openai/whisper-small
This model is a fine-tuned version of openai/whisper-medium on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.2195
- Wer: 19.56
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: 2
- eval_batch_size: 1
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 10000
- mixed_precision_training: Native AMP
Training results
| Epoch | Step | Wer |
|---|---|---|
| 0.1 | 1000 | 43.61 |
| 0.2 | 2000 | 36.79 |
| 0.3 | 3000 | 33.05 |
| 0.4 | 4000 | 29.53 |
| 0.5 | 5000 | 26.01 |
| 0.6 | 6000 | 23.44 |
| 0.7 | 7000 | 22.22 |
| 0.8 | 8000 | 21.88 |
| 0.9 | 9000 | 20.53 |
| 1.0 | 10000 | 19.56 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.0+cu117
- Datasets 2.7.1.dev0
- Tokenizers 0.13.2