sugafree/whisper-medium-hu

Whisper Medium HU

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 13 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2699
• Wer Ortho: 17.1763
• Wer: 14.8290

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: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.0804	1.38	2000	0.1977	19.2869	16.6612
0.038	2.76	4000	0.2028	18.2211	15.7494
0.014	4.14	6000	0.2190	17.9961	15.3466
0.0107	5.51	8000	0.2328	17.3490	14.9370
0.0144	6.89	10000	0.2376	17.4153	14.9559
0.0049	8.27	12000	0.2424	16.9984	14.6953
0.0071	9.65	14000	0.2594	17.6961	15.3586
0.0037	11.03	16000	0.2546	17.2007	14.8667
0.0078	12.41	18000	0.2644	17.5757	15.1495
0.0043	13.78	20000	0.2699	17.1763	14.8290

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0
• Datasets 2.17.0
• Tokenizers 0.15.2