ast_celeb_spoof / README.md
Kabatubare's picture
Upload
49d81d5 verified
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