deepdml/whisper-large-v3-turbo-ar-quran-mix-norm

Whisper Turbo ar-quran

This model is a fine-tuned version of deepdml/whisper-large-v3-turbo on the Quran dataset. It achieves the following results on the evaluation set:

• Loss: 0.0029
• Wer: 0.2401
• Cer: 0.0690

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: 16
• eval_batch_size: 16
• 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: 22000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0768	0.0455	1000	0.0944	7.1584	2.0729
0.0383	0.0909	2000	0.0788	6.2310	1.8630
0.0285	0.1364	3000	0.0496	4.1718	1.1321
0.0263	0.1818	4000	0.0412	3.3369	1.0024
0.0295	0.2273	5000	0.0335	2.8189	0.8138
0.0326	0.2727	6000	0.0324	2.7451	0.8045
0.0154	0.3182	7000	0.0327	2.6813	0.7991
0.0078	0.3636	8000	0.0268	2.2577	0.6867
0.0126	0.4091	9000	0.0171	1.6026	0.5705
0.0064	0.4545	10000	0.0170	1.3016	0.3734
0.0085	0.5	11000	0.0161	1.3246	0.3916
0.0374	0.5455	12000	0.0100	0.8497	0.2447
0.0032	0.5909	13000	0.0115	0.9043	0.2639
0.0063	0.6364	14000	0.0112	0.9316	0.3412
0.0112	0.6818	15000	0.0086	0.6987	0.1931
0.0021	0.7273	16000	0.0070	0.5593	0.1611
0.0019	0.7727	17000	0.0063	0.5051	0.1426
0.0006	0.8182	18000	0.0057	0.4519	0.1326
0.0004	0.8636	19000	0.0051	0.4313	0.1444
0.0012	0.9091	20000	0.0038	0.3149	0.0864
0.0025	0.9545	21000	0.0033	0.2693	0.0728
0.0007	1.0	22000	0.0029	0.2401	0.0690

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-large-v3-turbo-ar-quran-mix-norm,
      title={Fine-tuned Whisper turbo ASR model for speech recognition in Arabic},
      author={Jimenez, David},
      howpublished={\url{https://huggingface.co/deepdml/whisper-large-v3-turbo-ar-quran-mix-norm}},
      year={2026}
    }