This project aims at developing a robust system to differentiate between real and meticulously crafted deepfake audio clips. The necessity of such a system is paramount in the era of information manipulation, where verifying the authenticity of audio content has become crucial.

The core of our project is an end-to-end pipeline designed for voice anti-spoofing. It utilizes advanced machine learning techniques to identify and differentiate fake audio clips from genuine ones.

Please replace the image path with the actual location of your image on GitHub or another hosting service after uploading.

Our dataset is composed of real and fake audio samples:

• Real Audio Dataset: Consists of 13,100 short audio clips from the LJ Speech dataset, featuring a single speaker.
• Fake Audio Dataset: WaveFake, includes 104,885 generated audio clips using various architectures such as MelGAN, Parallel WaveGAN, Multi-band MelGAN, Full-band MelGAN, HiFi-GAN, and WaveGlow.

The diversity in generated audio types from WaveFake provides a broad basis for training and evaluating our anti-spoofing models.

git clone https://github.com/Lseonghak/MIE1517-Project.git
cd MIE1517-Project
pip install -r requirements.txt