asr-africa/whisper_BIG-C_BEMBA_5hr_v1

Whisper Small Bemba - Beijuka Bruno

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

• Loss: 1.3874
• Wer: 0.4833
• Cer: 0.1314

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: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.3572	1.0	638	0.9726	0.6131	0.2029
0.7143	2.0	1276	0.8546	0.5769	0.1907
0.4485	3.0	1914	0.8472	0.5749	0.1936
0.2568	4.0	2552	0.9305	0.6149	0.1963
0.1328	5.0	3190	0.9798	0.5448	0.1658
0.0678	6.0	3828	1.0441	0.5373	0.1647
0.0366	7.0	4466	1.1153	0.5366	0.1719
0.0223	8.0	5104	1.1549	0.5441	0.1626
0.0169	9.0	5742	1.2010	0.5364	0.1698
0.0122	10.0	6380	1.2321	0.5384	0.1560
0.0098	11.0	7018	1.2646	0.5202	0.1522
0.0092	12.0	7656	1.2855	0.5279	0.1570
0.0085	13.0	8294	1.3001	0.5218	0.1568
0.0094	14.0	8932	1.3090	0.5240	0.1573
0.008	15.0	9570	1.3016	0.5220	0.1524
0.0064	16.0	10208	1.3748	0.5195	0.1532
0.0061	17.0	10846	1.3741	0.5147	0.1561
0.0072	18.0	11484	1.3958	0.5211	0.1560
0.007	19.0	12122	1.3849	0.5233	0.1554
0.0042	20.0	12760	1.4370	0.5124	0.1521
0.0046	21.0	13398	1.4660	0.5177	0.1529
0.0048	22.0	14036	1.4625	0.5115	0.1509
0.0057	23.0	14674	1.4502	0.5197	0.1554
0.005	24.0	15312	1.4473	0.5174	0.1568
0.0037	25.0	15950	1.4782	0.5258	0.1545
0.0035	26.0	16588	1.4798	0.5154	0.1533
0.0034	27.0	17226	1.5074	0.5170	0.1539
0.0042	28.0	17864	1.4976	0.5227	0.1572
0.0045	29.0	18502	1.5364	0.5110	0.1565
0.0025	30.0	19140	1.5072	0.5206	0.1545

Framework versions

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