ST. LOUIS (KMOX) - Some scammers are using artificial intelligence to copy voices and dupe people out of money.
That's why the Federal Trade Commission (FTC) recently asked for solutions and are offering thousands of dollars in prize money.
One participant in that challenge is right here in St. Louis. Ning Zhang, an assistant professor at McKelvey School of Engineering at Washington University has proposed using an audio "water mark" to protect voices from cloning.
"In order for hackers to clone voice, they need audio clip from you which identifies you as a person or a speaker," said Zhang on Total Information A.M. Thursday. "In A.I. terms that generally involves tracking the signature of the sound."
Zhang says his solution to stop the A.I. from tracking that signature is to add a "water mark", noise that would hidden to the human ears, but when A.I. try to extract the identity, the voice embedding would be different.
Zhang says they have had recruited participants to test out the technology by asking them to record a clip, then put the clip through a synthesizer and ask Otter.AI users if the synthesizer audio is good and closely mimicking the participant. Once they do, they add a water marker to the audio and do the same thing again.
"Our tests reveal most human users see after the protection the voice is completely different," said Zhang.
Zhang thinks the technology can potentially be for every day use in the future and that why's they wanted to participate in the FTC challenge.
"We understand there's a gap between prototype and everyday technology, and we were working to reverse that" said Zhang. "Initial results are promising, however they are engineering challenges."
Zhang says he has been talking to voice actors and narrators and initially if the product gets rolled out, he would like them to get a first crack at it and then slowly work to improve depending on the feedback.