Begin configuration section. fbankdir=$1 fbankdir= nj=32 cmd=./utils/run.pl feats_dim=80 config= config_file= scale=1.0 echo "$0 $@" @@ -29,7 +29,7 @@ python utils/compute_cmvn.py \ --dim ${feats_dim} \ --wav_path $split_dir \ --config $config \ --config_file $config_file \ --idx JOB \ python utils/combine_cmvn_file.py --dim ${feats_dim} --cmvn_dir $split_dir --nj $nj --output_dir ${fbankdir}/cmvn egs/aishell2/conformer/run.sh
@@ -87,14 +87,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -132,7 +132,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell2/conformer/utils
@@ -1 +1 @@ ../transformer/utils/ ../../aishell/transformer/utils egs/aishell2/data2vec_pretrain/run.sh
@@ -66,7 +66,7 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt egs/aishell2/paraformer/run.sh
@@ -87,14 +87,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -132,7 +132,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell2/paraformer/utils
@@ -1 +1 @@ ../transformer/utils/ ../../aishell/transformer/utils egs/aishell2/paraformerbert/run.sh
@@ -90,14 +90,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -142,7 +142,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell2/paraformerbert/utils
@@ -1 +1 @@ ../transformer/utils/ ../../aishell/transformer/utils egs/aishell2/transformer/run.sh
@@ -87,14 +87,14 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/${lang}_token_list/char/tokens.txt 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/char/ mkdir -p ${feats_dir}/data/${lang}_token_list/$token_type/ echo "make a dictionary" echo "<blank>" > ${token_list} @@ -132,7 +132,7 @@ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type char \ --token_type $token_type \ --token_list $token_list \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ egs/aishell2/transformer/utils
New file @@ -0,0 +1 @@ ../../aishell/transformer/utils egs/aishell2/transformer/utils/apply_cmvn.py
File was deleted egs/aishell2/transformer/utils/apply_cmvn.sh
File was deleted egs/aishell2/transformer/utils/apply_lfr_and_cmvn.py
File was deleted egs/aishell2/transformer/utils/apply_lfr_and_cmvn.sh
File was deleted egs/aishell2/transformer/utils/cmvn_converter.py
File was deleted egs/aishell2/transformer/utils/combine_cmvn_file.py
File was deleted egs/aishell2/transformer/utils/compute_cmvn.py
File was deleted egs/aishell2/transformer/utils/compute_cmvn.sh
File was deleted egs/aishell2/transformer/utils/compute_fbank.py
File was deleted egs/aishell2/transformer/utils/compute_fbank.sh
File was deleted egs/aishell2/transformer/utils/compute_wer.py
File was deleted egs/aishell2/transformer/utils/download_model.py
File was deleted egs/aishell2/transformer/utils/error_rate_zh
File was deleted egs/aishell2/transformer/utils/extract_embeds.py
File was deleted egs/aishell2/transformer/utils/filter_scp.pl
File was deleted egs/aishell2/transformer/utils/fix_data.sh
File was deleted egs/aishell2/transformer/utils/fix_data_feat.sh
File was deleted egs/aishell2/transformer/utils/gen_ark_list.sh
File was deleted egs/aishell2/transformer/utils/gen_modelscope_configuration.py
File was deleted egs/aishell2/transformer/utils/parse_options.sh
File was deleted egs/aishell2/transformer/utils/print_args.py
File was deleted egs/aishell2/transformer/utils/proc_conf_oss.py
File was deleted egs/aishell2/transformer/utils/proce_text.py
File was deleted egs/aishell2/transformer/utils/run.pl
File was deleted egs/aishell2/transformer/utils/shuffle_list.pl
File was deleted egs/aishell2/transformer/utils/split_data.py
File was deleted egs/aishell2/transformer/utils/split_scp.pl
File was deleted egs/aishell2/transformer/utils/subset_data_dir_tr_cv.sh
