Taeyeun72/whisper-small

whisper-kor3

This model is a fine-tuned version of openai/whisper-small on the whisper-kor3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4157
• Wer: 24.6903
• Cer: 11.3851

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

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.2195	0.05	100	1.0198	34.4857	16.2544
0.7295	0.09	200	0.7220	32.6995	14.9684
0.5236	0.14	300	0.5703	31.4463	14.2549
0.4976	0.18	400	0.5461	31.8640	14.6274
0.479	0.23	500	0.5296	30.4091	14.0902
0.4544	0.28	600	0.5219	31.7920	16.3916
0.4672	0.32	700	0.5100	30.4955	13.9138
0.4305	0.37	800	0.5043	30.1354	14.5960
0.4561	0.42	900	0.4941	28.8101	13.2513
0.398	0.46	1000	0.4846	31.3166	14.2980
0.4338	0.51	1100	0.4780	28.0755	12.8945
0.4121	0.55	1200	0.4728	27.4128	12.5417
0.4217	0.6	1300	0.4693	28.2772	14.4392
0.3881	0.65	1400	0.4639	27.6577	13.0082
0.4035	0.69	1500	0.4593	26.9231	12.4436
0.4146	0.74	1600	0.4555	28.4212	13.7609
0.3837	0.78	1700	0.4511	28.8822	13.7845
0.3969	0.83	1800	0.4485	29.2135	14.2235
0.4368	0.88	1900	0.4414	26.5918	12.1457
0.3679	0.92	2000	0.4376	26.4477	12.1770
0.4496	0.97	2100	0.4335	30.1354	14.9018
0.3049	1.02	2200	0.4314	26.1164	12.9180
0.2213	1.06	2300	0.4325	25.9147	11.8046
0.2732	1.11	2400	0.4303	26.0012	11.8987
0.2568	1.15	2500	0.4293	25.9291	11.7576
0.2456	1.2	2600	0.4289	25.6986	11.7066
0.2702	1.25	2700	0.4262	25.8283	11.8203
0.2744	1.29	2800	0.4235	25.8139	11.8124
0.2742	1.34	2900	0.4254	25.6266	11.6360
0.2798	1.39	3000	0.4238	25.5546	11.6399
0.2593	1.43	3100	0.4219	26.1020	12.4632
0.2619	1.48	3200	0.4208	25.3241	11.4714
0.2633	1.52	3300	0.4210	26.6350	12.9964
0.2603	1.57	3400	0.4189	25.2809	11.4243
0.2992	1.62	3500	0.4189	25.2377	11.3969
0.2453	1.66	3600	0.4176	25.2377	11.5145
0.2475	1.71	3700	0.4172	24.8487	11.3969
0.2545	1.75	3800	0.4164	25.0216	11.4596
0.272	1.8	3900	0.4160	24.6471	11.2714
0.2339	1.85	4000	0.4157	24.6903	11.3851

Framework versions

• Transformers 4.33.2
• Pytorch 2.0.1+cu117
• Datasets 2.14.5
• Tokenizers 0.13.3