Whisper Small ar-quran

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

Loss: 0.0003
Wer: 0.0288

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.04
training_steps: 20905

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0499	0.0478	1000	0.0595	4.9726
0.0072	0.0957	2000	0.0255	2.2505
0.0518	0.1435	3000	0.0147	1.3491
0.0045	0.1913	4000	0.0107	0.9546
0.0035	0.2392	5000	0.0084	0.7682
0.001	0.2870	6000	0.0061	0.5717
0.0036	0.3348	7000	0.0052	0.4668
0.0035	0.3827	8000	0.0040	0.3609
0.0007	0.4305	9000	0.0036	0.3273
0.0009	1.0071	10000	0.0028	0.2727
0.0007	1.0550	11000	0.0023	0.2305
0.0005	1.1028	12000	0.0019	0.1883
0.0002	1.1506	13000	0.0018	0.1629
0.0004	1.1985	14000	0.0012	0.1217
0.0004	1.2463	15000	0.0011	0.1289
0.0002	1.2941	16000	0.0009	0.0935
0.0002	1.3420	17000	0.0007	0.0666
0.0002	1.3898	18000	0.0005	0.0537
0.0003	1.4376	19000	0.0004	0.0417
0.0001	2.0148	20000	0.0003	0.0288

Framework versions

Transformers 4.42.0.dev0
Pytorch 2.3.0+cu121
Datasets 2.19.1
Tokenizers 0.19.1

Citation

Please cite the model using the following BibTeX entry:

@misc{deepdml/whisper-small-ar-quran-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-quran-mix-norm}},
      year={2025}
    }