ASR2 recipe on Tedlium3 dataset

egs2/TEMPLATE/asr2/asr2.sh

@@ -821,7 +821,18 @@ if [ ${stage} -le 6 ] && [ ${stop_stage} -ge 6 ] && ! [[ " ${skip_stages} " =~ [
         for utt_extra_file in ${utt_extra_files}; do
             cp "${data_feats}/org/${dset}/${utt_extra_file}" "${data_feats}/${dset}"
         done
-        # TODO: Maybe Remove empty text
+
+        # Remove empty text
+        cat "${data_feats}/org/${dset}/text.${tgt_case}.${tgt_lang}" | awk ' { if( NF != 1 ) print $0; } ' > "${data_feats}/${dset}/text.${tgt_case}.${tgt_lang}"
+        utils/filter_scp.pl "${data_feats}/${dset}/text.${tgt_case}.${tgt_lang}" "${data_feats}/org/${dset}/utt2spk" > "${data_feats}/${dset}/utt2spk"
+        utils/fix_data_dir.sh \
+            --utt_extra_files "${utt_extra_files}" "${data_feats}/${dset}"
+
+        # Check how many samples are removed
+        org_num_samples=$(wc -l "${data_feats}/org/${dset}/utt2spk" | cut -d' ' -f1)
+        filtered_num_samples=$(wc -l "${data_feats}/${dset}/utt2spk" | cut -d' ' -f1)
+        echo "filter samples with empty texts: removed $((org_num_samples - filtered_num_samples)) samples with empty text"
+
         # TODO: Add other data cleaning -- currently being done as part of data.sh
     done
 
@@ -1291,16 +1302,16 @@ if [ ${stage} -le 13 ] && [ ${stop_stage} -ge 13 ] && ! [[ " ${skip_stages} " =~
             log "${_split_dir}/.done exists. Spliting is skipped"
         fi
 
-        _opts+="--train_data_path_and_name_and_type ${_split_dir}/text.${tgt_case}.${tgt_lang},text,text "
         _opts+="--train_data_path_and_name_and_type ${_split_dir}/text.${src_case}.${src_lang},src_text,text "
-        _opts+="--train_shape_file ${_split_dir}/text_shape.${tgt_token_type} "
+        _opts+="--train_data_path_and_name_and_type ${_split_dir}/text.${tgt_case}.${tgt_lang},text,text "
         _opts+="--train_shape_file ${_split_dir}/src_text_shape.${src_token_type} "
+        _opts+="--train_shape_file ${_split_dir}/text_shape.${tgt_token_type} "
         _opts+="--multiple_iterator true "
     else
-        _opts+="--train_data_path_and_name_and_type ${_asr_train_dir}/text.${tgt_case}.${tgt_lang},text,text "
         _opts+="--train_data_path_and_name_and_type ${_asr_train_dir}/text.${src_case}.${src_lang},src_text,text "
-        _opts+="--train_shape_file ${asr_stats_dir}/train/text_shape.${tgt_token_type} "
+        _opts+="--train_data_path_and_name_and_type ${_asr_train_dir}/text.${tgt_case}.${tgt_lang},text,text "
         _opts+="--train_shape_file ${asr_stats_dir}/train/src_text_shape.${src_token_type} "
+        _opts+="--train_shape_file ${asr_stats_dir}/train/text_shape.${tgt_token_type} "
     fi
 
     log "Generate '${asr_exp}/run.sh'. You can resume the process from stage 13 using this script"

