"Deepfake Audio Has a Tell – Researchers Use Fluid Dynamics to Spot Artificial Imposter Voices"
An audio deepfake detection technique developed by researchers at the University of Florida measures the acoustic and fluid dynamic differences between voice samples generated organically by human speakers and those generated synthetically by computers. Humans vocalize by forcing air through the vocal tract's various structures, which include vocal folds, the tongue, and lips. By rearranging these structures, it is possible to change the acoustical properties of the vocal tract and produce more than 200 distinct sounds or phonemes. However, the acoustic behavior of these different phonemes is fundamentally limited by human anatomy, resulting in a relatively small range of correct sounds for each. Audio deepfakes differ in that they are created by first allowing a computer to listen to audio recordings of a specific victim speaker. Depending on the techniques used, the computer may only need to hear 10 to 20 seconds of audio. This audio is used to extract information about the victim's voice. The attacker chooses a phrase for the deepfake to speak and then generates an audio sample that sounds like the victim saying the selected phrase using a modified text-to-speech algorithm. This process of creating a single deepfake audio sample takes only a few seconds, potentially giving attackers enough flexibility to use the deepfake voice in a conversation. Deepfakes, both audio and video, have become possible due to the advancement of sophisticated Machine Learning (ML) technologies. Such fraudulent media have added a new layer of uncertainty to the digital media landscape. Audio deepfakes may pose an even greater threat because people frequently communicate verbally without using video, such as through phone calls, radio broadcasts, and voice recordings. These voice-only communications greatly expand attackers' ability to use deepfakes. This article continues to discuss the potential impact of deepfakes and the method developed by researchers at the University of Florida to detect audio deepfakes.