metadata
license: mit
base_model: MIT/ast-finetuned-audioset-10-10-0.4593
tags:
- generated_from_trainer
metrics:
- accuracy
- f1
- precision
- recall
model-index:
- name: AST-VoxCelebSpoof-Synthetic-Voice-Detection
results: []
datasets:
- MattyB95/VoxCelebSpoof
language:
- en
AST-VoxCelebSpoof-Synthetic-Voice-Detection
This model is a fine-tuned version of MIT/ast-finetuned-audioset-10-10-0.4593 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 89136693248.0
- Accuracy: 0.9999
- F1: 0.9999
- Precision: 1.0
- Recall: 0.9998
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: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 | Precision | Recall |
|---|---|---|---|---|---|---|---|
| 2218896740319.232 | 1.0 | 29527 | 611463921664.0 | 0.9998 | 0.9998 | 0.9999 | 0.9997 |
| 522149441830.912 | 2.0 | 59054 | 284563668992.0 | 0.9997 | 0.9997 | 0.9999 | 0.9996 |
| 0.0 | 3.0 | 88581 | 89136693248.0 | 0.9999 | 0.9999 | 1.0 | 0.9998 |
Framework versions
- Transformers 4.36.2
- Pytorch 2.1.2
- Datasets 2.16.1
- Tokenizers 0.15.0