egs2/tedlium3/asr2/README.md

@@ -0,0 +1,70 @@
+# ASR2 recipe for Tedlium3
+## Related work:
+   Chang, Xuankai, et al. "Exploration of Efficient End-to-End ASR using Discretized Input from Self-Supervised Learning." InterSpeech 2023.
+   <details>
+   <summary>bib info</summary>
+
+   ```
+   @article{chang2023exploration,
+        title={Exploration of Efficient End-to-End ASR using Discretized Input from Self-Supervised Learning},
+        author={Chang, Xuankai and Yan, Brian and Fujita, Yuya and Maekaku, Takashi and Watanabe, Shinji},
+        journal={arXiv preprint arXiv:2305.18108},
+        year={2023}
+   }
+   ```
+   </details>
+
+
+# E-Branchformer ASR2 Discrete tokens with WavLM_large_Layer21_Kmeans1000_nBPE2000 (~14.5 hours for 35epochs with A5000 x 1)
+
+## Environments
+- date: `Thu Oct 19 22:11:12 JST 2023`
+- python version: `3.10.8 (main, Nov 24 2022, 14:13:03) [GCC 11.2.0]`
+- espnet version: `espnet 202308`
+- pytorch version: `pytorch 1.13.1`
+- Git hash: `7bcdab47ff7f47e55d52061e55db4128913f32b6`
+  - Commit date: `Thu Aug 31 20:42:18 2023 +0900`
+
+## Model info
+- Model link: https://huggingface.co/espnet/kohei0209_ted3_asr2_e_branchformer1_raw_wavlm_large_21_km1000_bpe_rm2000_bpe_ts500_sp
+- ASR config: [conf/tuning/train_discrete_asr_e_branchformer1.yaml](conf/tuning/train_discrete_asr_e_branchformer1.yaml)
+- Decode config: [conf/tuning/decode_ctc0.3.yaml](conf/tuning/decode_ctc0.3.yaml)
+
+
+## exp/asr_train_discrete_asr_e_branchformer1_raw_wavlm_large_21_km1000_bpe_rm2000_bpe_ts500_sp/
+### WER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_asr_model_valid.acc.ave/test|1155|27500|94.6|3.4|2.0|3.5|8.9|79.0|
+
+### CER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_asr_model_valid.acc.ave/test|1155|145066|97.4|0.9|1.7|4.2|6.7|79.0|
+
+### TER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_asr_model_valid.acc.ave/test|1155|54206|96.1|2.2|1.7|3.8|7.7|79.0|
+
+## exp/asr_train_discrete_asr_e_branchformer1_raw_wavlm_large_21_km1000_bpe_rm2000_bpe_ts500_sp/decode_asr_model_valid.acc.ave
+### WER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|org/dev|507|17783|94.2|3.7|2.2|3.2|9.0|84.8|
+
+### CER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|org/dev|507|95429|97.2|0.9|1.9|3.6|6.3|84.8|
+
+### TER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|org/dev|507|36002|95.8|2.3|1.9|3.2|7.4|84.8|

