Taeyeun72/whisper-small-noising_5

whisper-kor_noising_5

This model is a fine-tuned version of openai/whisper-small on the whisper-kor_noising_5 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0488
• Wer: 3.6157
• Cer: 1.8298

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: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.4197	0.19	100	0.4320	27.3865	13.2459
0.4509	0.37	200	0.3964	26.1729	12.8052
0.3414	0.56	300	0.3536	23.3204	10.5281
0.358	0.75	400	0.3171	22.9576	11.1100
0.2625	0.93	500	0.2747	20.5930	9.9462
0.1761	1.12	600	0.2419	17.1400	7.7968
0.1427	1.31	700	0.2118	14.9756	6.9055
0.1444	1.5	800	0.1740	12.4984	5.7619
0.126	1.68	900	0.1509	12.2732	6.4951
0.1047	1.87	1000	0.1282	10.3340	5.4255
0.0328	2.06	1100	0.1073	7.7318	3.8547
0.0321	2.24	1200	0.1017	7.3564	3.7437
0.0314	2.43	1300	0.0934	6.6058	3.2728
0.0271	2.62	1400	0.0824	6.1804	3.2089
0.0272	2.8	1500	0.0728	5.6174	2.8725
0.0213	2.99	1600	0.0690	5.9427	3.3434
0.0106	3.18	1700	0.0649	4.8668	2.5732
0.0087	3.36	1800	0.0619	4.8417	2.4992
0.0087	3.55	1900	0.0599	4.4414	2.2704
0.0086	3.74	2000	0.0563	4.5165	2.3646
0.0059	3.93	2100	0.0542	4.4039	2.2570
0.0033	4.11	2200	0.0540	4.0786	2.0350
0.0029	4.3	2300	0.0536	4.1787	2.1090
0.0028	4.49	2400	0.0514	3.9660	1.9946
0.003	4.67	2500	0.0511	3.9034	1.9475
0.003	4.86	2600	0.0505	3.6032	1.8937
0.0021	5.05	2700	0.0493	3.6157	1.8903
0.0018	5.23	2800	0.0495	3.5781	1.8130
0.0025	5.42	2900	0.0496	3.8033	1.9307
0.0018	5.61	3000	0.0495	3.6407	1.8500
0.0017	5.79	3100	0.0495	3.7408	1.8903
0.0017	5.98	3200	0.0491	3.6782	1.8903
0.0015	6.17	3300	0.0493	3.6532	1.8702
0.0015	6.36	3400	0.0490	3.6532	1.8702
0.0017	6.54	3500	0.0490	3.6282	1.8567
0.0014	6.73	3600	0.0490	3.7408	1.9273
0.0014	6.92	3700	0.0488	3.6282	1.8332
0.0014	7.1	3800	0.0488	3.6282	1.8365
0.0014	7.29	3900	0.0488	3.6157	1.8332
0.0013	7.48	4000	0.0488	3.6157	1.8298

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.1+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0