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: 0.2568
- Wer: 11.6487
- Cer: 4.4764
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 |
|---|---|---|---|---|---|
| 0.2476 | 0.92 | 500 | 0.2900 | 13.2049 | 4.9765 |
| 0.1886 | 1.84 | 1000 | 0.2611 | 12.2804 | 4.6173 |
| 0.1066 | 2.76 | 1500 | 0.2568 | 11.6487 | 4.4764 |
| 0.0698 | 3.68 | 2000 | 0.2701 | 11.8798 | 4.6145 |
| 0.0403 | 4.6 | 2500 | 0.2831 | 11.8644 | 4.4405 |
| 0.019 | 5.52 | 3000 | 0.3148 | 11.7565 | 4.4653 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.12.1+cu116
- Datasets 2.7.1.dev0
- Tokenizers 0.13.2