KasuleTrevor/whisper-ln-afrivoice-10hr-v4r

Whisper Small Hi - Sanchit Gandhi

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

• Loss: 1.1668
• Wer: 0.3479
• Cer: 0.1439

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: 3e-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_ratio: 0.1
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.6425	1.0	123	1.0239	0.5476	0.2091
0.8057	2.0	246	0.7465	0.4388	0.1797
0.563	3.0	369	0.6723	0.3866	0.1537
0.4016	4.0	492	0.6578	0.3876	0.1594
0.2677	5.0	615	0.6799	0.3976	0.1730
0.1708	6.0	738	0.7148	0.3929	0.1585
0.1146	7.0	861	0.7365	0.3840	0.1574
0.0786	8.0	984	0.7790	0.3839	0.1555
0.0646	9.0	1107	0.8027	0.3846	0.1584
0.0587	10.0	1230	0.8323	0.3857	0.1589
0.0527	11.0	1353	0.8354	0.3870	0.1548
0.0442	12.0	1476	0.8494	0.3800	0.1554
0.0352	13.0	1599	0.8863	0.3957	0.1577
0.0317	14.0	1722	0.8744	0.4027	0.1716
0.0289	15.0	1845	0.8952	0.3723	0.1566
0.0249	16.0	1968	0.9174	0.3728	0.1513
0.022	17.0	2091	0.9266	0.3764	0.1608
0.0197	18.0	2214	0.9473	0.3840	0.1608
0.0182	19.0	2337	0.9502	0.3780	0.1549
0.0161	20.0	2460	0.9454	0.3731	0.1536
0.0139	21.0	2583	0.9579	0.3716	0.1557
0.0128	22.0	2706	0.9722	0.3658	0.1510
0.0107	23.0	2829	0.9775	0.3738	0.1585
0.0116	24.0	2952	0.9750	0.3721	0.1587
0.0118	25.0	3075	0.9746	0.3706	0.1584
0.0108	26.0	3198	1.0020	0.3743	0.1617
0.0092	27.0	3321	0.9905	0.3639	0.1531
0.0086	28.0	3444	1.0039	0.3683	0.1546
0.0092	29.0	3567	1.0071	0.3697	0.1557
0.0071	30.0	3690	1.0074	0.3656	0.1502
0.0049	31.0	3813	1.0162	0.3621	0.1558
0.0074	32.0	3936	1.0396	0.3667	0.1592
0.0057	33.0	4059	1.0447	0.3684	0.1592
0.005	34.0	4182	1.0465	0.3658	0.1541
0.0051	35.0	4305	1.0388	0.3662	0.1539
0.0061	36.0	4428	1.0512	0.3581	0.1470
0.005	37.0	4551	1.0480	0.3668	0.1552
0.005	38.0	4674	1.0476	0.3641	0.1558
0.0046	39.0	4797	1.0402	0.3656	0.1499
0.0028	40.0	4920	1.0726	0.3613	0.1481
0.0028	41.0	5043	1.0624	0.3575	0.1463
0.0018	42.0	5166	1.0882	0.3599	0.1510
0.0033	43.0	5289	1.0653	0.3594	0.1506
0.0042	44.0	5412	1.0725	0.3586	0.1494
0.0039	45.0	5535	1.0754	0.3610	0.1489
0.003	46.0	5658	1.0761	0.3682	0.1551
0.0026	47.0	5781	1.0815	0.3667	0.1561
0.0026	48.0	5904	1.0740	0.3676	0.1527
0.0026	49.0	6027	1.0762	0.3604	0.1501
0.0019	50.0	6150	1.1023	0.3563	0.1485
0.0016	51.0	6273	1.0775	0.3564	0.1475
0.0017	52.0	6396	1.0821	0.3535	0.1479
0.001	53.0	6519	1.0904	0.3523	0.1448
0.0005	54.0	6642	1.0975	0.3523	0.1453
0.0004	55.0	6765	1.1008	0.3503	0.1463
0.0003	56.0	6888	1.1026	0.3547	0.1493
0.0001	57.0	7011	1.1079	0.3510	0.1454
0.0001	58.0	7134	1.1128	0.3503	0.1449
0.0001	59.0	7257	1.1175	0.3501	0.1454
0.0001	60.0	7380	1.1217	0.3503	0.1448
0.0001	61.0	7503	1.1255	0.3506	0.1449
0.0	62.0	7626	1.1289	0.3499	0.1446
0.0	63.0	7749	1.1320	0.3501	0.1447
0.0	64.0	7872	1.1349	0.3493	0.1443
0.0	65.0	7995	1.1377	0.3491	0.1442
0.0	66.0	8118	1.1403	0.3490	0.1441
0.0	67.0	8241	1.1427	0.3486	0.1442
0.0	68.0	8364	1.1452	0.3486	0.1441
0.0	69.0	8487	1.1474	0.3489	0.1441
0.0	70.0	8610	1.1495	0.3490	0.1441
0.0	71.0	8733	1.1517	0.3485	0.1439
0.0	72.0	8856	1.1538	0.3484	0.1437
0.0	73.0	8979	1.1557	0.3483	0.1437
0.0	74.0	9102	1.1577	0.3483	0.1437
0.0	75.0	9225	1.1597	0.3483	0.1438
0.0	76.0	9348	1.1615	0.3483	0.1438
0.0	77.0	9471	1.1633	0.3484	0.1441
0.0	78.0	9594	1.1651	0.3482	0.1439
0.0	79.0	9717	1.1668	0.3479	0.1439
0.0	80.0	9840	1.1685	0.3480	0.1439
0.0	81.0	9963	1.1701	0.3481	0.1439
0.0	82.0	10086	1.1717	0.3480	0.1439
0.0	83.0	10209	1.1732	0.3486	0.1441
0.0	84.0	10332	1.1747	0.3485	0.1437
0.0	85.0	10455	1.1761	0.3489	0.1438
0.0	86.0	10578	1.1774	0.3489	0.1439
0.0	87.0	10701	1.1788	0.3489	0.1437
0.0	88.0	10824	1.1800	0.3490	0.1436
0.0	89.0	10947	1.1812	0.3491	0.1438

Framework versions

• Transformers 4.44.2
• Pytorch 2.1.0+cu118
• Datasets 3.0.0
• Tokenizers 0.19.1