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: 0.0408
• Wer: 0.3145
• Cer: 0.1128

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: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.1136	1.0	94	0.0777	5.3133	1.8561
0.0458	2.0	188	0.0478	2.6311	1.1803
0.0292	3.0	282	0.0402	1.8776	0.9786
0.0239	4.0	376	0.0369	0.9783	0.4714
0.0182	5.0	470	0.0368	0.6109	0.2680
0.0163	6.0	564	0.0357	0.6387	0.2671
0.0132	7.0	658	0.0363	0.3701	0.1281
0.01	8.0	752	0.0369	0.5068	0.1884
0.0067	9.0	846	0.0392	0.3609	0.1184
0.0056	10.0	940	0.0354	0.3155	0.1000
0.0041	11.0	1034	0.0382	0.3131	0.0964
0.0027	12.0	1128	0.0363	0.3419	0.1087
0.0018	13.0	1222	0.0387	0.3640	0.1111
0.0015	14.0	1316	0.0388	0.3428	0.1179
0.0012	15.0	1410	0.0382	0.3730	0.1359
0.0012	16.0	1504	0.0386	0.2863	0.0933
0.0007	17.0	1598	0.0387	0.3721	0.1560
0.0005	18.0	1692	0.0391	0.2780	0.0847
0.0004	19.0	1786	0.0405	0.2985	0.1114
0.0003	20.0	1880	0.0409	0.3452	0.1222
0.0002	21.0	1974	0.0405	0.3475	0.1441
0.0	22.0	2068	0.0413	0.2614	0.0754
0.0	23.0	2162	0.0423	0.3339	0.1156
0.0	24.0	2256	0.0426	0.2676	0.0854
0.0	25.0	2350	0.0442	0.3357	0.1101

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.2.0
• Tokenizers 0.19.1