Whisper Tiny ar-quran
This model is a fine-tuned version of openai/whisper-tiny on the Quran dataset. It achieves the following results on the evaluation set:
- Loss: 0.0344
- Wer Raw: 3.5416
- Cer Raw: 1.0757
- Wer: 3.3920
- Cer: 1.0872
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: 20000
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer Raw | Cer Raw | Wer | Cer |
|---|---|---|---|---|---|---|---|
| 0.6209 | 0.05 | 1000 | 0.3911 | 36.4009 | 12.3406 | 34.2952 | 11.8662 |
| 0.3487 | 0.1 | 2000 | 0.2256 | 22.4230 | 7.1651 | 21.1641 | 6.9660 |
| 0.1141 | 0.15 | 3000 | 0.1559 | 15.5910 | 4.9591 | 14.7442 | 4.8461 |
| 0.2314 | 0.2 | 4000 | 0.1305 | 14.3373 | 5.2439 | 13.6194 | 5.1185 |
| 0.125 | 0.25 | 5000 | 0.0992 | 10.4622 | 3.3425 | 10.0046 | 3.3191 |
| 0.1765 | 0.3 | 6000 | 0.0848 | 8.8922 | 2.7646 | 8.4628 | 2.7475 |
| 0.3065 | 0.35 | 7000 | 0.0801 | 8.7327 | 2.8984 | 8.2793 | 2.8643 |
| 0.044 | 0.4 | 8000 | 0.0716 | 7.2403 | 2.1966 | 6.8857 | 2.1833 |
| 0.5188 | 0.45 | 9000 | 0.0646 | 6.8962 | 2.1783 | 6.5449 | 2.1737 |
| 0.5267 | 0.5 | 10000 | 0.0622 | 6.9441 | 2.2432 | 6.6365 | 2.2551 |
| 0.1795 | 0.55 | 11000 | 0.0528 | 5.5275 | 1.7637 | 5.2663 | 1.7581 |
| 0.222 | 0.6 | 12000 | 0.0475 | 5.0622 | 1.6295 | 4.8446 | 1.6363 |
| 0.487 | 0.65 | 13000 | 0.0468 | 5.0828 | 1.7026 | 4.8552 | 1.6783 |
| 0.2944 | 0.7 | 14000 | 0.0446 | 4.6040 | 1.4377 | 4.4004 | 1.4395 |
| 0.3001 | 0.75 | 15000 | 0.0450 | 4.7722 | 1.5126 | 4.5609 | 1.5161 |
| 0.1493 | 0.8 | 16000 | 0.0400 | 4.0836 | 1.2418 | 3.9111 | 1.2501 |
| 0.2001 | 0.85 | 17000 | 0.0405 | 4.3107 | 1.4219 | 4.1253 | 1.4433 |
| 0.2617 | 0.9 | 18000 | 0.0371 | 3.7888 | 1.1473 | 3.6321 | 1.1582 |
| 0.1278 | 0.95 | 19000 | 0.0351 | 3.5938 | 1.1006 | 3.4467 | 1.1157 |
| 0.554 | 1.0 | 20000 | 0.0344 | 3.5416 | 1.0757 | 3.3920 | 1.0872 |
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-tiny-ar-quran-mix-norm,
title={Fine-tuned Whisper tiny ASR model for speech recognition in Arabic},
author={Jimenez, David},
howpublished={\url{https://huggingface.co/deepdml/whisper-tiny-ar-quran-mix-norm}},
year={2026}
}