Whisper Small Japanese Elite
This model is a fine-tuned version of openai/whisper-small on the Elite35P-Server/EliteVoiceProject youtube dataset. It achieves the following results on the evaluation set:
- Loss: 1.1596
- Wer: 31.5364
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: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant_with_warmup
- lr_scheduler_warmup_steps: 100
- training_steps: 10000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0003 | 52.0 | 1000 | 0.8053 | 28.8410 |
| 0.0 | 105.0 | 2000 | 0.8636 | 28.5714 |
| 0.0 | 157.0 | 3000 | 0.9056 | 28.0323 |
| 0.0 | 210.0 | 4000 | 0.9414 | 28.8410 |
| 0.0 | 263.0 | 5000 | 0.9842 | 31.2668 |
| 0.0 | 315.0 | 6000 | 1.0223 | 31.2668 |
| 0.0 | 368.0 | 7000 | 1.0677 | 31.2668 |
| 0.0 | 421.0 | 8000 | 1.1079 | 31.2668 |
| 0.0 | 473.0 | 9000 | 1.1468 | 31.5364 |
| 0.0 | 526.0 | 10000 | 1.1596 | 31.5364 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.8.1.dev0
- Tokenizers 0.13.2
- Downloads last month
- 5
Dataset used to train noflm/whisper-small-ja-elite
Evaluation results
- Wer on Elite35P-Server/EliteVoiceProject youtubetest set self-reported31.536