As an actor, DiMercurio is interested in how many emotions and “micro observations” you can pick up on just by the way someone says a word. Some actors trust their gut, or do an impersonation, and others look at voice granularly, observing, re-creating, and manipulating the speed of the speech, the inflection and the placement, to function as a set of “levers” for, say, producing different audiobook characters.
When it comes to voice-over more generally, she thinks AI is now passable and that we may get to the point where it’s almost as nuanced as talking to a person, but “I don’t think it will ever hit quite the same.”
In the short term, she expects a flattening in advertising audio, similar to the sudden homogeneity in graphic design a few years ago when it seemed like all brands started to look the same. “Almost every voice you hear, there’s someone behind that,” she says, “even the AI ones were a person who recorded that at one point.” But AI voices are designed to be palatable to the widest audience possible, “therefore we’re losing the specificity, the identity, the little quirks—like nobody’s s’s whistle like mine do. You don’t think about it, you don’t even hear it, because it’s so neutral.”
Ultimately DiMercurio predicts that voice actors will become a high-end refinement in some industries. “A human voice is going to become bespoke,” she says. “We’re going to become a luxury item, almost thinking of it like artisanship. So if you’re a luxury brand, you’ll have a real person’s voice instead of AI in your commercials and in your products. In the same way that you can get handmade ceramics and bowls or you can buy them from Wal-Mart.”
A now infamous case study showing the power of a single, distinctive human voice came last May when OpenAI was forced to pause the use of its Sky voice for GPT-4o, one of five initial voices for the chatbot. This came after Scarlett Johansson—yes, her—hired legal counsel, claiming that OpenAI had imitated her after she refused a request from its CEO, Sam Altman, to license her voice for the product and after Altman had tweeted this single-word tweet: her.
OpenAI denied that Sky was intended to resemble the star, and The Washington Post then compared recordings of the Sky AI voice and those provided by an anonymous actress, reporting them to sound identical. The actress’ agent also sent the Post documents to prove that she was hired for the Sky voice role after an open casting call, months before Altman approached Johansson. The actress also said that neither Scarlett Johansson nor the movie Her were mentioned by OpenAI staff, who specified that they wanted a “warm, engaging” and “charismatic” voice.
DiMercurio refers to the whole ChatGPT ScarJo incident as a “debacle” but notes there is a funny twist related to her own voice acting. In 2023, the developers behind the Zombies, Run! fitness game released a Marvel tie-in for iPhone and Android named Marvel Move, featuring interactive adventures where you’re instructed to run around with Thor, the X-Men, and The Hulk in order to save the world.
“I got hired to play Black Widow in that game. I was thinking, ‘I finally get to do Scarlett Johansson,’” she says with a laugh. “Then we get in there and they say, ‘We don’t want this to sound like her. We want this to be completely new.’ I was like, when do I get to have my Her moment? But I got to go home and tell my mom I played Black Widow.”
Still, DiMercurio probably gets her Her moment every day, every time thousands, maybe millions of people hear the words “Bluetooth connected.”
_.png "The Best Food Dehydrators for Meat, Fruit, and Vegetables")