egs2/tedlium3/asr2/asr2.sh

@@ -0,0 +1 @@
+../../TEMPLATE/asr2/asr2.sh

egs2/tedlium3/asr2/cmd.sh

@@ -0,0 +1,110 @@
+# ====== About run.pl, queue.pl, slurm.pl, and ssh.pl ======
+# Usage: <cmd>.pl [options] JOB=1:<nj> <log> <command...>
+# e.g.
+#   run.pl --mem 4G JOB=1:10 echo.JOB.log echo JOB
+#
+# Options:
+#   --time <time>: Limit the maximum time to execute.
+#   --mem <mem>: Limit the maximum memory usage.
+#   -–max-jobs-run <njob>: Limit the number parallel jobs. This is ignored for non-array jobs.
+#   --num-threads <ngpu>: Specify the number of CPU core.
+#   --gpu <ngpu>: Specify the number of GPU devices.
+#   --config: Change the configuration file from default.
+#
+# "JOB=1:10" is used for "array jobs" and it can control the number of parallel jobs.
+# The left string of "=", i.e. "JOB", is replaced by <N>(Nth job) in the command and the log file name,
+# e.g. "echo JOB" is changed to "echo 3" for the 3rd job and "echo 8" for 8th job respectively.
+# Note that the number must start with a positive number, so you can't use "JOB=0:10" for example.
+#
+# run.pl, queue.pl, slurm.pl, and ssh.pl have unified interface, not depending on its backend.
+# These options are mapping to specific options for each backend and
+# it is configured by "conf/queue.conf" and "conf/slurm.conf" by default.
+# If jobs failed, your configuration might be wrong for your environment.
+#
+#
+# The official documentation for run.pl, queue.pl, slurm.pl, and ssh.pl:
+#   "Parallelization in Kaldi": http://kaldi-asr.org/doc/queue.html
+# =========================================================~
+
+
+# Select the backend used by run.sh from "local", "stdout", "sge", "slurm", or "ssh"
+cmd_backend='local'
+
+# Local machine, without any Job scheduling system
+if [ "${cmd_backend}" = local ]; then
+
+    # The other usage
+    export train_cmd="run.pl"
+    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
+    export cuda_cmd="run.pl"
+    # Used for "*_recog.py"
+    export decode_cmd="run.pl"
+
+# Local machine logging to stdout and log file, without any Job scheduling system
+elif [ "${cmd_backend}" = stdout ]; then
+
+    # The other usage
+    export train_cmd="stdout.pl"
+    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
+    export cuda_cmd="stdout.pl"
+    # Used for "*_recog.py"
+    export decode_cmd="stdout.pl"
+
+
+# "qsub" (Sun Grid Engine, or derivation of it)
+elif [ "${cmd_backend}" = sge ]; then
+    # The default setting is written in conf/queue.conf.
+    # You must change "-q g.q" for the "queue" for your environment.
+    # To know the "queue" names, type "qhost -q"
+    # Note that to use "--gpu *", you have to setup "complex_value" for the system scheduler.
+
+    export train_cmd="queue.pl"
+    export cuda_cmd="queue.pl"
+    export decode_cmd="queue.pl"
+
+
+# "qsub" (Torque/PBS.)
+elif [ "${cmd_backend}" = pbs ]; then
+    # The default setting is written in conf/pbs.conf.
+
+    export train_cmd="pbs.pl"
+    export cuda_cmd="pbs.pl"
+    export decode_cmd="pbs.pl"
+
+
+# "sbatch" (Slurm)
+elif [ "${cmd_backend}" = slurm ]; then
+    # The default setting is written in conf/slurm.conf.
+    # You must change "-p cpu" and "-p gpu" for the "partition" for your environment.
+    # To know the "partion" names, type "sinfo".
+    # You can use "--gpu * " by default for slurm and it is interpreted as "--gres gpu:*"
+    # The devices are allocated exclusively using "${CUDA_VISIBLE_DEVICES}".
+
+    export train_cmd="slurm.pl"
+    export cuda_cmd="slurm.pl"
+    export decode_cmd="slurm.pl"
+
+elif [ "${cmd_backend}" = ssh ]; then
+    # You have to create ".queue/machines" to specify the host to execute jobs.
+    # e.g. .queue/machines
+    #   host1
+    #   host2
+    #   host3
+    # Assuming you can login them without any password, i.e. You have to set ssh keys.
+
+    export train_cmd="ssh.pl"
+    export cuda_cmd="ssh.pl"
+    export decode_cmd="ssh.pl"
+
+# This is an example of specifying several unique options in the JHU CLSP cluster setup.
+# Users can modify/add their own command options according to their cluster environments.
+elif [ "${cmd_backend}" = jhu ]; then
+
+    export train_cmd="queue.pl --mem 2G"
+    export cuda_cmd="queue-freegpu.pl --mem 2G --gpu 1 --config conf/queue.conf"
+    export decode_cmd="queue.pl --mem 4G"
+
+else
+    echo "$0: Error: Unknown cmd_backend=${cmd_backend}" 1>&2
+    return 1
+fi

egs2/tedlium3/asr2/conf/decode.yaml

@@ -0,0 +1 @@
+tuning/decode_ctc0.3.yaml

egs2/tedlium3/asr2/conf/fbank.conf

@@ -0,0 +1,2 @@
+--sample-frequency=16000
+--num-mel-bins=80

egs2/tedlium3/asr2/conf/pbs.conf

@@ -0,0 +1,11 @@
+# Default configuration
+command qsub -V -v PATH -S /bin/bash
+option name=* -N $0
+option mem=* -l mem=$0
+option mem=0          # Do not add anything to qsub_opts
+option num_threads=* -l ncpus=$0
+option num_threads=1  # Do not add anything to qsub_opts
+option num_nodes=* -l nodes=$0:ppn=1
+default gpu=0
+option gpu=0
+option gpu=* -l ngpus=$0

egs2/tedlium3/asr2/conf/pitch.conf

@@ -0,0 +1 @@
+--sample-frequency=16000

egs2/tedlium3/asr2/conf/queue.conf

@@ -0,0 +1,12 @@
+# Default configuration
+command qsub -v PATH -cwd -S /bin/bash -j y -l arch=*64*
+option name=* -N $0
+option mem=* -l mem_free=$0,ram_free=$0
+option mem=0          # Do not add anything to qsub_opts
+option num_threads=* -pe smp $0
+option num_threads=1  # Do not add anything to qsub_opts
+option max_jobs_run=* -tc $0
+option num_nodes=* -pe mpi $0  # You must set this PE as allocation_rule=1
+default gpu=0
+option gpu=0
+option gpu=* -l gpu=$0 -q g.q

