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: 0.0288
• Wer: 15.7981
• Cer: 4.8926

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
0.0015	1.0	157	0.0176	17.0508	5.4660
0.0007	2.0	314	0.0205	17.9173	5.9564
0.0007	3.0	471	0.0241	20.2647	6.7086
0.0009	4.0	628	0.0254	20.3409	6.5112
0.0007	5.0	785	0.0263	20.7123	6.4479
0.0006	6.0	942	0.0279	19.9695	6.0607
0.0004	7.0	1099	0.0269	20.1219	6.6016
0.0003	8.0	1256	0.0280	18.9982	6.0934
0.0003	9.0	1413	0.0294	20.1505	6.0196
0.0001	10.0	1570	0.0286	19.0791	6.0391
0.0	11.0	1727	0.0308	18.3935	5.9303
0.0	12.0	1884	0.0302	17.2793	5.5675
0.0	13.0	2041	0.0308	16.5556	5.3800
0.0	14.0	2198	0.0311	16.3984	5.2890
0.0	14.9088	2340	0.0312	16.4365	5.3134

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0