deepdml/whisper-small-ar-mix-norm

Whisper Small ar

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

• Loss: 0.2849
• Wer: 26.1068
• Cer: 7.6373

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: 32
• eval_batch_size: 32
• 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: 18000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.5538	0.0556	1000	0.4236	37.3864	11.5198
0.2862	0.1111	2000	0.3772	33.8003	10.0942
0.1789	0.1667	3000	0.3545	32.8400	9.7685
0.1261	0.2222	4000	0.3370	30.9396	9.3539
0.0889	0.2778	5000	0.3228	30.5852	9.0891
0.0584	0.3333	6000	0.3132	29.2521	8.5476
0.05	0.3889	7000	0.3066	29.9958	8.7841
0.051	0.4444	8000	0.3034	28.6260	8.3920
0.0643	0.5	9000	0.2916	28.6297	8.3994
0.0466	0.5556	10000	0.2952	28.3671	8.3171
0.0609	0.6111	11000	0.2897	27.7722	8.2164
0.0719	0.6667	12000	0.2822	26.9826	7.9877
0.0508	0.7222	13000	0.2883	26.7164	7.9671
0.032	0.7778	14000	0.2824	26.1123	7.6930
0.031	0.8333	15000	0.2810	27.0524	7.9896
0.0336	0.8889	16000	0.2805	26.2206	7.7793
0.04	0.9444	17000	0.2808	26.7145	7.9394
0.0371	1.0	18000	0.2849	26.1068	7.6373

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