Shallow Fusionを試してみる

[abePclWaseda]

What?

Why?

See also

[abePclWaseda]

RESULTS

Environments

• date: Mon Sep 23 22:32:43 JST 2024
• python version: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0]
• espnet version: espnet 202402
• pytorch version: pytorch 1.13.1
• Git hash: 2ec7ab035ba83ad9fcfafcf7547114c341f1082c
  • Commit date: Sun Sep 22 16:57:56 2024 +0900

exp/asr_train_asr_raw_en_hugging_face_facebook-opt-125m_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_clean	2703	54402	5.7	31.0	63.4	1.1	95.4	100.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_other	2864	50948	5.4	34.7	59.9	1.1	95.7	100.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_clean	2620	52576	5.6	32.8	61.6	0.9	95.3	100.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_other	2939	52343	5.4	35.0	59.6	1.1	95.7	100.0

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_clean	2703	288456	21.1	9.4	69.5	1.1	79.9	100.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_other	2864	265951	22.3	11.0	66.7	1.3	79.0	100.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_clean	2620	281530	21.5	9.9	68.6	1.1	79.6	100.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_other	2939	272758	22.3	11.3	66.4	1.4	79.0	100.0

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err

[abePclWaseda]

RESULTS

Environments

• date: Mon Oct 7 00:33:19 JST 2024
• python version: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0]
• espnet version: espnet 202402
• pytorch version: pytorch 1.13.1
• Git hash: aeeae86e3c288f10a34efcc53f1f67eed07d2c36
  • Commit date: Sun Oct 6 19:09:11 2024 +0900

exp/asr_train_asr_conformer_lr2e-3_warmup15k_amp_nondeterministic_raw_en_hugging_face_facebook-opt-125m_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_clean	2703	54402	95.8	3.9	0.3	0.6	4.8	47.8
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_other	2864	50948	87.8	11.0	1.2	1.7	14.0	72.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_clean	2620	52576	95.7	4.0	0.4	0.6	5.0	47.7
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_other	2939	52343	87.6	11.0	1.4	1.5	13.9	74.5

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_clean	2703	288456	98.6	0.8	0.6	0.5	2.0	47.8
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_other	2864	265951	94.5	3.3	2.2	1.6	7.1	72.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_clean	2620	281530	98.6	0.8	0.6	0.5	1.9	47.7
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_other	2939	272758	94.6	3.2	2.2	1.5	6.9	74.5

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err

[abePclWaseda]

RESULTS

Environments

• date: Mon Oct 7 15:36:24 JST 2024
• python version: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0]
• espnet version: espnet 202402
• pytorch version: pytorch 1.13.1
• Git hash: 649f9b64897a4cb9b22242704c806ba2a3796f28
  • Commit date: Mon Oct 7 11:41:29 2024 +0900

exp/asr_train_asr_conformer_lr2e-3_warmup15k_amp_nondeterministic_raw_en_hugging_face_facebook-opt-125m_sp

WER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_clean	2703	54402	94.5	5.1	0.4	0.8	6.3	56.7
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_other	2864	50948	84.6	14.0	1.4	2.1	17.5	80.7
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_clean	2620	52576	94.3	5.3	0.4	0.9	6.6	57.4
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_other	2939	52343	84.4	14.0	1.6	2.0	17.6	81.9
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave_shallowfusion1007/dev_clean	2703	54402	95.8	3.9	0.3	0.6	4.8	47.8
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave_shallowfusion1007/dev_other	2864	50948	87.8	11.0	1.2	1.7	14.0	72.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave_shallowfusion1007/test_clean	2620	52576	95.7	4.0	0.4	0.6	5.0	47.7
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave_shallowfusion1007/test_other	2939	52343	87.6	11.0	1.4	1.5	13.9	74.5

CER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_clean	2703	288456	98.3	1.0	0.7	0.7	2.4	56.7
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/dev_other	2864	265951	93.5	4.1	2.5	1.9	8.4	80.7
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_clean	2620	281530	98.2	1.1	0.7	0.6	2.4	57.4
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave/test_other	2939	272758	93.5	3.9	2.6	1.8	8.3	81.9
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave_shallowfusion1007/dev_clean	2703	288456	98.6	0.8	0.6	0.5	2.0	47.8
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave_shallowfusion1007/dev_other	2864	265951	94.5	3.3	2.2	1.6	7.1	72.0
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave_shallowfusion1007/test_clean	2620	281530	98.6	0.8	0.6	0.5	1.9	47.7
decode_asr_lm_lm_train_transformer_opt_notLSM_en_hugging_face_valid.loss.ave_asr_model_valid.acc.ave_shallowfusion1007/test_other	2939	272758	94.6	3.2	2.2	1.5	6.9	74.5

TER

dataset	Snt	Wrd	Corr	Sub	Del	Ins	Err	S.Err