OpenAI’s ‘Sky’ looks a lot like ScarJo, according to voice researchers

Researchers at Arizona State University discovered that Scarlett Johansson’s voice actually sounds like “Sky,” OpenAI’s currently discontinued GPT-4o voice, according to NPR, who commissioned the comparison. After using AI models developed to analyze vocal similarities to compare Sky to about 600 other actresses, the lab found that Johansson’s voice was “more similar to Sky’s than 98% of other actresses.”

But models also “often” discovered that Anne Hathaway and Keri Russell’s voices sounded more like Sky’s than Johansson’s. Visar Berisha, the professor who led the analysis, said NPR that Johansson’s voice is “similar, but probably not identical.” Berisha’s other work includes Origin Story (PDF), an F.T.C. challenge winning microphone which marks voice recordings as created by humans.

However, the researchers concluded that there were “undeniable commonalities” between the voices. For example, the analysis would have shown that Sky and Johansson’s vocal tracts would have “identical” lengths (“vocal tract” refers to the throat, mouth, and nasal passages that would produce a given voice).

Some differences included that Sky’s voice was a little higher-pitched and more expressive than Johansson’s, while hers was “a little more breathy” than the model’s, according to NPR. We asked Berisha for more information about his analysis and its limitations, and will update if he responds. OpenAI did not immediately respond to a request for comment.

Both OpenAI CEO Sam Altman and CTO Mira Murati have denied that Sky intends to sound like Johansson. Following his GPT-4o demonstration earlier this month – after which Altman published a one word post which simply said “her” – Johansson said that Altman asked her to lend her voice to the model, which she refused, and that he tried again two days before the demonstration.