File was deleted egs/aishell2/transformer/utils/text2token.py
File was deleted egs/aishell2/transformer/utils/text_tokenize.py
File was deleted egs/aishell2/transformer/utils/text_tokenize.sh
File was deleted egs/aishell2/transformer/utils/textnorm_zh.py
File was deleted egs/aishell2/transformerLM/utils
New file @@ -0,0 +1 @@ ../../aishell/transformer/utils egs/aishell2/transformerLM/utils/parse_options.sh
File was deleted egs/aishell2/transformerLM/utils/run.pl
File was deleted egs/aishell2/transformerLM/utils/split_scp.pl
File was deleted egs/librispeech/conformer/run.sh
@@ -97,7 +97,7 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/lang_char/${train_set}_${bpemode}${nbpe}_units.txt egs/librispeech_100h/conformer/run.sh
@@ -93,7 +93,7 @@ if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config $asr_config --scale 1.0 utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 1.0 fi token_list=${feats_dir}/data/lang_char/${train_set}_${bpemode}${nbpe}_units.txt @@ -120,7 +120,7 @@ # ASR Training Stage world_size=$gpu_num # run on one machine if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4; then if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then echo "stage 4: ASR Training" mkdir -p ${exp_dir}/exp/${model_dir} mkdir -p ${exp_dir}/exp/${model_dir}/log egs/wenetspeech/conformer/conf/decode_asr_transformer_5beam.yaml
New file @@ -0,0 +1,6 @@ beam_size: 5 penalty: 0.0 maxlenratio: 0.0 minlenratio: 0.0 ctc_weight: 0.5 lm_weight: 0.7 egs/wenetspeech/conformer/conf/train_asr_conformer.yaml
New file @@ -0,0 +1,104 @@ # network architecture # encoder related encoder: conformer encoder_conf: output_size: 512 # dimension of attention attention_heads: 8 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 rel_pos_type: latest 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 related decoder: transformer decoder_conf: attention_heads: 8 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 # CTC realted ctc_conf: ignore_nan_grad: true # frontend related frontend: wav_frontend frontend_conf: fs: 16000 window: hamming n_mels: 80 frame_length: 25 frame_shift: 10 lfr_m: 1 lfr_n: 1 # hybrid CTC/attention model_conf: ctc_weight: 0.3 lsm_weight: 0.1 # label smoothing option length_normalized_loss: false # optimization related accum_grad: 4 grad_clip: 5 patience: none max_epoch: 30 val_scheduler_criterion: - valid - acc best_model_criterion: - - valid - acc - max keep_nbest_models: 10 optim: adam optim_conf: lr: 0.0015 scheduler: warmuplr scheduler_conf: warmup_steps: 30000 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 dataset_conf: data_names: speech,text data_types: sound,text_nospace shuffle: True shuffle_conf: shuffle_size: 2048 sort_size: 500 batch_conf: batch_type: token batch_size: 32000 num_workers: 8 log_interval: 50 normalize: None egs/wenetspeech/conformer/local/data.sh
New file @@ -0,0 +1,102 @@ #!/usr/bin/env bash # 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 log() { local fname=${BASH_SOURCE[1]##*/} echo -e "$(date '+%Y-%m-%dT%H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*" } SECONDS=0 # general configuration nj=10 stage=2 stop_stage=100 set=L data_dir="data" WENETSPEECH= train_cmd= log "$0 $*" . utils/parse_options.sh if [ ! -d "${WENETSPEECH}/audio" ] && [ ! -f "${WENETSPEECH}/WenetSpeech.json" ]; then echo "Valid WENETSPEECH data not found in ${WENETSPEECH}." echo "Please follow the instruction in https://wenet-e2e.github.io/WenetSpeech/" echo "and re-construct the data." exit 1 fi train_set=train_"$(echo "${set}" | tr "[:upper:]" "[:lower:]")" dev_set=dev test_sets="test_net test_meeting" if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then