Hungarians/whisper-tiny-cv16-hu

Whisper Tiny Hungarian

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 16 dataset of Mozilla Foundation. It achieves the following results on the evaluation set:

• Loss: 0.3628
• Wer Ortho: 34.7985
• Wer: 32.2247

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: 3.75e-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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.7288	0.17	500	0.7093	59.8298	57.4443
0.5483	0.33	1000	0.5648	52.3541	49.3122
0.4647	0.5	1500	0.4912	46.1159	42.9533
0.3925	0.67	2000	0.4463	42.8674	39.9838
0.3682	0.84	2500	0.4258	41.1739	38.0487
0.3219	1.0	3000	0.3932	37.5828	34.7286
0.2638	1.17	3500	0.3909	37.8060	35.0311
0.2507	1.34	4000	0.3881	36.7856	34.1199
0.2483	1.51	4500	0.3737	35.5778	32.9881
0.2444	1.67	5000	0.3628	34.7985	32.2247

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.0
• Tokenizers 0.15.0