whisper-small-DL-Twi
This model is a fine-tuned version of openai/whisper-tiny on the Twi- native Ghanaian language. dataset. It achieves the following results on the evaluation set:
- Loss: 0.0056
- Wer: 21.9877
- Cer: 19.3600
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: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 0.0228 | 0.5701 | 1000 | 0.0239 | 31.6736 | 25.7396 |
| 0.0055 | 1.1403 | 2000 | 0.0089 | 19.4690 | 14.9189 |
| 0.0028 | 1.7104 | 3000 | 0.0061 | 27.2975 | 22.3501 |
| 0.0034 | 2.2805 | 4000 | 0.0056 | 21.9877 | 19.3600 |
Framework versions
- Transformers 4.48.0.dev0
- Pytorch 2.5.1+cu121
- Datasets 3.2.0
- Tokenizers 0.21.0