CasperMuz/whisper-base-sna-ven-cleaned

Whisper Base Shona + Tshivenda - Cleaned Data

This model is a fine-tuned version of openai/whisper-base on the Cleaned WAXAL Shona + Swivuriso Tshivenda dataset. It achieves the following results on the evaluation set:

• Loss: 0.5899
• Wer: 48.7657

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: 8
• seed: 42
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1846	0.0882	200	1.2058	79.2304
0.8066	0.1764	400	0.8677	65.0463
0.7158	0.2646	600	0.7538	57.1436
0.6391	0.3527	800	0.6928	55.1938
0.6140	0.4409	1000	0.6560	52.6030
0.5697	0.5291	1200	0.6290	51.2560
0.5357	0.6173	1400	0.6114	48.9857
0.5352	0.7055	1600	0.5992	48.9450
0.5247	0.7937	1800	0.5924	48.7657
0.5443	0.8818	2000	0.5899	48.7657

Framework versions

• Transformers 5.14.1
• Pytorch 2.13.0+cu130
• Datasets 5.0.1
• Tokenizers 0.22.2