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: 0.0061
• Wer: 0.0763
• Cer: 0.0310

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
0.0046	1.0	701	0.0052	0.0782	0.0297
0.0035	2.0	1402	0.0049	0.0717	0.0281
0.0036	3.0	2103	0.0052	0.0719	0.0290
0.0026	4.0	2804	0.0055	0.0671	0.0267
0.0012	5.0	3505	0.0058	0.0699	0.0275
0.0017	6.0	4206	0.0062	0.0691	0.0283
0.0012	7.0	4907	0.0067	0.0710	0.0285
0.0007	8.0	5608	0.0071	0.0681	0.0273
0.0005	9.0	6309	0.0075	0.0704	0.0287
0.0005	10.0	7010	0.0077	0.0695	0.0278
0.0003	11.0	7711	0.0079	0.0693	0.0270
0.0001	12.0	8412	0.0080	0.0728	0.0285
0.0002	13.0	9113	0.0081	0.0738	0.0289
0.0002	14.0	9814	0.0093	0.0770	0.0318
0.0001	14.9793	10500	0.0083	0.0717	0.0284

Framework versions

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