Whisper Small Khmer - Seanghay Yath
This model is a fine-tuned version of openai/whisper-small on the Google FLEURS & OpenSLR dataset. It achieves the following results on the evaluation set:
- Loss: 0.4484
- Wer: 1.0704
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 6.25e-06
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 800
- training_steps: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.2052 | 3.33 | 1000 | 0.3582 | 1.0233 |
| 0.0465 | 6.67 | 2000 | 0.3129 | 1.0105 |
| 0.0089 | 10.0 | 3000 | 0.3977 | 1.0214 |
| 0.0016 | 13.33 | 4000 | 0.4484 | 1.0704 |
Framework versions
- Transformers 4.28.0.dev0
- Pytorch 1.12.1
- Datasets 2.11.1.dev0
- Tokenizers 0.13.3