emason/whisper-base-sw

Whisper Base Swahili - Munene Mutuma

This model is a fine-tuned version of openai/whisper-base on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6033
• Wer: 39.3269

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: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5121	1.0893	1000	0.7500	50.1768
0.3802	2.1786	2000	0.6447	43.3016
0.31	3.2680	3000	0.6139	39.4864
0.2768	4.3573	4000	0.6029	39.2423
0.266	5.4466	5000	0.6033	39.3269

Framework versions

• Transformers 4.48.0
• Pytorch 2.7.0a0+79aa17489c.nv25.04
• Datasets 3.6.0
• Tokenizers 0.21.4