According to voice researchers, OpenAI’s “Sky” sounds very similar to ScarJo

Researchers at Arizona State University have discovered that Scarlett Johansson’s voice actually sounds like “Sky” – the voice of GPT-4o, which is now retired from OpenAI, According to NPR, which commissioned the comparison. After using artificial intelligence models developed to analyze vocal similarities to compare Sky to approximately 600 other actresses, the lab reportedly concluded that Johansson’s voice was “more similar to Sky’s than 98% of other actresses.”

But the models also “often” said that Anne Hathaway and Keri Russell’s voices were more reminiscent of Sky’s than Johansson’s. – said Visar Berisha, the professor leading the analysis NPR that Johansson’s voice is “similar, but probably not identical”. Berisha’s other works include Origin story (PDF), FTC winning microphone watermarking voice recordings as man-made.

Nevertheless, the researchers reportedly concluded that there were “undeniable similarities” between the voices. For example, the analysis reportedly found that Sky’s and Johansson’s vocal tracts would have been “identical” in length (the “vocal tract” refers to the throat, mouth, and nasal passages from which a given voice comes).

According to some differences, Sky’s voice was slightly higher and more expressive than Johansson’s, while her voice is “slightly more breathable” than the model’s. NPR. We asked Berisha for more information about his analysis and its limitations. We will update this information if we receive a response. OpenAI did not immediately respond to a request for comment.

Both OpenAI CEO Sam Altman and CTO Mira Murati denied that Sky was intended to sound like Johansson. Following a GPT-4o demonstration earlier this month, after which Altman released the file one-word entry which was simply “she” – Johansson said that Altman asked her to voice the model, which she declined, and that he tried again only two days before the demonstration.