Baselhany/Graduation_Project_Whisper_tiny_fine_tune_Quran

Whisper tiny AR - BH

This model is a fine-tuned version of openai/whisper-tiny on the quran-ayat-speech-to-text dataset. It achieves the following results on the evaluation set:

• Loss: 0.0344
• Wer: 16.1004
• Cer: 5.1378

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0023	1.0	157	0.0291	18.4363	5.9503
0.0007	2.0	314	0.0258	19.4172	6.2648
0.0006	3.0	471	0.0290	19.4172	6.4596
0.0007	4.0	628	0.0278	20.3124	6.5744
0.0007	5.0	785	0.0307	21.0409	7.0886
0.0005	6.0	942	0.0311	20.6647	6.3780
0.0004	7.0	1099	0.0321	21.0028	6.7774
0.0003	8.0	1256	0.0347	19.5172	6.0479
0.0002	9.0	1413	0.0356	20.1647	6.2282
0.0001	10.0	1570	0.0358	18.5078	5.7090
0.0	11.0	1727	0.0370	18.4649	5.8249
0.0	12.0	1884	0.0384	17.8316	5.5625
0.0	13.0	2041	0.0384	17.1984	5.4460
0.0	14.0	2198	0.0384	16.7270	5.3628
0.0	14.9088	2340	0.0385	16.6841	5.3334

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0