Whisper Small Portuguese
This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 pt dataset. It achieves the following results on the evaluation set:
- Loss: 0.3056
- Wer: 14.6841
- Cer: 5.8856
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: 5e-06
- train_batch_size: 32
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- training_steps: 10000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.2817 | 0.92 | 500 | 0.3352 | 15.9476 | 6.3609 |
| 0.2245 | 1.84 | 1000 | 0.3047 | 15.0231 | 5.9326 |
| 0.1587 | 2.76 | 1500 | 0.2985 | 15.0847 | 5.9326 |
| 0.1181 | 3.68 | 2000 | 0.3056 | 14.6841 | 5.8856 |
| 0.0741 | 4.6 | 2500 | 0.3162 | 14.9923 | 5.9906 |
| 0.0438 | 5.52 | 3000 | 0.3466 | 15.4700 | 6.2255 |
| 0.0294 | 6.45 | 3500 | 0.3799 | 15.2234 | 6.1647 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.12.1+cu116
- Datasets 2.7.1.dev0
- Tokenizers 0.13.2