log "data preparation" mkdir -p ${data_dir} abs_data_dir=$(readlink -f ${data_dir}) log "making Kaldi format data directory in ${abs_data_dir}" local/wenetspeech_data_prep.sh \ --train-subset ${set} \ --stage 1 \ ${WENETSPEECH} \ ${abs_data_dir} # prepare utt2spk and spk2utt files for x in ${train_set} ${dev_set} ${test_sets}; do dir=${data_dir}/${x} paste -d " " <(cut -f 1 ${dir}/segments) <(cut -f 1 ${dir}/segments) | \ sort -u > ${dir}/utt2spk utils/utt2spk_to_spk2utt.pl ${dir}/utt2spk > ${dir}/spk2utt done fi if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then log "process the long term opus audio file, may take about 3 hours" for x in ${train_set} ${dev_set} ${test_sets}; do log "process audio for ${data_dir}/${x}" dir=${data_dir}/${x} mkdir -p ${dir}/logs nutt=$(<${dir}/segments wc -l) nj=$((nj<nutt?nj:nutt)) split_scps="" for n in $(seq ${nj}); do split_scps="${split_scps} ${dir}/logs/segments.${n}" done utils/split_scp.pl ${dir}/segments ${split_scps} ${train_cmd} "JOB=1:${nj}" "${dir}/logs/process_audio.JOB.log"\ python3 local/process_opus.py \ ${dir}/wav.scp \ ${dir}/logs/segments.JOB \ ${dir}/logs/wav.JOB.scp # modify the `wav.scp` file and rename the `segments` file # rename the `segments` file to avoid the audio file formatting process in stage 3 of `asr.sh` mv ${dir}/wav.scp ${dir}/wav.scp.org mv ${dir}/segments ${dir}/segments.org for n in $(seq ${nj}); do cat ${dir}/logs/wav.${n}.scp || exit 1; done | sort -u > ${dir}/wav.scp done fi if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then log "format text file" for x in ${train_set} ${dev_set} ${test_sets}; do log "format text for ${data_dir}/${x}" dir=${data_dir}/${x} mv ${dir}/text ${dir}/text.org paste -d " " <(cut -f 1 ${dir}/text.org) \ <(cut -f 2- ${dir}/text.org | local/text_normalize.pl) | \ sort -u > ${dir}/text utils/fix_data_dir.sh ${dir} done fi log "Successfully finished. [elapsed=${SECONDS}s]" egs/wenetspeech/conformer/local/extract_meta.py
New file @@ -0,0 +1,114 @@ # Copyright 2021 Xiaomi Corporation (Author: Yongqing Wang) # Mobvoi Inc(Author: Di Wu, Binbin Zhang) # 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. import argparse import json import os import sys def get_args(): parser = argparse.ArgumentParser( description=""" This script is used to process raw json dataset of WenetSpeech, where the long wav is splitinto segments and data of wenet format is generated. """ ) parser.add_argument("input_json", help="""Input json file of WenetSpeech""") parser.add_argument("output_dir", help="""Output dir for prepared data""") args = parser.parse_args() return args def meta_analysis(input_json, output_dir): input_dir = os.path.dirname(input_json) if not os.path.exists(output_dir): os.makedirs(output_dir) try: with open(input_json, "r") as injson: json_data = json.load(injson) except Exception: sys.exit(f"Failed to load input json file: {input_json}") else: if json_data["audios"] is not None: with open(f"{output_dir}/text", "w") as utt2text, open( f"{output_dir}/segments", "w" ) as segments, open(f"{output_dir}/utt2dur", "w") as utt2dur, open( f"{output_dir}/wav.scp", "w" ) as wavscp, open( f"{output_dir}/utt2subsets", "w" ) as utt2subsets, open( f"{output_dir}/reco2dur", "w" ) as reco2dur: for long_audio in json_data["audios"]: try: long_audio_path = os.path.realpath( os.path.join(input_dir, long_audio["path"]) ) aid = long_audio["aid"] segments_lists = long_audio["segments"] duration = long_audio["duration"] assert os.path.exists(long_audio_path) except AssertionError: print( f"""Warning: {aid} something is wrong, maybe AssertionError, skipped""" ) continue