Hubert-noisy-cv-kakeiken

This model is a fine-tuned version of rinna/japanese-hubert-base on the ORIGINAL_NOISY_COMMON_VOICE_AND_KAKEIKEN - JA dataset. It achieves the following results on the evaluation set:

• Loss: 0.9441
• Wer: 1.0
• Cer: 0.3276

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 12500
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3007	1.0	3463	0.9433	1.0	0.3277
0.1409	2.0	6926	1.0068	1.0	0.3606
0.1444	3.0	10389	1.0954	1.0	0.3839
0.1518	4.0	13852	1.2021	1.0016	0.4125
0.1691	5.0	17315	1.3227	1.0224	0.4465
0.1612	6.0	20778	1.2268	1.0087	0.4165
0.155	7.0	24241	1.3089	1.0160	0.4389
0.1529	8.0	27704	1.2341	1.0017	0.4234
0.1458	9.0	31167	1.2319	1.0095	0.4250
0.1371	10.0	34630	1.1689	1.0041	0.4131
0.1295	11.0	38093	1.2024	1.0278	0.4175
0.1347	12.0	41556	1.2089	1.0142	0.4192
0.1161	13.0	45019	1.1461	1.0371	0.3998
0.1162	14.0	48482	1.1236	1.0311	0.3920
0.1107	15.0	51945	1.0697	1.0276	0.3797
0.1029	16.0	55408	1.0551	1.0108	0.3806
0.0992	17.0	58871	1.0634	1.0187	0.3727
0.0906	18.0	62334	1.0299	1.0273	0.3657
0.0793	19.0	65797	1.0217	1.0149	0.3602
0.0769	20.0	69260	1.0025	1.0334	0.3533
0.0727	21.0	72723	1.0101	1.0386	0.3510
0.0654	22.0	76186	1.0316	1.0345	0.3494
0.0605	23.0	79649	1.0584	1.0254	0.3438
0.0566	24.0	83112	1.0380	1.0479	0.3431
0.0507	25.0	86575	1.0691	1.0427	0.3413
0.0498	26.0	90038	1.1261	1.0399	0.3407
0.0444	27.0	93501	1.1671	1.0578	0.3417
0.0444	28.0	96964	1.1998	1.0621	0.3414
0.0439	29.0	100427	1.1988	1.0568	0.3406
0.0441	29.9915	103860	1.2041	1.0594	0.3410

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3