deepdml/whisper-base-ta-mix-norm

Whisper Base ta

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: 0.2168
• Wer: 44.3470
• Cer: 9.3419

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: 64
• seed: 42
• optimizer: Use 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_ratio: 0.04
• training_steps: 8000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.1907	0.125	1000	0.2800	54.7140	12.3286
0.1294	0.25	2000	0.2466	50.1204	10.8475
0.1125	0.375	3000	0.2416	48.2552	10.6527
0.0839	0.5	4000	0.2322	46.5587	10.0303
0.0965	0.625	5000	0.2219	45.3337	9.5889
0.0719	0.75	6000	0.2191	44.6793	9.3874
0.0753	0.875	7000	0.2155	44.4101	9.2483
0.0883	1.0	8000	0.2168	44.3470	9.3419

Framework versions

• Transformers 4.48.0.dev0
• Pytorch 2.5.1+cu121
• Datasets 3.6.0
• Tokenizers 0.21.0

Citation

Please cite the model using the following BibTeX entry:

@misc{deepdml/whisper-base-ta-mix-norm,
      title={Fine-tuned Whisper base ASR model for speech recognition in Tamil},
      author={Jimenez, David},
      howpublished={\url{https://huggingface.co/deepdml/whisper-base-ta-mix-norm}},
      year={2026}
    }