wav2vec2-base-960h-asv19-deepfake
This model is a fine-tuned version of facebook/wav2vec2-base-960h on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.0962
- Accuracy: 0.9845
- Far: 0.0090
- Frr: 0.0162
- Eer: 0.0126
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-06
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 4.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Far | Frr | Eer |
|---|---|---|---|---|---|---|---|
| 0.3383 | 0.79 | 5000 | 0.2361 | 0.8974 | 1.0 | 0.0 | 0.5 |
| 0.0829 | 1.58 | 10000 | 0.1134 | 0.9739 | 0.0122 | 0.0277 | 0.0199 |
| 0.0441 | 2.36 | 15000 | 0.0922 | 0.9841 | 0.0118 | 0.0163 | 0.0140 |
| 0.0484 | 3.15 | 20000 | 0.1215 | 0.9798 | 0.0086 | 0.0215 | 0.0151 |
| 0.0335 | 3.94 | 25000 | 0.0962 | 0.9845 | 0.0090 | 0.0162 | 0.0126 |
Framework versions
- Transformers 4.38.0.dev0
- Pytorch 2.1.2+cu121
- Datasets 2.16.2.dev0
- Tokenizers 0.15.2