Whisper Small Slovak - Robust
This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 sk dataset. It achieves the following results on the evaluation set:
- Loss: 0.7397
- Wer: 43.6221
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-05
- train_batch_size: 64
- eval_batch_size: 32
- 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: 5000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0232 | 14.29 | 1000 | 0.7425 | 51.8801 |
| 0.0083 | 28.57 | 2000 | 0.7698 | 48.4888 |
| 0.0006 | 42.86 | 3000 | 0.7640 | 47.5964 |
| 0.0005 | 57.14 | 4000 | 0.7649 | 44.8953 |
| 0.0002 | 71.43 | 5000 | 0.7440 | 44.3598 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.8.0
- Tokenizers 0.13.2