except Exception: print( f"""Warning: {aid} something is wrong, maybe the error path: {long_audio_path}, skipped""" ) continue else: wavscp.write(f"{aid}\t{long_audio_path}\n") reco2dur.write(f"{aid}\t{duration}\n") for segment_file in segments_lists: try: sid = segment_file["sid"] start_time = segment_file["begin_time"] end_time = segment_file["end_time"] dur = end_time - start_time text = segment_file["text"] segment_subsets = segment_file["subsets"] except Exception: print( f"""Warning: {segment_file} something is wrong, skipped""" ) continue else: utt2text.write(f"{sid}\t{text}\n") segments.write( f"{sid}\t{aid}\t{start_time}\t{end_time}\n" ) utt2dur.write(f"{sid}\t{dur}\n") segment_sub_names = " ".join(segment_subsets) utt2subsets.write(f"{sid}\t{segment_sub_names}\n") def main(): args = get_args() meta_analysis(args.input_json, args.output_dir) if __name__ == "__main__": main() egs/wenetspeech/conformer/local/path.sh
egs/wenetspeech/conformer/local/process_opus.py
New file @@ -0,0 +1,88 @@ # Copyright 2021 NPU, ASLP Group (Author: Qijie Shao) # 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. # process_opus.py: segmentation and downsampling of opus audio # usage: python3 process_opus.py wav.scp segments output_wav.scp import os import sys import torchaudio def read_file(wav_scp, segments): wav_scp_dict = {} with open(wav_scp, "r", encoding="UTF-8") as fin: for line_str in fin: wav_id, path = line_str.strip().split() wav_scp_dict[wav_id] = path utt_list = [] seg_path_list = [] start_time_list = [] end_time_list = [] with open(segments, "r", encoding="UTF-8") as fin: for line_str in fin: arr = line_str.strip().split() assert len(arr) == 4 utt_list.append(arr[0]) seg_path_list.append(wav_scp_dict[arr[1]]) start_time_list.append(float(arr[2])) end_time_list.append(float(arr[3])) return utt_list, seg_path_list, start_time_list, end_time_list # TODO(Qijie): Fix the process logic def output(output_wav_scp, utt_list, seg_path_list, start_time_list, end_time_list): num_utts = len(utt_list) step = int(num_utts * 0.01) with open(output_wav_scp, "w", encoding="UTF-8") as fout: previous_wav_path = "" for i in range(num_utts): utt_id = utt_list[i] current_wav_path = seg_path_list[i] output_dir = (os.path.dirname(current_wav_path)).replace( "audio", "audio_seg" ) seg_wav_path = os.path.join(output_dir, utt_id + ".wav") os.makedirs(output_dir, exist_ok=True) if current_wav_path != previous_wav_path: waveform, sample_rate = torchaudio.load(current_wav_path) previous_wav_path = current_wav_path start = int(start_time_list[i] * sample_rate) end = int(end_time_list[i] * sample_rate) target_audio = waveform[:, start:end] torchaudio.save(seg_wav_path, target_audio, sample_rate) fout.write("{} {}\n".format(utt_id, seg_wav_path)) if i % step == 0: print("seg wav finished: {}%".format(int(i / step))) def main(): wav_scp = sys.argv[1] segments = sys.argv[2] output_wav_scp = sys.argv[3] utt_list, seg_path_list, start_time_list, end_time_list = read_file( wav_scp, segments ) output(output_wav_scp, utt_list, seg_path_list, start_time_list, end_time_list) if __name__ == "__main__": main() egs/wenetspeech/conformer/local/text_normalize.pl
New file @@ -0,0 +1,24 @@ #!/usr/bin/env perl use utf8; use open qw(:std :utf8); use warnings; while (<STDIN>) { chomp; # remove non UTF-8 whitespace character if ($_ =~ / /) {$_ =~ s: ::g;} if ($_ =~ / /) {$_ =~ s: ::g;} # upper letters if ($_ =~ /[a-zA-Z]/) {$_ =~ uc $_;} # add "_" before and after each English word if ($_ =~ /([A-Z]+)\s+([A-Z]+)/) {$_ =~ s/([A-Z]+)\s+([A-Z]+)/$1\_$2/g;} if ($_ =~ /([A-Z]+)\s+([A-Z]+)/) {$_ =~ s/([A-Z]+)\s+([A-Z]+)/$1\_$2/g;} if ($_ =~ m/([A-Z]+)(\p{Han}+)/) {$_ =~ s/([A-Z]+)(\p{Han}+)/$1\_$2/g;} if ($_ =~ m/(\p{Han}+)([A-Z]+)/) {$_ =~ s/(\p{Han}+)([A-Z]+)/$1\_$2/g;} # remove UTF-8 whitespace charcter if ($_ =~ /\s+/) {$_ =~ s:\s+::g;} # replace "_" with a normal whitespace if ($_ =~ /\_/) {$_ =~ s:\_: :g;} print "$_\n"; } egs/wenetspeech/conformer/local/wenetspeech_data_prep.sh
