Yes, antagonizing high-profile voice actors seems like a good way to avoid establishing a bad precedent in court. Judges love seeing intentional, spiteful harassment from a for-profit corporation
If people want voice actors to 'get with the times' in this regard, they need to find a good way to convince them. And if 'get with the times' means 'let someone make money off your main skill that you spent years/decades developing without compensating you' it's an especially hard sell.
Instead people are just going "i'm gonna do this! try and stop me!" and expecting a good response from laymen who don't necessarily understand technology or the law.
This is how you end up with copyright/IP maximalism because it just looks and feels bad to people who don't understand the fine details.
I understand the fine details and it looks terrible. What it looks like, from top to bottom, is that GenAI researchers and developers are engaged in mass theft and free-riding of everyone else's work.
People have rights to their image and likeness. Get with the law and industry kid. This is akin to photoshopping a person into scenes they didn't authorise with the intent to profit from it.
Another AI company on the grift, with unauthorized voice cloning.
Good luck in court. There is absolutely nothing fair use about that so this AI company needs to find another excuse for their scheming before the lawsuits pile up to bankrupt them.
Where exactly is the legal boundary? Are you allowed to clone the voice of a former politician like President Obama for use in a museum exhibit? What about dead people like JFK or Michael Jackson? This particular instance of voice cloning seems hard to defend, but some guidance would be helpful.
I realize it's a company, but aside from a hacker's toolkit, this may be the first big, catastrophic moment in history where normal non-technical people come face to face with the uncontrollable nature of open source and the absolute ephemerality of software. Given things like Stable Diffusion and locally-runnable language models have been around since the inception, it's not that the cat got out of the bag, it's that a bag was never even made. The non-technical crowd is going to have a very difficult time understanding that AI proliferation is completely out of their control and is not ever going to be in their control.
Another coin in the well for just how new and fertile the computer industry is. So many monumental things yet to happen.
AI renditions. Or more appropriately, artificial renditions. How close are they to the real thing? Is it noticeable, like kids wearing and acting out in Brad Pitt masks? Or is it unnoticeable, and therefore deceptive?
Solution is to have laws prohibiting deception by AI plus regulation that AI generators need to have explicit tells that this was generated by AI, for image, audio, and video.
Once deception issue is cleared, then address a person's claim of RIGHTS, that they own their face, their style of body movements, and their voices, and its depiction. No laws are made on this. The only law that could be leveraged is copyright law. If a person's voice, style of body movement, and face is copyrighted, copyright law can be leveraged. What is the process of copyrighting? Copyrighting a movie? Copyrighting an audio tape? Copyrighting on image? Is everything automatically a copyright? Or is the distribution explicitly made under copyright?
I guess? This is extremely common in the arts (superhero movies, 007, Star Wars, cover bands, and theater productions) where the original actor does not do every performance.
Seems like an odd thing from the voice actor to be upset about.
I’m an actor and a software developer. And AI replacing us (the nobodies, the extras, ie not Tom Cruise) scares the shit out of me. Long live the theatre I guess.
AI won't replace you. It will enhance you and give you way too much work to cram into a small window of time. AI will be a series of trials and errors first, before it becomes something that you will learn how to finetune.
I remember the Web was this chaotic place of info "at your fingertip" and no one mentioned how difficult it is to find stuff. AI is just as fancy and shiny.
"AI will enhance you" is a really hard sell for voice acting. Voice actors are paid to use their physical training and experience in order to perform as directed in a unique way. Even if a voice actor were to go "now I trained this model on my own voice so I don't have to perform anymore", no sensible game studio or animation director would want to pay full price for an AI performance. The AI isn't going to be able to satisfy the precise demands of the director at a recording session, at least without a lot of domain specific training and extra work.
It's like telling a guitarist that AI will empower them because now the computer will play music for them.
An actor could probably additionally sell access to an AI version of their voice on the side, but doing that is devaluing their voice, arguably.
