When the generative synthetic intelligence startup OpenAI launched a demo of its new ChatGPT 4o mannequin final week, it included intensive video of its “Voice Mode,” which options an emotive voice answering person questions.
Whereas there are a selection of voices obtainable, viewers observed that certainly one of them, “Sky,” sounded suspiciously like actress Scarlett Johansson, who portrayed the voice of an emotive AI within the 2013 movie Her (in actual fact, OpenAI founder Sam Altman posted “her” on X in the course of the demo).
Now, OpenAI says that it’s “pausing” the usage of the Sky voice because it seeks to deal with the issues from customers about such a well-known voice getting used.
“We’ve heard questions on how we selected the voices in ChatGPT, particularly Sky,” the corporate posted Monday morning. “We’re working to pause the usage of Sky whereas we handle them.”
In a weblog put up, the corporate acknowledged the issues, and defined its course of for creating the voices, noting that it ran an in depth casting course of
“We consider that AI voices mustn’t intentionally mimic a star’s distinctive voice — Sky’s voice shouldn’t be an imitation of Scarlett Johansson however belongs to a distinct skilled actress utilizing her personal pure talking voice,” the weblog put up mentioned. “To guard their privateness, we can not share the names of our voice skills.”
OpenAI says that it started working with “well-known, award-winning” casting administrators and producers in early 2023 to establish totally different voice actors that might turn into the voices within the product, and acquired over 400 submissions. That listing was whittled all the way down to 14.
“We spoke with every actor concerning the imaginative and prescient for human-AI voice interactions and OpenAI, and mentioned the expertise’s capabilities, limitations, and the dangers concerned, in addition to the safeguards we’ve got carried out. It was essential to us that every actor understood the scope and intentions of Voice Mode earlier than committing to the mission,” the weblog put up continued, including that they might ultimately choose the 5 closing voices.
These actors flew to San Francisco, the place the corporate led recording periods, earlier than releasing the voices into ChatGPT final fall.
The tech firm says that it’s going to add new voices to the platform over time.
“We assist the inventive neighborhood and labored carefully with the voice appearing business to make sure we took the fitting steps to forged ChatGPT’s voices,” it mentioned within the weblog put up. “Every actor receives compensation above top-of-market charges, and this may proceed for so long as their voices are utilized in our merchandise.”