zlacker

There is a distinction between the image of a celebrity and their voice. The image of a celebrity is usually pretty cut and dried, it’s them, or obviously intended to be them. If the use of their image isn’t meant to be satirical, it’s problematic. The Crispin Glover/Back to the Future 2 case is a good example of non-satirical use that was problematic. Zemeckis used previous footage of Glover, plus used facial molds of Glover to craft prosthetics for another actor.

Voices…are usually not so distinctive. However, certain voices are very distinct—Tom Waits, Miley Cyrus, James Earl Jones, Matt Berry. Those voices are pretty distinctively those people and simulating their voices it would be obvious who you are simulating. Other celebrity voices are much more generic. Scarlett fits into this with a pretty generic female voice with a faint NY/NJ accent.

Open AI screwed up by taking a generic voice and making it specific to the celebrity by reference and by actually pursuing the actor for the use of their voice.