asr-africa/whisper_BIG-C_BEMBA_10hr_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: 1.1631
• Model Preparation Time: 0.0092
• Wer: 0.4525
• Cer: 0.1227

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	Model Preparation Time	Wer	Cer
1.1298	1.0	1281	0.8440	0.0092	0.5704	0.1924
0.6239	2.0	2562	0.7778	0.0092	0.5489	0.1840
0.4117	3.0	3843	0.7974	0.0092	0.5208	0.1680
0.2409	4.0	5124	0.8600	0.0092	0.5432	0.1883
0.1249	5.0	6405	0.9344	0.0092	0.5202	0.1696
0.0619	6.0	7686	1.0230	0.0092	0.5079	0.1667
0.0349	7.0	8967	1.0739	0.0092	0.5135	0.1649
0.0232	8.0	10248	1.1178	0.0092	0.5039	0.1673
0.0173	9.0	11529	1.1459	0.0092	0.5154	0.1670
0.0139	10.0	12810	1.2013	0.0092	0.5139	0.1672
0.0124	11.0	14091	1.2302	0.0092	0.5133	0.1658
0.011	12.0	15372	1.2629	0.0092	0.5142	0.1732
0.0084	13.0	16653	1.3002	0.0092	0.5135	0.1650
0.0075	14.0	17934	1.3475	0.0092	0.4972	0.1629
0.0081	15.0	19215	1.3522	0.0092	0.4931	0.1618
0.0077	16.0	20496	1.3592	0.0092	0.5087	0.1623
0.0073	17.0	21777	1.3655	0.0092	0.5067	0.1662
0.0061	18.0	23058	1.3930	0.0092	0.5074	0.1669
0.0057	19.0	24339	1.3912	0.0092	0.5055	0.1636
0.0065	20.0	25620	1.4236	0.0092	0.4995	0.1641
0.0052	21.0	26901	1.4587	0.0092	0.5035	0.1609
0.0044	22.0	28182	1.4459	0.0092	0.5034	0.1653
0.006	23.0	29463	1.4685	0.0092	0.5036	0.1684
0.0051	24.0	30744	1.4455	0.0092	0.5029	0.1651
0.0043	25.0	32025	1.4682	0.0092	0.5410	0.1875

Framework versions

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