Whisper Small Fine-tuned on 不懂上海话的老王
This model is a fine-tuned version of openai/whisper-small on the Shanghai Dialect Conversational Speech Corpus dataset. It achieves the following results on the evaluation set:
- Loss: 0.6211
- Cer Ortho: 33.3881
- Cer: 32.7486
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: 16
- seed: 42
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 50
- training_steps: 500
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Cer Ortho | Cer |
|---|---|---|---|---|---|
| 0.9990 | 0.6024 | 100 | 0.9789 | 53.6472 | 53.2233 |
| 0.5426 | 1.2048 | 200 | 0.7389 | 41.0817 | 40.5007 |
| 0.4840 | 1.8072 | 300 | 0.6436 | 38.6281 | 38.0243 |
| 0.2562 | 2.4096 | 400 | 0.6285 | 34.8212 | 34.2246 |
| 0.2487 | 3.0120 | 500 | 0.6211 | 33.3881 | 32.7486 |
Framework versions
- Transformers 5.16.1
- Pytorch 2.11.0+cu128
- Datasets 4.8.5
- Tokenizers 0.23.1