Can You Believe Your Ears? FTC Awards Prizes for Tech to Detect AI-Generated Deepfake Voices

Amid growing concerns about the impact of deepfakes on elections and consumer fraud, the Federal Trade Commission (FTC) has named four organizations that have developed technology that can distinguish between real human voices and voices generated by AI. Awarded.

NPR report The winners of the FTC contest are said to be showcasing different approaches to detecting AI-generated audio, also known as audio deepfakes. Created by researchers at Arizona State University, OriginStory uses sensors to detect human characteristics such as breathing, movement, and heart rate information. “We have a very unique sound generation mechanism. So by sensing it in parallel with the acoustic signal, we can confirm that the sound came from a human. ” explained Bisar Berisha, a professor in the ASU College of Health Solutions and a member of the winning team.

Another winner, DeFake, created by Ning Zhang at Washington University in St. Louis, adds data to recordings of real voices to prevent AI-generated voice clones from sounding like real humans. Inject. The technology was inspired by an earlier tool developed by researchers at the University of Chicago that made hidden changes to images to prevent them from being imitated by AI algorithms.

Developed by startup Omni Speech, AI Detect uses AI to catch AI. CEO David Przygoda explained that the company’s machine learning algorithms extract features such as intonation from audio clips and use them to teach models to distinguish between real and fake audio. I am. “Humans can be amused in one sentence and distraught in the next, but humans have great difficulty switching between different emotions,” Przygoda said.

There are existing commercial detection tools that rely on machine learning, but they can be unreliable due to factors such as audio quality and media format. Berisha believes that similar to the challenges faced by AI-generated text detectors, AI-based detectors will become less effective over time. This led him to develop a process to authenticate the human voice as words are spoken.