Can I Hear Your Face? Pervasive Attack on Voice Authentication Systems with a Single Face Image

Learn about a groundbreaking security research presentation from USENIX Security '24 that demonstrates Foice, an innovative deepfake attack capable of generating synthetic voices from single facial images. Explore how researchers from the National University of Singapore and KAIST developed a system that exploits the correlation between facial and voice features to bypass voice authentication systems without requiring voice samples. Discover the implications of this 12-minute talk that reveals vulnerabilities in major commercial voice authentication platforms, including WeChat's Voiceprint and Microsoft Azure, through extensive testing with both offline participants and a large online dataset of over 1,000 individuals. Understand how this attack method could potentially enable large-scale security breaches by leveraging the widespread availability of facial images.