egs2/tedlium3/asr2/conf/slurm.conf

@@ -0,0 +1,14 @@
+# Default configuration
+command sbatch --export=PATH
+option name=* --job-name $0
+option time=* --time $0
+option mem=* --mem-per-cpu $0
+option mem=0
+option num_threads=* --cpus-per-task $0
+option num_threads=1 --cpus-per-task 1
+option num_nodes=* --nodes $0
+default gpu=0
+option gpu=0 -p cpu
+option gpu=* -p gpu --gres=gpu:$0 -c $0  # Recommend allocating more CPU than, or equal to the number of GPU
+# note: the --max-jobs-run option is supported as a special case
+# by slurm.pl and you don't have to handle it in the config file.

egs2/tedlium3/asr2/conf/train.yaml

@@ -0,0 +1 @@
+./tuning/train_discrete_asr_e_branchformer1.yaml

egs2/tedlium3/asr2/conf/tuning/decode_ctc0.3.yaml

@@ -0,0 +1,6 @@
+beam_size: 10
+ctc_weight: 0.3
+lm_weight: 0.0
+maxlenratio: 1.0
+minlenratio: 0.0
+penalty: 0.0

egs2/tedlium3/asr2/conf/tuning/train_discrete_asr_e_branchformer1.yaml

@@ -0,0 +1,93 @@
+# Trained with A5000 (24 GB) x 1 GPU for Kmeans1000+nbpe2000. It takes 24 minutes per epoch.
+# BPE-Dropout (https://github.com/google/sentencepiece#subword-regularization-and-bpe-dropout)
+src_tokenizer_encode_conf:
+    enable_sampling: true # If set to true, bpe-dropout is used.
+    alpha: 0.4
+    nbest_size: -1
+
+frontend: embed # embedding + positional encoding
+frontend_conf:
+    embed_dim: 512
+    positional_dropout_rate: 0.1
+
+specaug: specaug
+specaug_conf:
+    apply_time_warp: false
+    time_warp_window: 5
+    time_warp_mode: bicubic
+    apply_freq_mask: false
+    freq_mask_width_range:
+        - 0
+        - 10
+    num_freq_mask: 0
+    apply_time_mask: true
+    time_mask_width_ratio_range:
+        - 0.
+        - 0.05
+    num_time_mask: 10
+
+encoder: e_branchformer
+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: conv1d2
+    layer_drop_rate: 0.0
+    linear_units: 1024
+    positionwise_layer_type: linear
+    use_ffn: true
+    macaron_ffn: true
+    merge_conv_kernel: 31
+
+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: 0.1
+    src_attention_dropout_rate: 0.1
+    layer_drop_rate: 0.0
+
+model: discrete_asr
+model_conf:
+    ctc_weight: 0.3
+    lsm_weight: 0.1
+    length_normalized_loss: false
+    share_decoder_input_output_embed: false
+    share_encoder_decoder_input_embed: false
+
+use_amp: true
+num_att_plot: 1
+log_interval: 500
+num_workers: 4
+batch_type: numel
+batch_bins: 80000000
+accum_grad: 2
+max_epoch: 35
+patience: none
+init: none
+best_model_criterion:
+    - - valid
+      - acc
+      - max
+keep_nbest_models: 10
+
+optim: adam
+optim_conf:
+    lr: 0.001
+    weight_decay: 0.000001
+scheduler: warmuplr
+scheduler_conf:
+    warmup_steps: 10000

egs2/tedlium3/asr2/db.sh

@@ -0,0 +1 @@
+../../TEMPLATE/asr1/db.sh

egs2/tedlium3/asr2/local/data.sh

@@ -0,0 +1 @@
+../../asr1/local/data.sh

egs2/tedlium3/asr2/local/download_data.sh

@@ -0,0 +1 @@
+../../asr1/local/download_data.sh

egs2/tedlium3/asr2/local/join_suffix.py

@@ -0,0 +1 @@
+../../asr1/local/join_suffix.py

egs2/tedlium3/asr2/local/prepare_data.sh

@@ -0,0 +1 @@
+../../asr1/local/prepare_data.sh

egs2/tedlium3/asr2/path.sh

@@ -0,0 +1 @@
+../../TEMPLATE/asr1/path.sh

egs2/tedlium3/asr2/pyscripts

@@ -0,0 +1 @@
+../../TEMPLATE/asr1/pyscripts