Baselhany/Graduation_Project_Whisper_tiny

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: 0.0095
• Wer: 0.1037
• Cer: 0.0382

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: 5e-06
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0104	1.0	407	0.0098	0.1182	0.0449
0.0068	2.0	814	0.0088	0.1055	0.0373
0.0075	3.0	1221	0.0088	0.1008	0.0356
0.0057	4.0	1628	0.0091	0.0992	0.0345
0.0047	5.0	2035	0.0097	0.0997	0.0349
0.0038	6.0	2442	0.0103	0.0994	0.0340
0.0024	7.0	2849	0.0109	0.1033	0.0357
0.0031	8.0	3256	0.0113	0.1015	0.0351
0.0014	9.0	3663	0.0118	0.1003	0.0350
0.0018	10.0	4070	0.0123	0.1014	0.0349
0.0013	11.0	4477	0.0128	0.1122	0.0405
0.0011	12.0	4884	0.0130	0.1037	0.0379
0.0004	13.0	5291	0.0132	0.1032	0.0379
0.0019	14.0	5698	0.0141	0.1055	0.0397
0.001	14.9643	6090	0.0135	0.1017	0.0371

Framework versions

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