rahafvii/STT-EGY

Egyptian Whisper Small

This model is a fine-tuned version of openai/whisper-small on the Egyptian-ASR-MGB-3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.9939
• Wer: 46.2893

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4615	6.8966	100	0.6496	48.0162
0.072	13.7931	200	0.7459	46.3990
0.0122	20.6897	300	0.8380	45.6863
0.0054	27.5862	400	0.8981	45.0764
0.0033	34.4828	500	0.9322	45.2820
0.0025	41.3793	600	0.9555	45.4670
0.002	48.2759	700	0.9724	46.1454
0.0017	55.1724	800	0.9843	45.9467
0.0016	62.0690	900	0.9916	46.0769
0.0015	68.9655	1000	0.9939	46.2893

Framework versions

• Transformers 4.44.2
• Pytorch 2.3.0+cu118
• Datasets 3.0.0
• Tokenizers 0.19.1