whisper-small-ar

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

• Loss: 0.1581
• Wer Ortho: 27.8630
• Wer: 5.0804

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.7097	0.0842	50	0.5331	51.0035	11.0584
0.3752	0.1684	100	0.3281	42.8571	8.3150
0.221	0.2525	150	0.2201	36.9540	6.8078
0.1948	0.3367	200	0.1996	33.5891	6.3167
0.1947	0.4209	250	0.1833	31.1098	5.8425
0.1708	0.5051	300	0.1752	29.3388	5.2667
0.1667	0.5892	350	0.1680	28.5714	5.3514
0.1606	0.6734	400	0.1631	28.8076	5.1312
0.1525	0.7576	450	0.1600	28.6305	5.2159
0.1513	0.8418	500	0.1581	27.8630	5.0804

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1