© 2024 KRWG
Play Live Radio
Next Up:
0:00
0:00
0:00 0:00
Available On Air Stations

Even OpenAI says its new voice cloning tool has major potential for misuse

MARY LOUISE KELLY, HOST:

A 15-second audio clip - that is all someone needs now to create a replica of your voice with a new tool from the company that makes ChatGPT. Experts fear the tool will make it easier to execute scams and cons because 15 seconds is less time than it took for me to introduce this story from NPR's Dara Kerr.

DARA KERR, BYLINE: OK, let's do a quiz. Can you tell which one of these clips is a real person? Here's option A.

UNIDENTIFIED PERSON #1: Imagine you're riding a bike down a hill.

KERR: And now option B.

AUTOMATED VOICE #1: Have you ever wondered why a soccer ball soars through the air the way it does?

KERR: Which one was the real person? It was option A. The second clip was made by artificial intelligence using a new tool called Voice Engine. It was created by the company OpenAI. Ari Lightman of Carnegie Mellon University says it's really, really good.

ARI LIGHTMAN: People who are audio engineers and really understand voice and all of the various different intonations and catenations between natural voice and synthetic voice, they might be able to pick out something different, but to me, it looks fantastic.

KERR: OpenAI touts it as a tool that can help people learn to read. And for people who have trouble speaking, it can manufacture voices for them. It can also do simultaneous translation. Here's a person speaking English and an AI clone translating to German. Note it sounds like the exact same person in both languages.

UNIDENTIFIED PERSON #2: Friendship is a universal treasure. It brings joy, support...

AUTOMATED VOICE #2: (Speaking German).

KERR: Voice cloning has been around for years, but it used to take eight hours of audio to get a substandard replica. Now with OpenAI's new tool, it can be done well in seconds.

HANY FARID: Here we are driving the train at 100 miles an hour.

KERR: Hany Farid is a computer science professor at the University of California at Berkeley. He says scammers are already having a field day with voice cloning.

FARID: I can create scams where I call somebody's parents and say, Mom, Dad - sounds like their son or their daughter - I've been hurt. You need to send money.

KERR: Con artists have also impersonated police officers, IRS officials and the U.S. marshals. They've used voice cloning to access people's bank accounts. The tools have also been used to manipulate voters.

(SOUNDBITE OF ARCHIVED RECORDING)

AUTOMATED VOICE #3: Voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again. Your vote makes a difference in November, not this Tuesday.

KERR: This is a robocall sent to voters in New Hampshire before January's primary. It used a voice clone of President Joe Biden to encourage people not to vote. All of this has been done with technology far more basic than what OpenAI is now showing off. The company says it recognizes the risks here, especially during an election year. So it's put up a bunch of guardrails. For now, it's only letting a few people use it, and they can only use it in a specific way. Farid says even with that, it's not enough.

FARID: You are still developing and deploying technologies that is leading to harm, and you're doing it to profit. It's a simple equation.

KERR: Profit is the bottom line. Artificial intelligence is a new frontier where billions of dollars can be made, and it seems no company wants to miss out on that. Dara Kerr, NPR News.

(SOUNDBITE OF MOTOR CITY DRUM ENSEMBLE'S "THE STRANGER") Transcript provided by NPR, Copyright NPR.

NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.

Dara Kerr
Dara Kerr is a tech reporter for NPR. She examines the choices tech companies make and the influence they wield over our lives and society.