> no sensible game studio or animation director would want to pay full price for an AI performance. The AI isn't going to be able to satisfy the precise demands of the director at a recording session, at least without a lot of domain specific training and extra work.
You’re assuming that it will be used for production. With a model than can spit out sentences on demand, game writers can try out a lot of different dialogue for draft storylines then have the final version recorded by the voice actor.
I'd 100% expect game studios to move to pure AI voices as soon as it's feasible to do so. There are even very strong use cases for it, like using it to voice generated dialogue so that NPCs don't need to be scripted word-for-word.
The entire genre of Internet video, too, is likely to do it, albeit because of cost.
Movies and TV may keep using voice actors if it's the prestigious thing to do. Who knows. The point about directing the actor is also good, but I don't know how long it would last.
TTS is already used for placeholder dialogue in games, yeah. That has been the case for a long time. There's not really a precedent for using fake versions of an actor's voice, but placeholder TTS dialogue has accidentally slipped into large releases before, and a few notable games have come out using AI instead of voice actors (The Finals, one of Cyan's games, etc)
> It's like telling a guitarist that AI will empower them because now the computer will play music for them.
I don't need AI for that - we've had drum machines for ages and now VSTs that can emulate pretty much any instrument. They do empower me because now I can create music that isn't limited to the instruments I own and can play without needing to hire an orchestra.
That's not empowering a guitarist to play guitar in some new or better or more efficient way.
Drum machines and VSTs are tools for any composer to synthesize a composition. They solve a different problem for a different set of people that can include but is not limited to guitarists. And as you pointed out they predate AI.
How do drum machines and VSTs make you better at playing guitar, or allow you to do that more efficiently or for higher pay? How will AI do that?
Even if you're right that AI will "enhance" actors' skill, then that will raise the bar to an un-realistic standard that consumers will then expect. That's not a good thing.
And another thing > no AI is going to replace your Daniel Day Lewises. Your Tilda Swintons. These people are real human beings who are perfect on screen without AI adjustment.
Simulate a character acting in various fashions and styles and saying the same exact thing. Show off the character in your likeness to other people. Ask them which simulation is best.
The theatre's an intersting concept, like live music. I understand that every town once had plenty of performances and shows, but then came record record players, and the "best" would outcompete the local shows, ditto for movies and tv. Seems this is a continuation of such.
Already happened (in multiple ways) with Tupac. Released posthumous albums and reproduced his likeness holographically. I think its inevitable that actors will do the same.
It's not such a big difference between watching at home a concert, and going to a fake concert. Both are not connecting you to the artist, just giving you their music and performance. Arguably it's much more impressive if you go to the holographic concert - you have a bunch of vibing people to mingle, impressive sound systems and fancy visuals all around.
I'm resigned that it was all inevitable, but also just upset that we spent all this effort learning to automate away the parts of life that are actually interesting first.
How did people decide that writing and art and music and theatre were the things that we needed to automate out? Especially with these fields, it's not even about removing drudgery - the very process of creating is the interesting part!
I don’t believe it’s the inevitable yet. What if it’s shit and you don’t want to see hundreds of AI extras? Never mind an AI Tom Cruise (who won’t be getting any academy awards if he’s entirely computer generated).
It has also been super gross to me to watch these infant technologies be thrown at removing humans from creative tasks. The whole point was supposed to be to automate the mundane, boring bullshit so we could do more things involving higher thought, not less.
This is how industrial factories work though. Instead of things like pottery, food and such being crafted it's now just bog standard and crappy. There are benefits in bog standard and crappy though.
There are no benefits to bog-standard, crappy replacements for genuine human-crafted creative products.
Period.
These tools should be applied to solving real problems. Instead, they're being used to solve nonproblems like "how can creatively-bankrupt but affluent parasites extract more money from human creativity while paying as few other human beings as possible?".
It's gross, and it's also pathetic, because the models aren't even /there/ yet. They're certainly gonna get there, but I can't help but feel my future is being robbed at the behest of people who want to remove humans from humanity, rather than remove humans from mundanity, and all we're going to end up with is a pile of statistically-average mediocrity.
