LogOn: AI helps spot audio deepfakes amid election disinformation threat | VOANews

LogOn: AI helps spot audio deepfakes amid election disinformation threat | VOANews

The ability to make synthetic voice messages known as deep fakes is now Within Reach of anyone with an internet connection. With more than 50 countries holding elections this year, efforts are ramping up to detect disinformation before it causes harm.

Sarah Barington researches audio deep fakes at the University of California Berkeley. “We need to just take some data from you in just a few minutes,” she says. She has cloned this reporter's Voice using an online tool and created a fake version of a short statement.

The real voice sounds like this:

This is a testing sentence that I am using to tell the difference between real and fake.

Then we have the fake version:

This is a testing sentence that I am using to tell the difference between real.

Quite incredible. Barington describes, "what we start to do is put this into a computer and look at the waveforms and the noise of the signal and start to deduce differences. Some clues are visually apparent to a trained eye, like pauses that humans tend to make. Generally speaking, there are more of these pauses in the real voice than there are in the fake."

Barington also analyzes the synthetic voice using an AI trained to spot fakes. “These patterns to the AI are very obvious and they can be things like very small subtle artifacts that we can't hear, but if the fake has been made using an unfamiliar system, then the AI is going to really struggle to understand what to do with that; it might not show the exact same patterns that the previous fakes did.”

Barington says detection, even when successful, may come too late to stop a viral post, and social media companies and their users need to be more involved in reducing the spread of content that could be fake. “We are heading towards a solution that is partially technical but also partially due to changes in society and activities from policymakers. We all have a role to play in keeping it real.”

Matt Dibble VOA News Berkeley

Keyword

• Audio deepfakes
• Synthetic voice messages
• Disinformation detection
• University of California Berkeley
• AI detection
• Voice cloning
• Social media companies
• Policy changes
• Viral posts

FAQ

Q: What are deep fakes? A: Deep fakes are synthetic voice or video messages that are made to imitate real people, often used to spread disinformation.

Q: Why are deep fakes a concern this election year? A: With more than 50 countries holding elections, deep fakes can be used to manipulate public perception and spread false information.

Q: Who is researching solutions for detecting deep fakes? A: Sarah Barington at the University of California Berkeley is actively researching audio deep fakes and their detection.

Q: How does the AI detect deep fakes? A: The AI can spot very subtle artifacts and patterns in the audio that are not easily detected by the human ear.

Q: Are social media companies involved in combating deep fakes? A: Yes, social media companies and their users need to work together to reduce the spread of potentially fake content.

Q: How effective is AI in detecting unfamiliar deep fake systems? A: If the fake is made using an unfamiliar system, the AI might struggle as the patterns may not be consistent with previously identified fakes.

Q: What roles do policymakers play in the context of deep fakes? A: Policymakers are expected to implement changes and guidelines that help in reducing the creation and spread of disinformation through deep fakes.