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