New file @@ -0,0 +1,135 @@ #!/usr/bin/env bash # Copyright 2021 Xiaomi Corporation (Author: Yongqing Wang) # Seasalt AI, Inc (Author: Guoguo Chen) # Mobvoi Inc(Author: Di Wu, Binbin Zhang) # NPU, ASLP Group (Author: Qijie Shao) # 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. set -e set -o pipefail stage=1 prefix= train_subset=L . utils/parse_options.sh || exit 1; filter_by_id () { idlist=$1 input=$2 output=$3 field=1 if [ $# -eq 4 ]; then field=$4 fi cat $input | perl -se ' open(F, "<$idlist") || die "Could not open id-list file $idlist"; while(<F>) { @A = split; @A>=1 || die "Invalid id-list file line $_"; $seen{$A[0]} = 1; } while(<>) { @A = split; @A > 0 || die "Invalid file line $_"; @A >= $field || die "Invalid file line $_"; if ($seen{$A[$field-1]}) { print $_; } }' -- -idlist="$idlist" -field="$field" > $output ||\ (echo "$0: filter_by_id() error: $input" && exit 1) || exit 1; } subset_data_dir () { utt_list=$1 src_dir=$2 dest_dir=$3 mkdir -p $dest_dir || exit 1; # wav.scp text segments utt2dur filter_by_id $utt_list $src_dir/utt2dur $dest_dir/utt2dur ||\ (echo "$0: subset_data_dir() error: $src_dir/utt2dur" && exit 1) || exit 1; filter_by_id $utt_list $src_dir/text $dest_dir/text ||\ (echo "$0: subset_data_dir() error: $src_dir/text" && exit 1) || exit 1; filter_by_id $utt_list $src_dir/segments $dest_dir/segments ||\ (echo "$0: subset_data_dir() error: $src_dir/segments" && exit 1) || exit 1; awk '{print $2}' $dest_dir/segments | sort | uniq > $dest_dir/reco filter_by_id $dest_dir/reco $src_dir/wav.scp $dest_dir/wav.scp ||\ (echo "$0: subset_data_dir() error: $src_dir/wav.scp" && exit 1) || exit 1; rm -f $dest_dir/reco } if [ $# -ne 2 ]; then echo "Usage: $0 [options] <wenetspeech-dataset-dir> <data-dir>" echo " e.g.: $0 --train-subset L /disk1/audio_data/wenetspeech/ data/" echo "" echo "This script takes the WenetSpeech source directory, and prepares the" echo "WeNet format data directory." echo " --prefix <prefix> # Prefix for output data directory." echo " --stage <stage> # Processing stage." echo " --train-subset <L|M|S|W> # Train subset to be created." exit 1 fi wenetspeech_dir=$1 data_dir=$2 declare -A subsets subsets=( [L]="train_l" [M]="train_m" [S]="train_s" [W]="train_w" [DEV]="dev" [TEST_NET]="test_net" [TEST_MEETING]="test_meeting") prefix=${prefix:+${prefix}_} corpus_dir=$data_dir/${prefix}corpus/ if [ $stage -le 1 ]; then echo "$0: Extract meta into $corpus_dir" # Sanity check. [ ! -f $wenetspeech_dir/WenetSpeech.json ] &&\ echo "$0: Please download $wenetspeech_dir/WenetSpeech.json!" && exit 1; [ ! -d $wenetspeech_dir/audio ] &&\ echo "$0: Please download $wenetspeech_dir/audio!" && exit 1; [ ! -d $corpus_dir ] && mkdir -p $corpus_dir # Files to be created: # wav.scp text segments utt2dur python3 local/extract_meta.py \ $wenetspeech_dir/WenetSpeech.json $corpus_dir || exit 1; fi if [ $stage -le 2 ]; then echo "$0: Split data to train, dev, test_net, and test_meeting" [ ! -f $corpus_dir/utt2subsets ] &&\ echo "$0: No such file $corpus_dir/utt2subsets!" && exit 1; for label in $train_subset DEV TEST_NET TEST_MEETING; do if [ ! ${subsets[$label]+set} ]; then echo "$0: Subset $label is not defined in WenetSpeech.json." && exit 1; fi subset=${subsets[$label]} [ ! -d $data_dir/${prefix}$subset ] && mkdir -p $data_dir/${prefix}$subset cat $corpus_dir/utt2subsets | \ awk -v s=$label '{for (i=2;i<=NF;i++) if($i==s) print $0;}' \ > $corpus_dir/${prefix}${subset}_utt_list|| exit 1; subset_data_dir $corpus_dir/${prefix}${subset}_utt_list \ $corpus_dir $data_dir/${prefix}$subset || exit 1; done fi echo "$0: Done" egs/wenetspeech/conformer/path.sh
