© 2021 WFAE
90.7 Charlotte 93.7 Southern Pines 90.3 Hickory 106.1 Laurinburg
Play Live Radio
Next Up:
0:00
0:00
Available On Air Stations

How Animators Are Using Artificial Intelligence For 'The Simpsons'

(SOUNDBITE OF SONG, "THE SIMPSONS THEME")

UNIDENTIFIED SINGERS: (Singing) The Simpsons.

MICHEL MARTIN, HOST:

The longest-running cartoon in TV history, "The Simpsons," aired its 700th episode last weekend. It debuted in 1989, and since then, not one character has aged. Even the voice actors have stuck around for decades. But a few weeks ago, "The Simpsons" included a voice not heard in years.

(SOUNDBITE OF TV SHOW, "THE SIMPSONS")

UNIDENTIFIED SINGERS: (As Edna Krabappel) Remember, if you can teach one kid one thing, then today will be a success.

MARTIN:

You may recognize that voice. It's Bart's teacher, Edna Krabappel, voiced by Marcia Wallace for 25 seasons. But there is a twist.

AMIT KATWALA: Now, Marcia Wallace, who's the voice actor who plays Edna Krabappel, died a few years ago. But they wanted to bring her back for a one-off episode, and to do it, they used recordings from previous episodes, which they spliced together.

MARTIN: That's Amit Katwala, a senior editor at Wired UK. He recently wrote about how animators are using technology to voice characters.

KATWALA: And I was wondering whether they might be able to train an AI to do this kind of thing. Could they, if Dan Castellaneta, who plays Homer, decided to leave the show, replace him with an AI model trained on recordings of his own voice? One of the first things I came across was this YouTube channel called Speaking of AI, which is run by a British guy who's based in Canada called Tim McSmythurs. And he's built a generic model that he's trained on kind of hundreds of hours of data. And then he can tune the model to mimic a particular person's voice,

MARTIN: Like, say, a famous actor. McSmythurs has a model of the voice, and he types up a script about good hygiene.

(SOUNDBITE OF ARCHIVED RECORDING)

COMPUTER-GENERATED VOICE #2: Hi, kids. My name is Adam Driver. And today, we're going to sing a special song. This is the way we wash our hands, wash our hands, wash our hands.

KATWALA: So what he's done on YouTube as he's slotted "Simpsons" voice characters into famous movies. So there's one scene where he takes a clip from "Notting Hill," and he puts Homer into the Julia Roberts role.

(SOUNDBITE OF FILM, "NOTTING HILL")

HUGH GRANT: (As William Thacker) Everyone in the world knows who you are. My mother has trouble remembering my name

KATWALA: Homer's on the doorstep, begging Hugh Grant to take him back.

(SOUNDBITE OF ARCHIVED RECORDING)

COMPUTER-GENERATED VOICE #3: (As Homer Simpson) But don't forget, I'm also just a boy standing in front of another boy asking him if this is a donut shop.

KATWALA: He also does it the other way round. He takes politicians like Donald Trump or Joe Biden and slots them into scenes from "The Simpsons." So you've got Donald Trump reading something that was originally said by Ralph Wiggum.

(SOUNDBITE OF ARCHIVED RECORDING)

COMPUTER-GENERATED VOICE #4: (As Ralph Wiggum) Hi, Lisa. Hi, Supernintendo (ph) Chalmers.

(SOUNDBITE OF CAT MEOWING)

COMPUTER-GENERATED VOICE #3: (As Ralph Wiggum) I'm learning.

MARTIN: These manufactured voices are called deepfakes. You may have heard about them in terms of disinformation as a way to deceive people through manipulated images. Now animators can use this technology to voice characters. But don't expect it to replace your favorite characters just yet. There are a bunch of legal issues that come up with technology like this, and Katwala says there are other limits.

KATWALA: So I can say something like, this is deepfake Amit.

MARTIN: And out comes an eerie but familiar voice.

COMPUTER-GENERATED VOICE #5: This is deepfake Amit. You can tell this is not the real me because the intonation is very flat and unemotional.

KATWALA: When we speak, we're drawing on, you know, years of experience of how to make our voice change to portray a particular emotion or stress a particular point. And the AI model of Homer doesn't know how to do that because it's only been fed on a very, very small subset of the things that Homer Simpson says.

MARTIN: Katwala says that is good news for voice actors.

KATWALA: It's much, much easier to just get an actor to do it or even hire a soundalike if Homer leaves. There's, you know, thousands of people that can do a Homer Simpson-esque voice rather than training a deepfake to do it.

MARTIN: That was Amit Katwala, senior editor at Wired UK, talking about "The Simpsons" and deepfake technology.

(SOUNDBITE OF UYAMA HIROTO'S "81 AUTUMNS")

MARTIN: This is NPR News, and it really is me. Transcript provided by NPR, Copyright NPR.