stt-april-1

This model is a fine-tuned version of openai/whisper-small on the Common Voice 17 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7422
• Wer Ortho: 32.1876
• Wer: 26.1493

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.6457	0.6180	500	0.6308	38.7728	31.6649
0.3566	1.2361	1000	0.5699	36.3837	29.0941
0.3687	1.8541	1500	0.5554	34.7230	27.5095
0.2412	2.4722	2000	0.5469	31.7690	25.0986
0.1654	3.0902	2500	0.6037	32.8070	26.1459
0.172	3.7083	3000	0.5907	32.5790	25.2142
0.1214	4.3263	3500	0.6065	31.8337	24.9354
0.1488	4.9444	4000	0.6024	31.5376	24.9966
0.1034	5.5624	4500	0.6288	32.1161	25.4931
0.0818	6.1805	5000	0.6470	31.9051	25.5373
0.0955	6.7985	5500	0.6566	32.1195	25.4795
0.0744	7.4166	6000	0.6748	31.9153	25.6257
0.0784	8.0346	6500	0.6908	32.3577	25.9759
0.0684	8.6527	7000	0.6959	32.6538	26.3772
0.0587	9.2707	7500	0.7318	32.2931	25.7991
0.0656	9.8888	8000	0.7182	32.1467	25.8535
0.0589	10.5068	8500	0.7361	32.4428	26.3500
0.049	11.1248	9000	0.7597	32.3884	25.9827
0.0562	11.7429	9500	0.7504	32.2318	25.7481
0.0503	12.3609	10000	0.7422	32.1876	26.1493

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0