asr-africa/whisper_DigitalUmuganda_Afrivoice_Shona_20hr_v1

Whisper Small Shona - Beijuka Bruno

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

• Loss: 0.9407
• Wer: 0.4097
• Cer: 0.1044

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
0.828	1.0	876	0.4722	0.3663	0.0789
0.3137	2.0	1752	0.4031	0.3228	0.0660
0.182	3.0	2628	0.4081	0.3196	0.0667
0.102	4.0	3504	0.4312	0.3207	0.0658
0.0537	5.0	4380	0.4600	0.3150	0.0617
0.0292	6.0	5256	0.4840	0.3196	0.0669
0.0178	7.0	6132	0.5033	0.3123	0.0615
0.0118	8.0	7008	0.5443	0.3068	0.0621
0.0092	9.0	7884	0.5597	0.3062	0.0610
0.0072	10.0	8760	0.5778	0.3176	0.0641
0.0063	11.0	9636	0.5991	0.3116	0.0623
0.006	12.0	10512	0.5886	0.2986	0.0597
0.0053	13.0	11388	0.6122	0.3099	0.0625
0.0053	14.0	12264	0.6129	0.3070	0.0614
0.0054	15.0	13140	0.6246	0.2990	0.0604
0.0051	16.0	14016	0.6465	0.3105	0.0608
0.0037	17.0	14892	0.6433	0.3040	0.0620
0.0036	18.0	15768	0.6522	0.3039	0.0613
0.004	19.0	16644	0.6465	0.2983	0.0595
0.0035	20.0	17520	0.6700	0.3049	0.0603
0.0035	21.0	18396	0.6835	0.2988	0.0596
0.0027	22.0	19272	0.6802	0.3068	0.0653

Framework versions

• Transformers 4.44.2
• Pytorch 2.2.0+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1