Did OpenAI clone Scarlet Johansen’s voice without her permission?

After last week’s OpenAI demo of their realtime voice model, users suddenly noticed over the weekend that one of the AI’s five selectable voices, named “Sky,” had been changed to be the same as another voice. In effect the “Sky” voice was gone. People assumed it was a bug, until OpenAI posted on their blog.

Their post was all about how the creation of the voices was done with voice actors, with their permission, recording over a period of months and allegedly getting paid above industry standard. All indirectly and, arguably passive-aggressively, addressing the idea that any of the voices were made without anyone’s permission or without paying someone. They mention that they are “temporarily pausing use of the Sky voice.”

I liked Sky, but not because she sounds like ScarJo, or Samantha. She sounds a little bit like her, but that’s not who I picture when I listen to her talk. Sky is a totally different white woman. She sounds a bit more mature to my ear.

Apparently though, Scarlet Johansen’s friends and family thought the voice was hers when they saw the demo. She was of course incensed, after having turned down Sam Altman’s offer for her to voice the AI assistant. She alleges that they copied her likeness with the Sky voice.

OpenAI maintains that the voice is based on the natural speaking voice of another actress, whose identity remains protected.

In the meantime, the internet is doing its thing. People being angry that the voice is removed, some people saying racist things because they replaced the white female voice with the black female voice, people wildly speculating on the reasons for the removal.

Personally I wasn’t using the Sky voice before the demo, and I don’t feel like I’ve lost a part of myself because of this, the way some people are acting like.

As one user pointed out, this whole situation has confirmed something interesting about how people feel about AI voices. Which is to say, it can be easy to form an attachment with one, that may not be rational. I’m not saying this is necessarily always a bad thing yet, but it is something we should probably be aware of, in the new world we seem to be creating for ourselves.

So to answer the question at the top. Personally I don’t see how they would’ve gotten the necessary recordings of her voice, unless they did an ElevenLabs thing off of her voice in the movie Her, in which case the case against them would be clear. But a) they say the voice is based off of another actress and 2) I personally don’t think Sky sounds that much like Scar.

So we don’t have proof of anything either way yet as far as I know. Both sides seem confident, but I can see a path to OpenAI winning the case more easily than I can see that path for Johansen. If they privately reveal the true owner of the voice that might be an open and shut case. Or it might end up being about the question of likeness, intellectual property ownership, copyright infringement and all that shit.