Deepfake Audio Detection

Overview

This repository provides tools and models for detecting deepfake audio using machine learning techniques. The project focuses on enhancing detection robustness through explainability and feature engineering.

Contents

• custom_data.py: Script to load and preprocess custom datasets.
• features.py: Feature extraction methods for audio analysis.
• transformer.py: Implementation of a transformer-based model for classification.
• xgboost.py: XGBoost model for baseline comparison.
• Jupyter Notebooks: Analysis, feature importance, and explainability.

Usage

1. Preprocess the data using custom_data.py.
2. Extract features using features.py.
3. Train and evaluate models using transformer.py and xgboost.py.
4. Analyze results and model explainability using the provided notebooks.
5. Explainability tooling found in .ipynb files.

Contributing

Contributions are welcome! Please submit issues or pull requests for improvements.

License

MIT License