New file @@ -0,0 +1,5 @@ export FUNASR_DIR=$PWD/../../.. # NOTE(kan-bayashi): Use UTF-8 in Python to avoid UnicodeDecodeError when LC_ALL=C export PYTHONIOENCODING=UTF-8 export PATH=$FUNASR_DIR/funasr/bin:$PATH egs/wenetspeech/conformer/run.sh
New file @@ -0,0 +1,223 @@ #!/usr/bin/env bash . ./path.sh || exit 1; # machines configuration CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" gpu_num=8 count=1 gpu_inference=true # Whether to perform gpu decoding, set false for cpu decoding # for gpu decoding, inference_nj=ngpu*njob; for cpu decoding, inference_nj=njob njob=5 train_cmd=utils/run.pl infer_cmd=utils/run.pl # general configuration feats_dir="../DATA" #feature output dictionary exp_dir="." lang=zh token_type=char type=sound scp=wav.scp speed_perturb="0.9 1.0 1.1" stage=3 stop_stage=5 # feature configuration feats_dim=80 nj=64 # data raw_data=/nfs/zhifu.gzf/wenetspeech_proc # exp tag tag="exp1" . 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 set=L train_set=train_l valid_set=dev test_sets="dev test_net test_meeting" asr_config=conf/train_asr_conformer.yaml model_dir="baseline_$(basename "${asr_config}" .yaml)_${lang}_${token_type}_${tag}" inference_config=conf/decode_asr_transformer_5beam.yaml inference_asr_model=valid.acc.ave_10best.pb # you can set gpu num for decoding here gpuid_list=$CUDA_VISIBLE_DEVICES # set gpus for decoding, the same as training stage by default ngpu=$(echo $gpuid_list | awk -F "," '{print NF}') if ${gpu_inference}; then inference_nj=$[${ngpu}*${njob}] _ngpu=1 else inference_nj=$njob _ngpu=0 fi if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then echo "For downloading data, please refer to https://github.com/wenet-e2e/WenetSpeech." exit 0; fi if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then echo "stage 0: Data preparation" # Data preparation local/data.sh --set ${set} --nj $nj --data_dir $feats_dir --WENETSPEECH $raw_data --train_cmd $train_cmd mkdir $feats_dir/data mv $feats_dir/$train_set $feats_dir/data/$train_set for x in $test_sets; do mv $feats_dir/$x $feats_dir/data/ done fi if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then echo "stage 1: Feature and CMVN Generation" utils/compute_cmvn.sh --fbankdir ${feats_dir}/data/${train_set} --cmd "$train_cmd" --nj $nj --feats_dim ${feats_dim} --config_file "$asr_config" --scale 0.1 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>" >> ${token_list} fi # LM Training Stage world_size=$gpu_num # run on one machine if [ ${stage} -le 3 ] && [ ${stop_stage} -ge 3 ]; then echo "stage 3: LM Training" fi # ASR Training Stage world_size=$gpu_num # run on one machine if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then echo "stage 4: ASR Training" mkdir -p ${exp_dir}/exp/${model_dir} mkdir -p ${exp_dir}/exp/${model_dir}/log INIT_FILE=${exp_dir}/exp/${model_dir}/ddp_init if [ -f $INIT_FILE ];then rm -f $INIT_FILE fi init_method=file://$(readlink -f $INIT_FILE) echo "$0: init method is $init_method" for ((i = 0; i < $gpu_num; ++i)); do { rank=$i local_rank=$i gpu_id=$(echo $CUDA_VISIBLE_DEVICES | cut -d',' -f$[$i+1]) train.py \ --task_name asr \ --gpu_id $gpu_id \ --use_preprocessor true \ --token_type $token_type \ --token_list $token_list \ --dataset_type large \ --data_dir ${feats_dir}/data \ --train_set ${train_set} \ --valid_set ${valid_set} \ --data_file_names "wav.scp,text" \ --cmvn_file ${feats_dir}/data/${train_set}/cmvn/am.mvn \ --speed_perturb ${speed_perturb} \ --resume true \ --output_dir ${exp_dir}/exp/${model_dir} \ --config $asr_config \ --ngpu $gpu_num \ --num_worker_count $count \ --dist_init_method $init_method \ --dist_world_size $world_size \ --dist_rank $rank \ --local_rank $local_rank 1> ${exp_dir}/exp/${model_dir}/log/train.log.