Whisper Small - Swedish
This model is a fine-tuned version of openai/whisper-small on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.4312
- Wer: 19.0503
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: 4
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 18000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0887 | 1.71 | 2000 | 0.2817 | 21.0831 |
| 0.0168 | 3.41 | 4000 | 0.3108 | 19.6338 |
| 0.0027 | 5.12 | 6000 | 0.3421 | 19.8731 |
| 0.0012 | 6.83 | 8000 | 0.3713 | 19.1229 |
| 0.0005 | 8.53 | 10000 | 0.3844 | 19.2036 |
| 0.0004 | 10.24 | 12000 | 0.3900 | 19.0369 |
| 0.0008 | 11.94 | 14000 | 0.4161 | 19.9511 |
| 0.0002 | 13.65 | 16000 | 0.4201 | 19.1283 |
| 0.0001 | 15.36 | 18000 | 0.4312 | 19.0503 |
Framework versions
- Transformers 4.25.0.dev0
- Pytorch 1.12.1
- Datasets 2.7.1
- Tokenizers 0.13.2