From ce3a1ca14ae7e7d973f0e91da6794ccb85775fb4 Mon Sep 17 00:00:00 2001 From: William Chen <39677488+wanchichen@users.noreply.github.com> Date: Mon, 12 Feb 2024 17:11:31 -0500 Subject: [PATCH 1/5] Update README.md --- egs2/fleurs/asr1/README.md | 30 ++++++++++++++++++++++++++++++ 1 file changed, 30 insertions(+) diff --git a/egs2/fleurs/asr1/README.md b/egs2/fleurs/asr1/README.md index ad8dc6fe9d2..6b0d482f816 100644 --- a/egs2/fleurs/asr1/README.md +++ b/egs2/fleurs/asr1/README.md @@ -44,6 +44,36 @@ Finally, the model needs to be configured to use the data. # RESULTS + +# Multilingual ASR - SSL + E-Branchformer + Self-condition [XLS-R, E-Branchformer, utt_mvn, 6500 BPE](conf/train_asr_ebf_scctc.yaml) + +## Environments +- date: `Tue Feb 7 05:54:39 CST 2023` +- python version: `3.8.16 (default, Jan 17 2023, 23:13:24) [GCC 11.2.0]` +- espnet version: `espnet 202211` +- pytorch version: `pytorch 1.13.1+cu116` +- Git hash: `7f37bf7270017eede7d77701b389d1412f30078c` + - Commit date: `Sun Jan 1 13:06:01 2023 -0500` + +## asr_train_asr_branchformer_scctc_raw_all_bpe6500_sp +### WER + +|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err| +|---|---|---|---|---|---|---|---|---| +|lm0.4_pen0.0_lm_lm_train_lm_all_bpe6500_valid.loss.ave_asr_model_valid.acc.ave/test_all|77809|1669969|75.2|22.3|2.5|3.0|27.8|95.7| + +### CER + +|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err| +|---|---|---|---|---|---|---|---|---| +|lm0.4_pen0.0_lm_lm_train_lm_all_bpe6500_valid.loss.ave_asr_model_valid.acc.ave/test_all|77809|10235271|93.1|4.3|2.6|2.3|9.2|95.6| + +### TER + +|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err| +|---|---|---|---|---|---|---|---|---| +|lm0.4_pen0.0_lm_lm_train_lm_all_bpe6500_valid.loss.ave_asr_model_valid.acc.ave/test_all|77809|9622352|92.2|5.1|2.6|2.4|10.2|95.5| + # Multilingual ASR - SSL + Conformer + Hierarchical LID Self-condition [XLS-R, Conformer, utt_mvn, 6500 BPE](conf/train_asr_conformer_hier_lid_utt.yaml) ## Environments From 00ed5b1c7677fa590ff047bd68a99883079f5b69 Mon Sep 17 00:00:00 2001 From: William Chen <39677488+wanchichen@users.noreply.github.com> Date: Mon, 12 Feb 2024 17:14:14 -0500 Subject: [PATCH 2/5] Update README.md --- egs2/fleurs/asr1/README.md | 1 + 1 file changed, 1 insertion(+) diff --git a/egs2/fleurs/asr1/README.md b/egs2/fleurs/asr1/README.md index 6b0d482f816..236e99f9213 100644 --- a/egs2/fleurs/asr1/README.md +++ b/egs2/fleurs/asr1/README.md @@ -54,6 +54,7 @@ Finally, the model needs to be configured to use the data. - pytorch version: `pytorch 1.13.1+cu116` - Git hash: `7f37bf7270017eede7d77701b389d1412f30078c` - Commit date: `Sun Jan 1 13:06:01 2023 -0500` +- Pre-trained model: https://huggingface.co/espnet/wanchichen_fleurs_asr_ebf_scctc ## asr_train_asr_branchformer_scctc_raw_all_bpe6500_sp ### WER From 0f7b533e7eed20b48cc43a1e168e49480c68b85d Mon Sep 17 00:00:00 2001 From: William Chen <39677488+wanchichen@users.noreply.github.com> Date: Mon, 12 Feb 2024 17:15:20 -0500 Subject: [PATCH 3/5] Create train_asr_ebf_scctc.yaml --- .../asr1/conf/tuning/train_asr_ebf_scctc.yaml | 94 +++++++++++++++++++ 1 file changed, 94 insertions(+) create mode 100644 egs2/fleurs/asr1/conf/tuning/train_asr_ebf_scctc.yaml diff --git a/egs2/fleurs/asr1/conf/tuning/train_asr_ebf_scctc.yaml b/egs2/fleurs/asr1/conf/tuning/train_asr_ebf_scctc.yaml new file mode 100644 index 00000000000..40c00ab822a --- /dev/null +++ b/egs2/fleurs/asr1/conf/tuning/train_asr_ebf_scctc.yaml @@ -0,0 +1,94 @@ +batch_type: numel +batch_bins: 40000000 +accum_grad: 4 +max_epoch: 20 +patience: none +# The initialization method for model parameters +init: xavier_uniform +best_model_criterion: +- - valid + - acc + - max +keep_nbest_models: 3 + +encoder: e_branchformer +encoder_conf: + output_size: 512 + attention_heads: 8 + attention_layer_type: rel_selfattn + pos_enc_layer_type: rel_pos + rel_pos_type: latest + cgmlp_linear_units: 3072 + 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: conv2d2 + layer_drop_rate: 0.1 + linear_units: 1024 + positionwise_layer_type: linear + macaron_ffn: true + use_ffn: true + merge_conv_kernel: 31 + interctc_layer_idx: [3, 6, 9] + interctc_use_conditioning: true + +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: 0.1 + src_attention_dropout_rate: 0.1 + +model_conf: + ctc_weight: 0.3 + lsm_weight: 0.1 + interctc_weight: 0.5 + length_normalized_loss: false + extract_feats_in_collect_stats: false + +optim: adam +optim_conf: + lr: 0.002 +scheduler: warmuplr +scheduler_conf: + warmup_steps: 25000 + +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 + +freeze_param: [ +"frontend.upstream" +] + +frontend: s3prl +frontend_conf: + frontend_conf: + upstream: wav2vec2_url # Note: If the upstream is changed, please change the input_size in the preencoder. + path_or_url: https://huggingface.co/s3prl/converted_ckpts/resolve/main/xlsr2_300m.pt + download_dir: ./hub + multilayer_feature: True + +preencoder: linear +preencoder_conf: + input_size: 1024 # Note: If the upstream is changed, please change this value accordingly. + output_size: 80 From 86409dfc5c5e5f8dfa731416098cce3fc5b99ee6 Mon Sep 17 00:00:00 2001 From: William Chen <39677488+wanchichen@users.noreply.github.com> Date: Mon, 12 Feb 2024 17:15:41 -0500 Subject: [PATCH 4/5] Update README.md --- egs2/fleurs/asr1/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/egs2/fleurs/asr1/README.md b/egs2/fleurs/asr1/README.md index 236e99f9213..3412bae5b9e 100644 --- a/egs2/fleurs/asr1/README.md +++ b/egs2/fleurs/asr1/README.md @@ -45,7 +45,7 @@ Finally, the model needs to be configured to use the data. -# Multilingual ASR - SSL + E-Branchformer + Self-condition [XLS-R, E-Branchformer, utt_mvn, 6500 BPE](conf/train_asr_ebf_scctc.yaml) +# Multilingual ASR - SSL + E-Branchformer + Self-condition [XLS-R, E-Branchformer, utt_mvn, 6500 BPE](conf/tuning/train_asr_ebf_scctc.yaml) ## Environments - date: `Tue Feb 7 05:54:39 CST 2023` From 77b6c0d60149b7ffd2d1d7b77bcd070a7f792437 Mon Sep 17 00:00:00 2001 From: "pre-commit-ci[bot]" <66853113+pre-commit-ci[bot]@users.noreply.github.com> Date: Mon, 12 Feb 2024 22:19:03 +0000 Subject: [PATCH 5/5] [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --- egs2/fleurs/asr1/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/egs2/fleurs/asr1/README.md b/egs2/fleurs/asr1/README.md index 3412bae5b9e..597a30bd41b 100644 --- a/egs2/fleurs/asr1/README.md +++ b/egs2/fleurs/asr1/README.md @@ -54,7 +54,7 @@ Finally, the model needs to be configured to use the data. - pytorch version: `pytorch 1.13.1+cu116` - Git hash: `7f37bf7270017eede7d77701b389d1412f30078c` - Commit date: `Sun Jan 1 13:06:01 2023 -0500` -- Pre-trained model: https://huggingface.co/espnet/wanchichen_fleurs_asr_ebf_scctc +- Pre-trained model: https://huggingface.co/espnet/wanchichen_fleurs_asr_ebf_scctc ## asr_train_asr_branchformer_scctc_raw_all_bpe6500_sp ### WER