$i 2>&1 } & done wait fi # Testing Stage if [ ${stage} -le 5 ] && [ ${stop_stage} -ge 5 ]; then echo "stage 5: Inference" for dset in ${test_sets}; do asr_exp=${exp_dir}/exp/${model_dir} inference_tag="$(basename "${inference_config}" .yaml)" _dir="${asr_exp}/${inference_tag}/${inference_asr_model}/${dset}" _logdir="${_dir}/logdir" if [ -d ${_dir} ]; then echo "${_dir} is already exists. if you want to decode again, please delete this dir first." exit 0 fi mkdir -p "${_logdir}" _data="${feats_dir}/data/${dset}" key_file=${_data}/${scp} num_scp_file="$(<${key_file} wc -l)" _nj=$([ $inference_nj -le $num_scp_file ] && echo "$inference_nj" || echo "$num_scp_file") split_scps= for n in $(seq "${_nj}"); do split_scps+=" ${_logdir}/keys.${n}.scp" done # shellcheck disable=SC2086 utils/split_scp.pl "${key_file}" ${split_scps} _opts= if [ -n "${inference_config}" ]; then _opts+="--config ${inference_config} " fi ${infer_cmd} --gpu "${_ngpu}" --max-jobs-run "${_nj}" JOB=1:"${_nj}" "${_logdir}"/asr_inference.JOB.log \ python -m funasr.bin.asr_inference_launch \ --batch_size 1 \ --ngpu "${_ngpu}" \ --njob ${njob} \ --gpuid_list ${gpuid_list} \ --data_path_and_name_and_type "${_data}/${scp},speech,${type}" \ --cmvn_file ${feats_dir}/data/${train_set}/cmvn/am.mvn \ --key_file "${_logdir}"/keys.JOB.scp \ --asr_train_config "${asr_exp}"/config.yaml \ --asr_model_file "${asr_exp}"/"${inference_asr_model}" \ --output_dir "${_logdir}"/output.JOB \ --mode asr \ ${_opts} for f in token token_int score text; do if [ -f "${_logdir}/output.1/1best_recog/${f}" ]; then for i in $(seq "${_nj}"); do cat "${_logdir}/output.${i}/1best_recog/${f}" done | sort -k1 >"${_dir}/${f}" fi done python utils/proce_text.py ${_dir}/text ${_dir}/text.proc python utils/proce_text.py ${_data}/text ${_data}/text.proc python utils/compute_wer.py ${_data}/text.proc ${_dir}/text.proc ${_dir}/text.cer tail -n 3 ${_dir}/text.cer > ${_dir}/text.cer.txt cat ${_dir}/text.cer.txt done fi # Prepare files for ModelScope fine-tuning and inference if [ ${stage} -le 6 ] && [ ${stop_stage} -ge 6 ]; then echo "stage 6: ModelScope Preparation" cp ${feats_dir}/data/${train_set}/cmvn/am.mvn ${exp_dir}/exp/${model_dir}/am.mvn vocab_size=$(cat ${token_list} | wc -l) python utils/gen_modelscope_configuration.py \ --am_model_name $inference_asr_model \ --mode asr \ --model_name conformer \ --dataset wenetspeech \ --output_dir $exp_dir/exp/$model_dir \ --vocab_size $vocab_size \ --tag $tag fi egs/wenetspeech/conformer/utils
New file @@ -0,0 +1 @@ ../../aishell/transformer/utils funasr/datasets/large_datasets/dataset.py
@@ -148,6 +148,12 @@ if "key" not in sample_dict: sample_dict["key"] = segs[0] sample_dict['hw_tag'] = 1 elif data_type == "text_nospace": text = item segs = text.strip().split(maxsplit=1) sample_dict[data_name] = [x for x in segs[1]] if "key" not in sample_dict: sample_dict["key"] = segs[0] else: text = item segs = text.strip().split() funasr/utils/prepare_data.py
@@ -185,7 +185,7 @@ for i in range(nj): path = "" for file_name in file_names: path = path + os.path.join(split_path, str(i + 1), file_name) path = path + " " + os.path.join(split_path, str(i + 1), file_name) f_data.write(path + "\n")
