UPDATE
Use an updated fine tunned version Sagicc/whisper-large-v3-sr-cmb with new 50+ hours of dataset.
Whisper Large v3 Sr
This model is a fine-tuned version of openai/whisper-large-v3 on Serbian Mozilla/Common Voice 13 and Google/Fleurs datasets. It achieves the following results on the evaluation set:
- Loss: 0.1628
- Wer Ortho: 0.1635
- Wer: 0.0556
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 2
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 8
- 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: 50
- training_steps: 1500
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer Ortho | Wer |
|---|---|---|---|---|---|
| 0.0567 | 1.34 | 500 | 0.1512 | 0.1676 | 0.0717 |
| 0.0256 | 2.67 | 1000 | 0.1482 | 0.1585 | 0.0610 |
| 0.0114 | 4.01 | 1500 | 0.1628 | 0.1635 | 0.0556 |
Framework versions
- Transformers 4.35.0
- Pytorch 2.0.1+cu117
- Datasets 2.14.5
- Tokenizers 0.14.1