No one decided that. Those are just the things with the most quantifiable datasets to absorb and also not quantifiable enough to have to provide sources for their origins.
WAME's response is offensive. I cannot honestly believe they thought this would actually go anywhere positive. I don't know if it was covered in the article (maybe I missed it) but the Voice Actor sure didn't seem to be _consulted before_ this was made. How the fuck did they honestly think the original Voice Actor would not even raise the slightest bit of concern?
Chances are people were paid to make this thing, and some fuck-head manager saw $$$ but didn't once stop to think about liabiility $$$.
Never heard of WAME before, but I suspect they're about to get WAME'd into the dirt.
Microsoft made a coding AI based on your code, yet software engineers seems to be happy having such bots. So software engineers seems to have a different mindset about these things.
Copilot's unlicensed handling of people's code is one of the few areas where "you can't just ignore the license when it comes to training/reproduction" is popular around here.
The coding bot can be used by me to make more money / work less. I dont think WAME are giving the AI voice acting bot to the original voice actor so he can put it to work making money are they ?
There are a lot of nuances based on the type of license associated with the code that was used to train copilot.
Have there been breaches of license agreements? I think so. I have no way to prove it (there may be organizations who have). Should those breaches be litigated? Yes.
However, in that same vein, this should be litigated as well.
I would love to be able to generate an open source application by using others open source code! Would be amazing if it could just collect all sources in a similar way to how one would make references/citations in an essay, and “forcing” the code to be released as open source as well. Basically build on all the open source snippets that people release to power the future.
My code there is all MIT licensed, I think it's totally fair to interpret permissive licenses as allowing AI model training. They don't train on private repos.
For people who work on GPL'd stuff sure, it's more questionable.
MIT licensed work still requires attribution, which (as far as I know) current AI models / training practices are unable to handle properly. From the MIT license text:
> The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
The violation of creators and artists with AI tools (copyright violation, impersonation/likeness stealing) is a serious matter that a lot of people aren't taking seriously or respecting.
The biggest problem seems to be with companies exploiting that law and/or courts haven't kept up with technology, and companies trying to run past that and grab billions/trillions before the government finally steps in (and it's too late, like Uber and Airbnb ignoring regulations).
There's a lesser problem with random people piggybacking onto those company efforts, and contributing to the problem like a thousand papercuts, as well as helping to normalize the violations.
A good example of normalization is that a twitch/YouTube streamer with a large following set up an automated TTS system for his chat to read user messages using ai versions of the characters from gta5, which presumably includes Ned Luke. It's cute and very casual but that also makes it kind of insidious in that now any random user can put words into Ned's mouth without even learning how to use the tech, and it'll be broadcast to a big audience.
It's not just one streamer doing this. There are services[0] providing tons of presumably unlicensed tts voices to any streamers who want them and I've seen them all over the place. It really bothers me how little thought people are putting into this before doing it.
The sum total of a life can’t be reduced to an algorithm. We are so much more than that!
A human actor brings the sum of their life to each role. Joy, anger, fear, sadness; these can’t be felt by an LLM running on an Nvida chip in some random data center.
Our humanity matters, and it saddens me that you don’t seem to share that belief.
Most voice recordings are not for artistic purposes. I feel you’re falling for a false choice fallacy, because there’s a world for both kinds of performances.
Furthermore small companies without a budget are enabled to create things now that they were not able to do before because they don’t have the money for many voice actors for multiple languages. Even large companies may prototype things and create placeholder recordings.
Is it really hard to use our imagination and move on to another person? In your example they replaced Roiland with a different sounding character (think English) on Solar Opposites and it wasn’t a big deal.
Nor is it hard to find people who sound similar either. AI voices have a hard time sounding emotional imo.
I see, I totally misunderstand your question now that I've read your reply.
I think shows should go on and while it can be tragic if someone dies prematurely it's not the end of the world to hire someone completely different to take the part.
It's one of the reasons why I'm not a fan of de-aging you see in actors, just hire someone new and green. Allow them to showcase their talents.
