Rohan Aware, Prajakta Sonawane, Shruti Walunj, Rajveer Patil, Sushma Gunjal

The rapid progress in artificial intelligence has given rise to advanced deepfake technologies, particularly in the audio space, where synthetic voices can imitate real people convincingly. Although this feature has creative potential, it also poses serious security threats, including misinformation, identity theft, and impersonation fraud. With the growing prevalence of deepfake audio, particularly on social media and onlineinteractions, guaranteeing the integrity and authenticity of audio content is now imperative. Through this +project, these issues are to be alleviated with the creation of a reliable system that can identify deepfake audio and maintain trust in audio interactions. Our method uses machine learning algorithms to scrutinize different acoustic and linguistic characteristics using methods such as Convolutional Neural Networks (CNNs)and transformer models to identify real audio from synthetic material. By targeting minute disparities usually inherent in deepfake audio, our detection mechanism is conceived to run in real-time with the ability to be adaptable in the face of different synthesis methods and platforms. Large-scale testing on a variety of real and deepfake audio datasets will guarantee high detection accuracy and robustness against novel deepfaketechniques. The end objective of this project is to offer an effective countermeasure against audio manipulation, improving security in media, legal, and secure communication contexts.