whisper-large-v3-bn-cv17
This model is a fine-tuned version of openai/whisper-large-v3 on the common_voice_17_0 dataset. And IIT Madras SPRING-INX Data. It achieves the following results on the evaluation set:
- Loss: 0.0669
- Wer: 30.8364
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: constant_with_warmup
- lr_scheduler_warmup_steps: 500
- training_steps: 5000
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.1229 | 0.2618 | 500 | 0.1370 | 109.1423 |
| 0.097 | 0.5236 | 1000 | 0.1005 | 41.9964 |
| 0.0853 | 0.7853 | 1500 | 0.0852 | 32.2293 |
| 0.0472 | 1.0471 | 2000 | 0.0786 | 56.4450 |
| 0.0503 | 1.3089 | 2500 | 0.0746 | 87.3848 |
| 0.0479 | 1.5707 | 3000 | 0.0716 | 30.8487 |
| 0.0511 | 1.8325 | 3500 | 0.0665 | 32.1765 |
| 0.0294 | 2.0942 | 4000 | 0.0710 | 48.9811 |
| 0.03 | 2.3560 | 4500 | 0.0695 | 52.9004 |
| 0.0333 | 2.6178 | 5000 | 0.0669 | 30.8364 |
Framework versions
- Transformers 4.46.3
- Pytorch 2.4.0
- Datasets 3.1.0
- Tokenizers 0.20.3