Doubt and division echo throughout Hollywood and Silicon Valley as Scarlett Johansson and OpenAI dispute whether or not OpenAI’s now discontinued Sky voice—which the company launched last September—resembles Johansson’s actual voice.
Rijul Gupta, a synthetic media expert and the CEO of DeepMedia AI—a deepfake detection and AI security company based in San Francisco, California—shared a file exclusively with IT Brew showing no evidence of a voice match between Johansson’s real voice and Sky.
“This evidence suggests that it is very unlikely that OpenAI cloned Scarlett’s voice for use in their Sky voice demo,” he told IT Brew in an email. Deep Media—whose customers include major tech companies and the Department of Defense (DOD)—received a $1.25 million grant from the Air Force Research Laboratory to “integrate its tools” into DOD apps, Tech Brew reported last year.
“It is worth noting that DeepMedia’s Voice Biometric AI algorithms are unique in that they are trained on both real and synthetically generated voice samples, which we believe makes them more robust in use cases involving the comparison between real and GenAI voices such as this,” he wrote.
Casting call. In a post on May 19, OpenAI detailed on its website how the voices for ChatGPT were chosen, writing that they narrowed down “over 400 submissions before selecting the five voices.”
“Each of the five distinct voices you hear has been carefully selected through an extensive process spanning five months, involving professional voice actors, talent agencies, casting directors, and industry advisors,” the company wrote.
Altman also said in a statement on May 20 that they cast the voice actor who lent their voice to Sky “before any outreach to Ms. Johansson.” Each of the five voices were “sampled from voice actors we partnered with to create them,” OpenAI’s site also stated.
Top insights for IT pros
From cybersecurity and big data to cloud computing, IT Brew covers the latest trends shaping business tech in our 4x weekly newsletter, virtual events with industry experts, and digital guides.
The Washington Post reported on May 22 that OpenAI did not clone Johansson’s voice for Sky and that a different actor was hired to provide the voice for Sky, “according to documents, recordings, casting directors, and the actress’s agent.” Another recent analysis by AI researchers at Arizona State University indicates that Sky and Johansson’s voices “are similar but likely not identical.”
“The researchers found that Sky was also reminiscent of other Hollywood stars, including Anne Hathaway and Keri Russell. The analysis of Sky often rated Hathaway and Russell as being even more similar to the AI than Johansson,” NPR reported.
Audio file by software engineer Peter Marreck shows OpenAI’s Sky voice (directly from the app) compared to a clone of Scarlett Johansson’s voice he created via ElevenLabs.
A cause for concern. Multiple victims of voice cloning and deepfakes have come forward, and some US lawmakers and cybersecurity experts have also sounded the alarm on AI and the way it can be used to disseminate misinformation and cause harm to everyday people as well as celebrities.
Despite the results of the analysis indicating no evidence of a voice match, Gupta, whose company is “dedicated to pursuing AI safety,” emphasized that no system is perfect, and that “only OpenAI knows if they used Scarlett Johansson’s voice” to train its Sky model.
“If OpenAI did in fact use Scarlett Johansson’s voice without her permission to build a commercially available/revenue-generating product, then we fully support Scarlett’s efforts to protect her identity, her brand, and—quite literally—her voice,” he wrote. “We are entering a society where the answers to these questions are not clear.”
IT Brew has reached out to Scarlett Johansson and OpenAI for comment.