Hungarians/whisper-tiny-cv17-hu

Whisper Tiny Hu CV17

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1387
• Wer Ortho: 10.7883
• Wer: 9.9259

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: 7.5e-05
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.6432	0.3298	250	0.7031	59.9211	57.3051
0.4839	0.6596	500	0.5256	50.4185	47.3406
0.4127	0.9894	750	0.4089	40.9250	37.8002
0.2858	1.3193	1000	0.3434	34.4725	31.8885
0.2616	1.6491	1250	0.3027	32.2956	29.2203
0.2448	1.9789	1500	0.2571	28.1705	25.3780
0.1474	2.3087	1750	0.2318	25.4426	22.8343
0.1434	2.6385	2000	0.2112	22.7749	20.3647
0.14	2.9683	2250	0.1922	20.1102	17.7883
0.0768	3.2982	2500	0.1797	18.2344	15.9917
0.0764	3.6280	2750	0.1678	16.5573	14.6161
0.0737	3.9578	3000	0.1573	15.0337	13.6199
0.0375	4.2876	3250	0.1538	14.0130	12.3243
0.0357	4.6174	3500	0.1495	13.1970	11.7106
0.033	4.9472	3750	0.1435	11.9234	10.8005
0.0138	5.2770	4000	0.1422	11.8873	10.7026
0.0137	5.6069	4250	0.1398	11.4477	10.4684
0.0135	5.9367	4500	0.1381	11.1436	10.2639
0.0072	6.2665	4750	0.1388	10.7853	9.9140
0.0065	6.5963	5000	0.1387	10.7883	9.9259

Framework versions

• Transformers 4.41.1
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1