News10NBC Investigates: Here’s what happened when we did a deep fake on Berkeley Brean’s voice
Right now we’re going to talk about audio deep fakes. Technology can take our real voices and use artificial intelligence to have us say anything.
That’s a deep fake.
We tested it on my voice. It happened in the engineering lab of Neil Zhang, a UofR graduate student who was just awarded a $64,000 grant from the National Institutes for Justice to fund his work on picking out the deep fakes.
First, we recorded my voice.
“My name is Berkeley Brean at News10NBC,” I said.
Then Zhang grabbed a woman’s voice already in his system.
It said, “We don’t see it as downgrading.”
Then Zhang converted that voice to my voice and played it back.
The whole process took three minutes.
“So this is a fake voice of yours and saying exactly the same thing as that woman,” he said.
Brean: “So your work is to try to identify that fake.”
Neil Zhang, UofR Ph.d. Engineering student: Yes. By having a system that can tell this is real or fake.”
Brean: “What are the implications of deep fakes? Why would it affect the average person?”
Neil Zhang: “So people may use the voice to verify it is your identity. For example the bank transaction, we need to verify it’s really you doing the action. What deep fake might hurt is if some other people generates your voice with some AI technology, artificial intelligence, and that voice may sound really similar to you. But it’s fake. It’s not your voice. It’s not your recording. They might steal your money. They might do kidnapping. So they can do many crimes with your voice, with your fake voice.”
Two weeks ago, University at Buffalo Professor David Doermann testified to Congress and urged them to work with social media companies to identify and label the deep fakes.
The fakes usually get the most attention.
“It’s things that people say – wow! I didn’t know this and they share it with other people,” he said. “But now they should be saying – wow! I didn’t know this. Am I sure this is real?”
Part of the problem is you don’t need to be a Ph.D. at UR to do this.
Brean: “So you found these clips of me from stories I’ve done here at News10NBC.”
Melody Emm, News10NBC producer: “Yeah, I used three clips that you recorded this week.”
News10NBC producer Melody Emm ran my voice through a simple website and prompted it to talk in my voice and say “this is an A-I generated voice.”
The playback sounded exactly like my voice.
Brean: “Hmm, sounds just like me.”
Melody Emm: “It does.”
Emm said the website is not complicated.
“It didn’t even take me five minutes to learn how to use this program,” Emm said.