asr-africa/whisper_BIG-C_BEMBA_167hr_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: 0.4520
• Wer: 0.3491
• Cer: 0.0971

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_ratio: 0.025
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.9127	1.0	5143	0.5881	0.4483	0.1252
0.5091	2.0	10286	0.4981	0.3918	0.1136
0.4171	3.0	15429	0.4668	0.3636	0.1024
0.3332	4.0	20572	0.4638	0.3551	0.1022
0.251	5.0	25715	0.4828	0.3585	0.1101
0.1689	6.0	30858	0.5249	0.3631	0.1102
0.0992	7.0	36001	0.5907	0.3645	0.1078
0.0548	8.0	41144	0.6471	0.3676	0.1082
0.034	9.0	46287	0.7023	0.3646	0.1071
0.0252	10.0	51430	0.7307	0.3707	0.1129
0.0207	11.0	56573	0.7652	0.3652	0.1071
0.0178	12.0	61716	0.7873	0.3653	0.1088
0.0161	13.0	66859	0.8036	0.3643	0.1093
0.0144	14.0	72002	0.8223	0.3573	0.1064

Framework versions

• Transformers 4.45.2
• Pytorch 2.2.0+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1