speech-accent-detection
This model is a fine-tuned version of facebook/wav2vec2-base on the VCTK dataset. It achieves the following results on the evaluation set:
- Loss: 0.0441
- Accuracy: 0.9955
Model description
I used the wav2vec2 model's weights and fine-tune over my dataset.
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: 3e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Use OptimizerNames.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_ratio: 0.1
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 0.8005 | 1.0 | 2205 | 0.6526 | 0.8270 |
| 0.0508 | 2.0 | 4410 | 0.3466 | 0.9374 |
| 0.3054 | 3.0 | 6615 | 0.2946 | 0.9524 |
| 0.0882 | 4.0 | 8820 | 0.1832 | 0.9737 |
| 0.0006 | 5.0 | 11025 | 0.1539 | 0.9757 |
| 0.0003 | 6.0 | 13230 | 0.0677 | 0.9896 |
| 0.3011 | 7.0 | 15435 | 0.1219 | 0.9859 |
| 0.0001 | 8.0 | 17640 | 0.0695 | 0.9916 |
| 0.0001 | 9.0 | 19845 | 0.0397 | 0.9955 |
| 0.0 | 10.0 | 22050 | 0.0441 | 0.9955 |
Framework versions
- Transformers 4.48.0
- Pytorch 2.5.1+cu124
- Datasets 3.2.0
- Tokenizers 0.21.0