Breaking Fake Voice Detection with Speaker-Irrelative Features

Watch a technical conference presentation from HITB2024BKK that explores SiFDetectCracker, a groundbreaking black-box adversarial attack framework designed to circumvent fake voice detection systems. Learn how researchers leverage Speaker-Irrelative Features like background noise and mute segments to manipulate synthesized speech, achieving over 80% success rate in bypassing state-of-the-art detection methods. Delve into the technical analysis explaining current detectors' vulnerabilities to these attacks, presented by researchers from Lanzhou University and Zhejiang University who specialize in AI security and privacy. Gain valuable insights into the evolving challenges of voice authentication security and the ongoing arms race between synthetic voice creation and detection technologies.