It worked for the Godfather, no reason why it can't work now.
Some notes on voice cloning since everyone seems to be wrong about both how it works and its potential impact on the industry:
- Voice cloning has many good open-source implementations so it's possible for any new company to spin up easily. (the "moat" is having a higher-quality model)
- Most implementations of voice cloning only require 10 seconds of clean audio, so they can be obtained quickly.
- Implementations like the open-source projects and ElevenLabs do not train a model; instead it creates an encoding which represents a person's voice, and is what is stored/loaded. The encoding is used with a "base model" (trained on an often-secret dataset due to legal liability) for the generations. It also makes it difficult to narrow down which voice is used if a malicious voice is used in the wild.
- As with all AI models, it's infeasible to remove a voice from the base model without retraining it.
- The audio output of SaaS voice clone models are 44.1kHz, which is CD quality. Voice acting tends to require much higher quality audio for mastering. Open source models are much lower quality.
On the ethics of voice cloning:
- Modern voice acting tends to be collaborative process between the actor and the audio director/producers to get the best performance.
- Current AI-generated voices cannot display contextual and important nuance, such as pauses and tone inflections. These often aren't things that can be added post-production.
- Voice actors tend to be contractors, with the clients being the company and "owning" the performance. Voice actors alone don't have legal leverage to get voices taken down, although companies like Rockstar do.
- Of course, getting the audio to say embarrassing things could land the voice actor in hot water with the company, especially those less tech-savvy.
And, as of last week, some observations from the SAG-AFTRA deal:
- Replica Studios notably doesn't have a individual-oriented product, instead requiring a significant cost and is more intended for teams.
- Replica Studios apparently doesn't encode the voice and does a full finetune of the model on a new voice.
Were people upset for Justin Roiland being unilaterally replaced by other voice actors (that try to recreate his style of playing the character and unique voice)?
What about Carrie Fisher's Princess Leia performance in her last movie?
Are you allowed to take hours of recordings of Sean Connery's voice, splice words into sentences, and make a radio commercial out of it without paying Connery? You aren't, but this is exactly what AI companies are doing. AI companies want to hide behind the fact that their splicing machine is very fancy.
> 129. GTA 5 actor goes nuclear on AI company that made a voice chatbot of him (pcgamer.com) 72 points by pavel_lishin 3 hours ago | flag | hide | 73 comments
I get that most everyone these days is freaked out about AI. There is a lot of smoke and very little fire here.
If the voice actor in question here has the option of hiring a lawyer and suing the company for wrongful use of his image. He's a professional voice actor and there is quite a bit of legal precedent^1 on his side. If they're making money off his voice/image, he has every right to sue to have all that cash transferred to his bank account, plus damages for theft of opportunity.
If they're doing it, not as quid pro quo, but instead as a marketing gimmick to sell their product, well there is legal precedent on his side as well.
If they're giving it away, though, isn't that just the same as fan fiction? Do we really want to crack down on what is essentially a fan site? Is that the precedent you want to set? It could actually drive more business to the real guy: Maybe he can strike a deal with someone to sell the rights to his voice for whatever he can get for it? Maybe this actually drives up his ability to bargain for a higher price?
Please don't let shortsighted defense of the unknown prevent you from seeing the opportunities presented.
92 comments
[ 8.6 ms ] story [ 157 ms ] threadInstead people are just going "i'm gonna do this! try and stop me!" and expecting a good response from laymen who don't necessarily understand technology or the law.
This is how you end up with copyright/IP maximalism because it just looks and feels bad to people who don't understand the fine details.
https://firstamendment.mtsu.edu/article/right-of-publicity/
Specifically, here, the issue would be "appropriation".
Good luck in court. There is absolutely nothing fair use about that so this AI company needs to find another excuse for their scheming before the lawsuits pile up to bankrupt them.
Another coin in the well for just how new and fertile the computer industry is. So many monumental things yet to happen.
