Arizona State University researchers have reportedly discovered that Scarlett Johansson’s voice does indeed sound like OpenAI’s now-discontinued GPT-4o voice “Sky” NPR, commissioned for comparison. According to reports, after using an artificial intelligence model developed to analyze voice similarities and comparing Sky to approximately 600 other actresses, the lab found that Johnson’s voice was “more similar to Sky than 98% of other actresses.”
But the model also “often” found that Anne Hathaway and Keri Russell’s voices were more similar to Skye’s than Johnson’s. Professor Visar Berisha, who led the analysis, said NPR Johnson’s voice was “similar, but probably not the same.” Berisha’s other works include OriginStory (PDF), an FTC-challenged microphone that adds a watermark to human-created recordings.
Nonetheless, the researchers reportedly concluded that there were “undeniable commonalities” between the voices. For example, analysis reportedly showed that Skye and Johnson had “the same” vocal tract length (the “vocal tract” refers to the throat, mouth, and nasal cavities that produce specific sounds).
Some of the differences reportedly included that Skye’s voice was higher and more expressive than Johnson’s, while her voice was “slightly more breathy” than the model’s. 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.
OpenAI CEO Sam Altman and chief technology officer Mira Murati both denied that Sky intended to sound like Johansson. After a demo of GPT-4o earlier this month, in which Altman published a one-word post that simply read “her,” Johnson said Altman had asked her to voice the model, but she refused and he tried again. Once just two days before the demo.
