openai/whisper-small
This model is a fine-tuned version of openai/whisper-small on the Hanhpt23/GermanMed-full dataset. It achieves the following results on the evaluation set:
- Loss: 0.6502
- Wer: 22.8325
- Cer: 15.2331
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: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 100
- num_epochs: 20
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.4898 | 1.0 | 194 | 0.4951 | 31.3586 | 21.1891 |
| 0.2315 | 2.0 | 388 | 0.4999 | 33.0968 | 22.6097 |
| 0.1009 | 3.0 | 582 | 0.5097 | 29.4456 | 19.8430 |
| 0.0558 | 4.0 | 776 | 0.5487 | 27.8412 | 19.0253 |
| 0.0379 | 5.0 | 970 | 0.5783 | 28.7566 | 19.9297 |
| 0.0232 | 6.0 | 1164 | 0.5849 | 24.9203 | 16.3505 |
| 0.0136 | 7.0 | 1358 | 0.5879 | 27.7486 | 19.3562 |
| 0.0143 | 8.0 | 1552 | 0.6237 | 25.2700 | 17.6567 |
| 0.0036 | 9.0 | 1746 | 0.6323 | 26.4939 | 18.1002 |
| 0.0035 | 10.0 | 1940 | 0.6338 | 26.5762 | 18.5022 |
| 0.0045 | 11.0 | 2134 | 0.6337 | 23.3364 | 15.9694 |
| 0.0041 | 12.0 | 2328 | 0.6363 | 23.6655 | 16.0144 |
| 0.0004 | 13.0 | 2522 | 0.6397 | 22.1125 | 14.6475 |
| 0.0002 | 14.0 | 2716 | 0.6415 | 22.7502 | 15.1846 |
| 0.0002 | 15.0 | 2910 | 0.6443 | 22.7193 | 15.1742 |
| 0.0002 | 16.0 | 3104 | 0.6461 | 22.7296 | 15.1915 |
| 0.0001 | 17.0 | 3298 | 0.6477 | 22.6885 | 15.1777 |
| 0.0002 | 18.0 | 3492 | 0.6490 | 22.6782 | 15.1586 |
| 0.0001 | 19.0 | 3686 | 0.6499 | 22.7296 | 15.1673 |
| 0.0001 | 20.0 | 3880 | 0.6502 | 22.8325 | 15.2331 |
Framework versions
- Transformers 4.41.1
- Pytorch 2.3.0
- Datasets 2.19.1
- Tokenizers 0.19.1