Taeyeun72/whisper-small-noising_3

whisper-kor_noising_3

This model is a fine-tuned version of openai/whisper-small on the whisper-kor_noising_3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2748
• Wer: 19.5871
• Cer: 8.9142

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
0.2751	0.05	100	0.3078	20.8754	10.0300
0.3055	0.09	200	0.2981	20.7046	9.6834
0.2684	0.14	300	0.2974	21.1703	9.7215
0.267	0.18	400	0.3019	21.7600	10.4611
0.2927	0.23	500	0.3014	20.7357	9.5862
0.287	0.28	600	0.3057	21.5117	9.9370
0.2913	0.32	700	0.3098	22.8620	10.9218
0.3201	0.37	800	0.3044	22.7534	10.9683
0.2929	0.42	900	0.2994	21.1237	9.7553
0.2661	0.46	1000	0.3023	22.3809	10.6218
0.2865	0.51	1100	0.3013	24.5538	11.4755
0.2668	0.55	1200	0.3011	23.4052	11.0951
0.2888	0.6	1300	0.2956	24.4296	13.4494
0.245	0.65	1400	0.3015	21.2323	9.8821
0.2718	0.69	1500	0.3009	21.4807	9.7468
0.2757	0.74	1600	0.2950	20.7357	9.5862
0.2943	0.78	1700	0.2965	21.1237	9.7510
0.2637	0.83	1800	0.2934	21.9618	10.5372
0.2593	0.88	1900	0.2911	21.9929	10.4231
0.2742	0.92	2000	0.2888	22.2257	11.2642
0.2682	0.97	2100	0.2866	20.9530	9.7806
0.172	1.02	2200	0.2858	19.8044	9.1044
0.165	1.06	2300	0.2875	19.6027	9.0452
0.1634	1.11	2400	0.2868	19.5871	9.0621
0.1928	1.15	2500	0.2853	22.0705	11.0233
0.1876	1.2	2600	0.2832	21.7911	10.8035
0.1795	1.25	2700	0.2826	19.7113	9.1255
0.1844	1.29	2800	0.2821	19.6803	9.0494
0.1532	1.34	2900	0.2797	19.7579	9.0790
0.1529	1.39	3000	0.2783	20.0683	9.0748
0.1334	1.43	3100	0.2795	19.7579	9.0579
0.1538	1.48	3200	0.2787	20.7667	10.0850
0.1537	1.52	3300	0.2785	19.5406	8.7704
0.1694	1.57	3400	0.2780	19.6492	8.8085
0.1811	1.62	3500	0.2766	19.5406	8.9015
0.163	1.66	3600	0.2772	21.2634	10.2033
0.1445	1.71	3700	0.2763	19.3854	8.8169
0.1548	1.75	3800	0.2750	19.4009	8.7958
0.1588	1.8	3900	0.2749	19.3854	8.8127
0.1575	1.85	4000	0.2748	19.5871	8.9142

Framework versions

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