deepdml/whisper-medium-ar-quran-mix-norm

Whisper Medium ar-quran

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

• Loss: 0.0002
• Wer: 0.0225
• Cer: 0.0065

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: 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: 41811

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0314	0.0239	1000	0.0630	4.8259	1.4337
0.022	0.0478	2000	0.0381	3.2238	1.2475
0.0201	0.0718	3000	0.0232	1.8863	0.5372
0.0056	0.0957	4000	0.0181	1.4976	0.4436
0.01	0.1196	5000	0.0138	1.2360	0.4163
0.0299	0.1435	6000	0.0097	0.8300	0.2353
0.005	0.1674	7000	0.0111	0.9460	0.2609
0.0041	0.1913	8000	0.0085	0.6724	0.1903
0.0031	0.2153	9000	0.0085	0.7467	0.2519
0.0015	0.2392	10000	0.0065	0.5152	0.1718
0.0024	0.2631	11000	0.0055	0.4879	0.1368
0.0013	0.2870	12000	0.0049	0.3987	0.1141
0.0013	0.3109	13000	0.0053	0.4605	0.1257
0.0015	0.3348	14000	0.0041	0.3633	0.1081
0.0011	0.3588	15000	0.0037	0.3359	0.1115
0.0041	0.3827	16000	0.0047	0.3666	0.1026
0.0031	0.4066	17000	0.0041	0.3522	0.1118
0.0008	0.4305	18000	0.0030	0.2569	0.0788
0.0012	0.4544	19000	0.0028	0.2674	0.0811
0.0012	1.0072	20000	0.0025	0.2415	0.0753
0.0011	1.0311	21000	0.0029	0.2689	0.0795
0.001	1.0550	22000	0.0022	0.1989	0.0608
0.001	1.0789	23000	0.0017	0.1840	0.0852
0.0006	1.1028	24000	0.0017	0.1711	0.0500
0.0003	1.1267	25000	0.0013	0.1591	0.0670
0.0	1.1507	26000	0.0013	0.1212	0.0362
0.0008	1.1746	27000	0.0013	0.1716	0.0679
0.0001	1.1985	28000	0.0012	0.1730	0.0727
0.0005	1.2224	29000	0.0013	0.1054	0.0314
0.0003	1.2463	30000	0.0009	0.1021	0.0284
0.0002	1.2702	31000	0.0009	0.0925	0.0235
0.0001	1.2942	32000	0.0008	0.0863	0.0223
0.0	1.3181	33000	0.0008	0.0695	0.0193
0.0002	1.3420	34000	0.0007	0.0623	0.0159
0.0001	1.3659	35000	0.0005	0.0613	0.0195
0.0	1.3898	36000	0.0004	0.0474	0.0148
0.0003	1.4137	37000	0.0003	0.0364	0.0125
0.0002	1.4377	38000	0.0003	0.0321	0.0113
0.0	1.4616	39000	0.0002	0.0211	0.0063
0.0001	2.0148	40000	0.0002	0.0220	0.0064
0.0	2.0387	41000	0.0002	0.0225	0.0065

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