Who Are You - I Really Wanna Know - Detecting Audio DeepFakes Through Vocal Tract Reconstruction

Explore a groundbreaking approach to detecting audio deepfakes through vocal tract reconstruction in this 14-minute conference talk from USENIX Security '22. Delve into the innovative research by a team from the University of Florida that applies fluid dynamics to estimate human vocal tract arrangements during speech generation. Learn how this method can identify impossible or highly unlikely anatomical configurations in deepfake audio, achieving an impressive 99.5% recall rate at 99.9% precision. Discover the limitations of current deepfake models in reproducing all aspects of speech and understand how biologically constrained elements of human speech generation can serve as powerful tools for detecting audio deepfakes. Gain insights into the challenges of deepfake detection and the potential implications for voice cloning technologies in both beneficial and malicious applications.