Doing More with Less: Data Augmentation for Sudanese Dialect
Collection
22 items • Updated
openai/whisper-medium
This model is a fine-tuned version of openai/whisper-medium on the None dataset. It achieves the following results on the evaluation set:
- Loss: 1.2201
- Wer: 44.6966
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: 32
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 5000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0566 | 6.02 | 1000 | 0.9354 | 47.1998 |
| 0.0025 | 13.01 | 2000 | 1.0806 | 47.5605 |
| 0.0012 | 19.03 | 3000 | 1.1642 | 47.6665 |
| 0.0002 | 26.01 | 4000 | 1.1866 | 44.9724 |
| 0.0001 | 33.0 | 5000 | 1.2201 | 44.6966 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.7.1.dev0
- Tokenizers 0.13.2
- Downloads last month
- 8