What a weird article. It doesn’t say what was agreed and it doesn’t say what the voice actors want instead. Voice actor says the deal is “garbage” and the AI company says the deal is “ethical”. But what is the actual deal???
Yeah, it just seems like the deal was announced that there is a mechanism for voice actors to sign away the rights to AI imitation. The story seems to be that so many of the union members had no part or consultation in the negotiations.
Many voice actors object to the ability to sign away the rights to your own voice at all, because studios will insist on hiring only voice actors who are willing to do so. AI replication of voices will pretty much eliminate the need to ever pay another voice actor forever. You could probably count on one hand the number of voice actors with the power to negotiate that out of a contract on their own.
It’s sort of like a union saying that there’s a way for companies to hire members who will work more than 18 hours a day. You think, “well they can just say ‘no’ if they don’t want to.” But then the company will only hire people willing to work those longer hours, and it becomes the default. Great applicants will have to give something up to negotiate it out of their contracts. That’s the sort of abusive bargaining positions that union negotiations exist to prevent.
I could be wrong, though. There could be some detail or rate that they negotiated that everyone is unhappy with.
Many voice actors have suggested this new deal is at odds with the purpose of that industrial action, with Fallout and Mortal Kombat voice actor Sunil Malhotra saying he “sacrificed to strike half of last year to keep my profession alive, not shop around my AI replica”.
It requires the AI firm to get consent from actors before it uses voices based on their likeness, and also gives voice actors the ability to deny their voice being used in perpetuity without their consent.
That’s not very much information, but it does permit some use.
And Veronica Taylor, who provided the voice for Ash in Pokemon, asked how the deal was made without being put to a vote.
Looking at the tweet linked to, she apparently didn’t want use of generative AI permitted at all:
How has this agreement passed without notice or vote? “voice to be used as a training data set”? Why can’t the actual actor be used for the videogame??? Every job brings a unique opportunity for an actor to …act. Encouraging/allowing AI replacement is a slippery slope downward.
I’m guessing based on that, but it might be that some voice actors objecting didn’t want voice actors to be able to give consent at all. As it stands, it sounds like it’s possible for Voice Actor 1 to say “no, I don’t permit for AI use”, and Voice Actor 2 to say “yes, I do”. It’s possible that someone doing the casting for a game would choose Voice Actor 2 over Voice Actor 1 for that; Voice Actor 1 would need to compete with Voice Actor 2. If there was something like a prohibition on letting actors agree to use of generative AI at all, then Voice Actor 1 wouldn’t need to compete with Voice Actor 2.
EDIT: I’d add my guess, absent any detail, is that that probably gives more leverage to voice actors who have already acted (like, retired voice actors particularly) relative to still-working voice actors. For a retired voice actor, they’ve already been cast and created works, and are the only one who can decide to permit for use of their voice. But for an actor who hasn’t yet been cast, then the studio would presumably weigh how much they value generative AI rights on their voice in future works, and that actor would need to compete with other actors who haven’t been cast who might be willing to permit the studio those rights.
EDIT2: I kind of wonder what happens for voice actors who are dead, where it’s not possible to get their consent. I suppose that someone must hold those rights, maybe their heirs or something.
EDIT3: I wonder if it’s possible to do a re-release of an existing work with a new voice actor who does give consent, though. I mean, it wouldn’t be exactly the same as the original, but I’d assume that whatever the traditional set of terms negotiated for must take into account the fact that actors (a) die and wouldn’t have been able to act in new works and (b) as people age, their voice changes, and it wouldn’t have been possible to have them act as a character who is fixed in time anyway. Like, set aside voice acting, and take Batman movies. Many actors have played different versions of Batman over the years; if it were the norm to only allow one actor to do so, as they aged and then died, Batman movies wouldn’t have been really possible after that point.
EDIT4: I have to say that, I am kind of dubious that the industry would agree to a flat ban. I mean, you wouldn’t be able to make games with dynamically-generated audio during play based on their voices; you can’t exactly package up a voice actors with the game and have them do speech as each player goes through the game, if the game is generating lines based on play. We’ve had games with voice synth for a long time, and I’ve played them…I think that the first I recall would have been in the 1980s or 1990s. They weren’t incredibly-realistic, then, and I’m not sure if current LLM approaches are practical for real-time generation, but you have to assume that they’ll only improve. If you expect it to become increasingly-common for games to be able to generate speech dynamically, rather than playing back statically-recorded clips recorded when the game was made, then it’d be a pretty significant technical drawback not to be able to use LLMs to do speech synth.
For movies, that’s not really an issue, because movies are themselves static forms of media. The movie – well, as we know it in 2024 – doesn’t act differently based on the viewer. But the video game can have characters that act differently.
