Fake Voice Verification in the Age of Generative AI

Kunyang Huang

 Co-Presenters: Individual Presentation
College: Dorothy and George Hennings College of Science, Mathematics and Technology
Major: Computer Science
Faculty Research Mentor: Bin Hu
 

Abstract:

With the rapid advancement of artificial intelligence (AI) technology, while AI-generated audio and video bring convenience, they also offer new avenues for malicious acts such as fraud, giving rise to severe security concerns. This study centers on the identification of AI-generated products, with the aim of reducing the success rate of fraud committed using AI-generated audio. In the course of the research, we selected the AST model from numerous models for replication and employed this model to detect AI-generated voices in prevalent short-video platforms, such as Lei Jun's AI voice, among others. The experimental outcomes indicate that the model has achieved satisfactory results in the detection of AI-generated voices. In future work, we propose to develop an AI-powered speech recognition system based on this model, designed to discriminate between human speech and AI-generated synthetic speech. This research not only establishes an effective technical safeguard against fraudulent risks posed by AI-generated content but also contributes significantly to upholding cybersecurity and social stability in digital ecosystems.