Solution is to have laws prohibiting deception by AI plus regulation that AI generators need to have explicit tells that this was generated by AI, for image, audio, and video.
Once deception issue is cleared, then address a person's claim of RIGHTS, that they own their face, their style of body movements, and their voices, and its depiction. No laws are made on this. The only law that could be leveraged is copyright law. If a person's voice, style of body movement, and face is copyrighted, copyright law can be leveraged. What is the process of copyrighting? Copyrighting a movie? Copyrighting an audio tape? Copyrighting on image? Is everything automatically a copyright? Or is the distribution explicitly made under copyright?
There's a crackdown underway on Elvis impersonators.[2] Not clear how that will come out.
It's not clear that the actor has a case. He's just an employee. Rockstar Games owns the trademark.
[1] https://lawyerdrummer.com/2020/01/are-tribute-acts-actually-...
[2] https://www.reviewjournal.com/entertainment/entertainment-co...
Seems like an odd thing from the voice actor to be upset about.
I remember the Web was this chaotic place of info "at your fingertip" and no one mentioned how difficult it is to find stuff. AI is just as fancy and shiny.
It's like telling a guitarist that AI will empower them because now the computer will play music for them.
An actor could probably additionally sell access to an AI version of their voice on the side, but doing that is devaluing their voice, arguably.
You’re assuming that it will be used for production. With a model than can spit out sentences on demand, game writers can try out a lot of different dialogue for draft storylines then have the final version recorded by the voice actor.
The entire genre of Internet video, too, is likely to do it, albeit because of cost.
Movies and TV may keep using voice actors if it's the prestigious thing to do. Who knows. The point about directing the actor is also good, but I don't know how long it would last.
I don't need AI for that - we've had drum machines for ages and now VSTs that can emulate pretty much any instrument. They do empower me because now I can create music that isn't limited to the instruments I own and can play without needing to hire an orchestra.
Drum machines and VSTs are tools for any composer to synthesize a composition. They solve a different problem for a different set of people that can include but is not limited to guitarists. And as you pointed out they predate AI.
How do drum machines and VSTs make you better at playing guitar, or allow you to do that more efficiently or for higher pay? How will AI do that?
I’m a good actor too btw.
Go to the stage acting just like that simulation.
How did people decide that writing and art and music and theatre were the things that we needed to automate out? Especially with these fields, it's not even about removing drudgery - the very process of creating is the interesting part!
The only hope is that people aggressively vote with their wallets against this, but I don't think this would last long.
Period.
These tools should be applied to solving real problems. Instead, they're being used to solve nonproblems like "how can creatively-bankrupt but affluent parasites extract more money from human creativity while paying as few other human beings as possible?".
It's gross, and it's also pathetic, because the models aren't even /there/ yet. They're certainly gonna get there, but I can't help but feel my future is being robbed at the behest of people who want to remove humans from humanity, rather than remove humans from mundanity, and all we're going to end up with is a pile of statistically-average mediocrity.
I looked up what this means, and I really wish I hadn't.
Chances are people were paid to make this thing, and some fuck-head manager saw $$$ but didn't once stop to think about liabiility $$$.
Never heard of WAME before, but I suspect they're about to get WAME'd into the dirt.
The lawsuit is ongoing.
https://www.courtlistener.com/docket/65669506/doe-1-v-github...
However, in that same vein, this should be litigated as well.
For people who work on GPL'd stuff sure, it's more questionable.
> The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
https://opensource.org/license/mit/#:~:text=The%20above%20co....
The biggest problem seems to be with companies exploiting that law and/or courts haven't kept up with technology, and companies trying to run past that and grab billions/trillions before the government finally steps in (and it's too late, like Uber and Airbnb ignoring regulations).
There's a lesser problem with random people piggybacking onto those company efforts, and contributing to the problem like a thousand papercuts, as well as helping to normalize the violations.
[0]https://tts.monster/
If there a difference between another human acting a character vs a model?
Yes. One is an actual living, breathing human who uses their lived experience in order to bring nuance and feeling to their performance.
The other is a machine.
