Home Technology AI software fools voice authentication systems almost flawlessly

AI software fools voice authentication systems almost flawlessly

AI software fools voice authentication systems almost flawlessly

Creating a fake voice is not a problem for artificial intelligence
(photo: CC0 Public Domain)

UUniversity of Waterloo scientists created deep voice spoofing software that can fool the voice authentication system 99% of the time. All it takes to use this machine learning enabled voice cloning software is five minutes of human voice audio recording.

Lead study author Andre Cassis, a PhD in computer security and privacy from the University of Waterloo, explains to PC Mag how voice authentication works:

“When you sign up using voice authentication, you’re asked to repeat a certain phrase with your voice. The system then extracts a unique voice signature (voiceprint) from this phrase and stores it on the server. On subsequent authentication attempts, you will be asked to say a different phrase and the features extracted from it are compared to the stored voiceprint to decide whether the system should grant you access”.

But even the anti-spoofing measures used by voice authentication systems do not detect AI fraud. The program by scientists at the University of Waterloo removes markers from deeply spoofed audio that “indicate that it was generated by a computer.”

After making multiple attempts to bypass the authentication system, the researchers claim to have succeeded 99% of the time.

Creating a fake voice to fool a voice authentication system is no surprise, but the new AI software is proving to be extremely effective. Computer science professor Urs Hengartner believes that companies relying on voice as their sole authentication factor will “consider implementing additional or stronger authentication measures.”


Please enter your comment!
Please enter your name here