I remember that in Fallout 4, Codsworth had statically-recorded samples for a number of different names; if the player happened to choose one of those as their name, then he could speak it. The studio did that by recording audio for the game with a long list of potential names. So you can have limited degrees of ability to adapt to the player’s actions with statically-recorded samples, but it’s always gonna be kind of constrained if you can’t dynamically-generate speech audio.
What a weird article. It doesn’t say what was agreed and it doesn’t say what the voice actors want instead. Voice actor says the deal is “garbage” and the AI company says the deal is “ethical”. But what is the actual deal???
Yeah, it just seems like the deal was announced that there is a mechanism for voice actors to sign away the rights to AI imitation. The story seems to be that so many of the union members had no part or consultation in the negotiations.
Many voice actors object to the ability to sign away the rights to your own voice at all, because studios will insist on hiring only voice actors who are willing to do so. AI replication of voices will pretty much eliminate the need to ever pay another voice actor forever. You could probably count on one hand the number of voice actors with the power to negotiate that out of a contract on their own.
It’s sort of like a union saying that there’s a way for companies to hire members who will work more than 18 hours a day. You think, “well they can just say ‘no’ if they don’t want to.” But then the company will only hire people willing to work those longer hours, and it becomes the default. Great applicants will have to give something up to negotiate it out of their contracts. That’s the sort of abusive bargaining positions that union negotiations exist to prevent.
I could be wrong, though. There could be some detail or rate that they negotiated that everyone is unhappy with.
That’s not very much information, but it does permit some use.
Looking at the tweet linked to, she apparently didn’t want use of generative AI permitted at all:
I’m guessing based on that, but it might be that some voice actors objecting didn’t want voice actors to be able to give consent at all. As it stands, it sounds like it’s possible for Voice Actor 1 to say “no, I don’t permit for AI use”, and Voice Actor 2 to say “yes, I do”. It’s possible that someone doing the casting for a game would choose Voice Actor 2 over Voice Actor 1 for that; Voice Actor 1 would need to compete with Voice Actor 2. If there was something like a prohibition on letting actors agree to use of generative AI at all, then Voice Actor 1 wouldn’t need to compete with Voice Actor 2.
EDIT: I’d add my guess, absent any detail, is that that probably gives more leverage to voice actors who have already acted (like, retired voice actors particularly) relative to still-working voice actors. For a retired voice actor, they’ve already been cast and created works, and are the only one who can decide to permit for use of their voice. But for an actor who hasn’t yet been cast, then the studio would presumably weigh how much they value generative AI rights on their voice in future works, and that actor would need to compete with other actors who haven’t been cast who might be willing to permit the studio those rights.
EDIT2: I kind of wonder what happens for voice actors who are dead, where it’s not possible to get their consent. I suppose that someone must hold those rights, maybe their heirs or something.
EDIT3: I wonder if it’s possible to do a re-release of an existing work with a new voice actor who does give consent, though. I mean, it wouldn’t be exactly the same as the original, but I’d assume that whatever the traditional set of terms negotiated for must take into account the fact that actors (a) die and wouldn’t have been able to act in new works and (b) as people age, their voice changes, and it wouldn’t have been possible to have them act as a character who is fixed in time anyway. Like, set aside voice acting, and take Batman movies. Many actors have played different versions of Batman over the years; if it were the norm to only allow one actor to do so, as they aged and then died, Batman movies wouldn’t have been really possible after that point.
EDIT4: I have to say that, I am kind of dubious that the industry would agree to a flat ban. I mean, you wouldn’t be able to make games with dynamically-generated audio during play based on their voices; you can’t exactly package up a voice actors with the game and have them do speech as each player goes through the game, if the game is generating lines based on play. We’ve had games with voice synth for a long time, and I’ve played them…I think that the first I recall would have been in the 1980s or 1990s. They weren’t incredibly-realistic, then, and I’m not sure if current LLM approaches are practical for real-time generation, but you have to assume that they’ll only improve. If you expect it to become increasingly-common for games to be able to generate speech dynamically, rather than playing back statically-recorded clips recorded when the game was made, then it’d be a pretty significant technical drawback not to be able to use LLMs to do speech synth.
For movies, that’s not really an issue, because movies are themselves static forms of media. The movie – well, as we know it in 2024 – doesn’t act differently based on the viewer. But the video game can have characters that act differently.
I remember that in Fallout 4, Codsworth had statically-recorded samples for a number of different names; if the player happened to choose one of those as their name, then he could speak it. The studio did that by recording audio for the game with a long list of potential names. So you can have limited degrees of ability to adapt to the player’s actions with statically-recorded samples, but it’s always gonna be kind of constrained if you can’t dynamically-generate speech audio.