I don't see why one is ok but not the other.
A human actor brings the sum of their life to each role. Joy, anger, fear, sadness; these can’t be felt by an LLM running on an Nvida chip in some random data center.
Our humanity matters, and it saddens me that you don’t seem to share that belief.
Furthermore small companies without a budget are enabled to create things now that they were not able to do before because they don’t have the money for many voice actors for multiple languages. Even large companies may prototype things and create placeholder recordings.
Nor is it hard to find people who sound similar either. AI voices have a hard time sounding emotional imo.
I think shows should go on and while it can be tragic if someone dies prematurely it's not the end of the world to hire someone completely different to take the part.
It's one of the reasons why I'm not a fan of de-aging you see in actors, just hire someone new and green. Allow them to showcase their talents.
It worked for the Godfather, no reason why it can't work now.
Yet billionaires can commit copyright infringement by mass derivative works creation on a exabyte scale with absolutely no worry?
Rules for me, not for thee.
- Voice cloning has many good open-source implementations so it's possible for any new company to spin up easily. (the "moat" is having a higher-quality model)
- Most implementations of voice cloning only require 10 seconds of clean audio, so they can be obtained quickly.
- Implementations like the open-source projects and ElevenLabs do not train a model; instead it creates an encoding which represents a person's voice, and is what is stored/loaded. The encoding is used with a "base model" (trained on an often-secret dataset due to legal liability) for the generations. It also makes it difficult to narrow down which voice is used if a malicious voice is used in the wild.
- As with all AI models, it's infeasible to remove a voice from the base model without retraining it.
- The audio output of SaaS voice clone models are 44.1kHz, which is CD quality. Voice acting tends to require much higher quality audio for mastering. Open source models are much lower quality.
On the ethics of voice cloning:
- Modern voice acting tends to be collaborative process between the actor and the audio director/producers to get the best performance.
- Current AI-generated voices cannot display contextual and important nuance, such as pauses and tone inflections. These often aren't things that can be added post-production.
- Voice actors tend to be contractors, with the clients being the company and "owning" the performance. Voice actors alone don't have legal leverage to get voices taken down, although companies like Rockstar do.
- Of course, getting the audio to say embarrassing things could land the voice actor in hot water with the company, especially those less tech-savvy.
And, as of last week, some observations from the SAG-AFTRA deal:
- Replica Studios notably doesn't have a individual-oriented product, instead requiring a significant cost and is more intended for teams.
- Replica Studios apparently doesn't encode the voice and does a full finetune of the model on a new voice.
Anyone (or anything?) can portray the character.
What about Carrie Fisher's Princess Leia performance in her last movie?
Machine bad.
Human good.
If you're creating a parody, yes it is legal (which sounds like is the case here).
If the "You" is the owner of the 007 trademark, then yes, they the 007 trademark owner is allowed to do that too.
If you're using that exact character in your own production, it is not fair use, then no you cannot do that.
> 129. GTA 5 actor goes nuclear on AI company that made a voice chatbot of him (pcgamer.com) 72 points by pavel_lishin 3 hours ago | flag | hide | 73 comments
If the voice actor in question here has the option of hiring a lawyer and suing the company for wrongful use of his image. He's a professional voice actor and there is quite a bit of legal precedent^1 on his side. If they're making money off his voice/image, he has every right to sue to have all that cash transferred to his bank account, plus damages for theft of opportunity.
If they're doing it, not as quid pro quo, but instead as a marketing gimmick to sell their product, well there is legal precedent on his side as well.
If they're giving it away, though, isn't that just the same as fan fiction? Do we really want to crack down on what is essentially a fan site? Is that the precedent you want to set? It could actually drive more business to the real guy: Maybe he can strike a deal with someone to sell the rights to his voice for whatever he can get for it? Maybe this actually drives up his ability to bargain for a higher price?
Please don't let shortsighted defense of the unknown prevent you from seeing the opportunities presented.
0: https://www.dmlp.org/legal-guide/using-name-or-likeness-anot...