AI Voice Clones Can Now Say What You Never Said

AI technology can now produce realistic clones of a person's voice, opening up both promising and troubling possibilities. "Today, with nothing more than a keyboard and an internet connection, you can type, and you can get an image, a voice, a video of anything, [of] anybody doing or saying anything," said Hany Farid, a computer scientist at Dartmouth University.

For people who have lost the ability to speak, voice clones can be deeply meaningful. Jules Rodriguez said hearing his cloned voice felt like he had reappeared, and his wife cried at how closely it resembled him before he lost his voice. But the technology can also cause harm through scams and deepfakes, and some uses fall into gray zones, such as voice clones that "speak" after a person has died.

In laboratory tests, people cannot reliably tell whether a voice was made by a machine or a real person, Farid said, describing their performance as essentially flipping a coin. With roughly 20 seconds of someone's voice, AI can produce a clone that listeners identify as the same person.

Farid described widespread fraud, from romance scams to callers claiming a child has been taken and demanding ransom, along with deepfakes of political leaders. His own voice was cloned by an attacker who called an attorney he was working with on a sensitive case; the attorney grew suspicious after about three minutes and called Farid back.

Questions about what happens to a person's voice after death are tricky, said Karina Nagin of Bridging Voice. Some people are writing their wishes into wills, deciding whether their voice may speak at their own memorial or leave messages for a child's graduation, and who may use it. Rodriguez made his choice clear: his voice dies with him.

Source: Science News