WavLM model ensemble for audio deepfake detection
Paper
•
2408.07414
•
Published
wavLM-base-UTCN
This model is a fine-tuned version of microsoft/wavLM-base on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.0234
- Accuracy: 0.9962
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: 3e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 0.0603 | 1.0 | 435 | 0.0307 | 0.9928 |
| 0.009 | 2.0 | 870 | 0.0201 | 0.9962 |
| 0.0109 | 3.0 | 1305 | 0.0176 | 0.9972 |
| 0.001 | 4.0 | 1740 | 0.0246 | 0.9950 |
| 0.0013 | 5.0 | 2175 | 0.0234 | 0.9962 |
Framework versions
- Transformers 4.30.0
- Pytorch 2.3.1+cu121
- Datasets 2.20.0
- Tokenizers 0.13.3
arxiv.org/abs/2408.07414
- Downloads last month
- 161