Baselhany/Quran_Whisper_base_fine_tune

Whisper base AR - BH

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

• Loss: 0.0151
• Wer: 17.8284
• Cer: 5.3577

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: 7
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
0.0425	0.3011	400	11.2952	0.0330	37.6901
0.0258	0.6023	800	7.9539	0.0218	25.4742
0.0197	0.9034	1200	7.0188	0.0194	22.3362
0.0083	1.2637	1600	6.1604	0.0183	20.2225
0.0066	1.5794	2000	5.9469	0.0176	19.4205
0.0065	1.8952	2400	5.8609	0.0174	19.0958
0.0161	2.1084	2800	5.8786	0.0167	18.8094
0.0157	2.4096	3200	5.6719	0.0162	18.4526
0.0157	2.7107	3600	5.6150	0.0160	18.3772
0.0063	3.1595	4000	5.4820	0.0160	18.0877
0.0052	3.4752	4400	5.4747	0.0161	18.2023
0.0049	3.7910	4800	5.5285	0.0161	17.9922
0.0154	3.9155	5200	5.5457	0.0159	18.1641
0.0142	4.2168	5600	5.4871	0.0158	17.9721
0.0143	4.5180	6000	5.5382	0.0157	18.0495
0.0155	4.8191	6400	5.4620	0.0156	17.9460
0.005	5.3710	6800	5.4875	0.0157	17.7691
0.0052	5.6868	7200	5.4416	0.0157	17.8646
0.0085	4.8653	7600	5.5002	0.0157	17.8817
0.0096	5.1216	8000	5.5124	0.0156	17.8264
0.0094	5.3776	8400	5.6387	0.0155	17.7500
0.0093	5.6336	8800	5.4429	0.0154	17.6736
0.01	5.8896	9200	5.3128	0.0153	17.2856
0.0091	6.1453	9600	5.2836	0.0153	17.2424
0.0088	6.4019	10000	0.0153	17.2434	5.2826
0.008	6.6579	10400	0.0153	17.1288	5.2957
0.007	6.9139	10800	0.0153	17.1680	5.2889

Framework versions

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