ZHProject23/whisper-small-audiobook

Whisper Small Vietnamese

This model is a fine-tuned version of openai/whisper-small on the Vietnamese ASR Custom Corpus with 1 audiobook dataset. It achieves the following results on the evaluation set:

• Loss: 0.6499
• Wer: 78.4463

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 8
• training_steps: 80

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.0759	0.03	2	4.3984	33.2959
3.8512	0.05	4	3.7516	31.7467
3.1744	0.07	6	2.9537	31.2528
2.7374	0.1	8	2.4814	32.4427
2.3471	0.12	10	2.1517	72.1374
2.018	0.15	12	1.8614	127.7279
1.5275	0.17	14	1.5862	118.3431
1.4899	0.2	16	1.3252	127.5932
1.2433	0.23	18	1.1108	120.5433
1.0353	0.25	20	1.0291	112.0790
1.1132	0.28	22	0.9825	93.9156
1.0524	0.3	24	0.9412	87.9883
0.8907	0.33	26	0.9065	72.9232
0.8172	0.35	28	0.8786	64.9753
0.8563	0.38	30	0.8584	62.3934
1.0131	0.4	32	1.0352	56.7131
0.752	0.42	34	0.8189	62.1015
0.7312	0.45	36	0.8031	57.0723
0.8391	0.47	38	0.7888	63.4710
0.8875	0.5	40	0.7756	60.8217
0.7641	0.53	42	0.7633	61.7647
0.737	0.55	44	0.7519	61.2259
0.7782	0.57	46	0.7417	63.5384
0.6495	0.6	48	0.7318	67.0409
0.7102	0.62	50	0.7219	67.2205
0.7225	0.65	52	0.7132	72.9232
0.6752	0.68	54	0.7053	71.7782
0.679	0.7	56	0.6978	72.0476
0.6642	0.72	58	0.6906	72.8334
0.7048	0.75	60	0.6836	72.5191
0.6554	0.78	62	0.6773	72.4742
0.7454	0.8	64	0.6714	72.5415
0.6286	0.82	66	0.6663	74.7643
0.7423	0.85	68	0.6620	72.9681
0.8805	0.88	70	0.6584	77.0094
0.5701	0.9	72	0.6554	80.0404
0.5509	0.93	74	0.6531	80.0180
0.7618	0.95	76	0.6514	80.3996
0.6455	0.97	78	0.6503	79.8383
0.6748	1.0	80	0.6499	78.4463

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.15.0
• Tokenizers 0.15.0