nkkbr/whisper-NG_word_detect

NG_word_detect

This model is a fine-tuned version of openai/whisper-large-v3 on the NG_word_detect dataset. It achieves the following results on the evaluation set:

• Loss: 0.2154
• Wer: 40.9602

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: 5e-06
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3435	0.1524	25	0.3604	69.1673
0.2776	0.3049	50	0.2718	63.0158
0.2031	0.4573	75	0.2177	55.4389
0.171	0.6098	100	0.1879	52.3631
0.1363	0.7622	125	0.1721	49.8875
0.1587	0.9146	150	0.1654	48.9872
0.0824	1.0671	175	0.1641	47.2618
0.0933	1.2195	200	0.1659	48.3121
0.1426	1.3720	225	0.1572	46.5116
0.1059	1.5244	250	0.1528	45.6864
0.095	1.6768	275	0.1540	46.0615
0.0855	1.8293	300	0.1528	44.1110
0.1124	1.9817	325	0.1525	45.1613
0.052	2.1341	350	0.1559	45.5364
0.0539	2.2866	375	0.1575	45.0863
0.0718	2.4390	400	0.1667	45.1613
0.0451	2.5915	425	0.1701	46.0615
0.0421	2.7439	450	0.1582	44.4861
0.0508	2.8963	475	0.1604	44.4111
0.0204	3.0488	500	0.1601	42.7607
0.0257	3.2012	525	0.1744	43.9610
0.0175	3.3537	550	0.1728	45.7614
0.0219	3.5061	575	0.1766	45.2363
0.0216	3.6585	600	0.1800	45.9115
0.0173	3.8110	625	0.1692	44.5611
0.0418	3.9634	650	0.1672	43.7359
0.0076	4.1159	675	0.1777	43.6609
0.0088	4.2683	700	0.1805	42.4606
0.0097	4.4207	725	0.1774	43.0608
0.0097	4.5732	750	0.1802	44.7112
0.0117	4.7256	775	0.1783	43.5859
0.0101	4.8780	800	0.1851	42.9107
0.0069	5.0305	825	0.1807	41.9355
0.006	5.1829	850	0.1865	42.2356
0.0029	5.3354	875	0.1878	42.6107
0.0079	5.4878	900	0.1994	44.1110
0.0118	5.6402	925	0.1889	43.9610
0.0125	5.7927	950	0.1905	44.6362
0.0115	5.9451	975	0.1846	44.0360
0.0054	6.0976	1000	0.1845	43.8110
0.0036	6.25	1025	0.1922	42.7607
0.0088	6.4024	1050	0.1937	42.8357
0.0043	6.5549	1075	0.1914	42.9107
0.0016	6.7073	1100	0.1958	42.6107
0.0103	6.8598	1125	0.1877	41.6354
0.0027	7.0122	1150	0.1873	41.7104
0.0018	7.1646	1175	0.1890	41.7854
0.0012	7.3171	1200	0.1918	41.7104
0.0054	7.4695	1225	0.1949	41.0353
0.0014	7.6220	1250	0.1965	41.6354
0.0009	7.7744	1275	0.2024	41.7104
0.0011	7.9268	1300	0.1970	41.1853
0.0007	8.0793	1325	0.1995	41.1103
0.0006	8.2317	1350	0.2012	41.4854
0.0006	8.3841	1375	0.2075	41.7854
0.0006	8.5366	1400	0.2077	41.5604
0.0034	8.6890	1425	0.2092	41.7854
0.0006	8.8415	1450	0.2079	41.2603
0.0023	8.9939	1475	0.2080	41.0353
0.0004	9.1463	1500	0.2095	41.0353
0.0021	9.2988	1525	0.2096	41.4854
0.0004	9.4512	1550	0.2095	41.3353
0.0015	9.6037	1575	0.2102	41.0353
0.0012	9.7561	1600	0.2106	41.1853
0.0006	9.9085	1625	0.2110	41.2603
0.0004	10.0610	1650	0.2111	41.1103
0.0003	10.2134	1675	0.2122	41.1853
0.0003	10.3659	1700	0.2122	40.9602
0.0006	10.5183	1725	0.2125	40.8102
0.0004	10.6707	1750	0.2131	40.8852
0.0004	10.8232	1775	0.2137	41.0353
0.0003	10.9756	1800	0.2141	40.9602
0.0003	11.1280	1825	0.2144	40.9602
0.0003	11.2805	1850	0.2147	40.9602
0.0021	11.4329	1875	0.2149	40.9602
0.0011	11.5854	1900	0.2152	40.9602
0.0003	11.7378	1925	0.2153	40.9602
0.0003	11.8902	1950	0.2153	40.9602
0.0024	12.0427	1975	0.2153	40.9602
0.0003	12.1951	2000	0.2154	40.9602

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1