The world changes. Tech improves, and we start be to able to do things we couldn’t do before.
Talks I give on Machine Learning and AI almost always start with a variant of that.
I do AI / ML as my job (amongst other things), and I love every part of it. I’ve also worked in the game industry, and sometimes (more and more often these days) the 2 intersect.
@Project made a comment around availability of voice actors, which will drive adoption of this tech pretty quickly.
Shows a bunch of Machine Learning systems all put together in a really interesting way - this will have implications for news, and video evidence.
But, it will also change voice acting in a big way. There is more than enough voice from the voice actors in overwatch that you can generate new content without having the actors there.
What does this mean? I guess it will come down to their contracts, but, I would hope that it means they at the very least get paid for it.
What if they start throwing in lines which is actor would disagree with having enough that they would quit the job? Is that a reasonable thing? You are putting their voice to those ideas, so do they get a say in the matter?
We are in for an interesting time in the games industry. The tech has advanced, it will be interesting to see what players in that arena do with it.
The implications on news and faking content is huge. But, people believe a picture of someone with a quote written under them that they never said - so I guess this doesn’t make it worse.
Well I’ve done ML, some work with Anaconda via Jupyter notebook. I understand the implications and fears for the future. But working with ML makes me realize how dumb it is too, since it’s filled with lots of inherited biases from human creators and human data.
Really comes down to their contract though. But I’m sure lots of low paying voice actors will be desperate to accept anything. And this will honestly lower the price considerably the more it gets polished.
Shout out to you for working with ML though, it bores the literal crap out of me. Though statistics did too, which is basically ML’s sister.
It’s interesting to think about and people absolutely should be considering the ethics of the tech but honestly, atm voice actors(and everything else game dev) are so terribly underpaid that it’s probably cheaper than getting someone who can maintain GANs and if it gets *aaS’d it’s probably gonna go to some mundane thing like call centers or chatbots for sales, or some NSAesque shenanigans.
I expect there are already lawyers wetting their pants at the concept of not only how complicated voicework contracts will be, but the amount of work they’ll get from “derivative works” made with voice samples.
I think that’d depend on how many lines you want them to say.
I could be mistaken but, I think much of Overwatch’s talent is union and union requirements run like $825.50 per day/four hours and that’s the minimum.
That’s not lots of money in the corporate sense but, it would probably limit what the writers are doing a fair bit.
Plus, you’ve got the logistics to consider. I’d be able to get the tech to spit out the line for me once I typed it out. Getting the voice actor to do it would take considerably more effort and time.
Studio time, with an engineer, is around $50-$100 per hour. Union rate for voice acting is around $250 per hour. You can probably record around 8000 words in one hour, assuming they’re mostly sentence-length and there are a few retakes and breaks now and then. A really efficient team could probably do 10k in an hour.
That’s a lot of dialogue, and it sounds right when recorded naturally. No AI will ever sound fully human, especially in a stress-timed language like English where inflection changes meaning. I know you think it can be done, but it can’t. Decades ago, AI people thought they could make virtual actors, too – even bring some of them back from the dead. But it never looked right, and never will. The uncanny valley applies to sound, too.
Hypothetically speaking, voice actors have unions that would likely fight very hard against that practice. Unions always have their own problems, but at the very least the Unions like SAG-AFTRA seem to be very aggressive about their actors, and have been clamping down hard on a lot of companies (Nintendo is a prime example actually. They apparently have awful conditions for va’s and lost quite a few for it).
I see this as mostly a problem for “up and coming” voice actors who aren’t union. Such is the sacrifice for the advancement of technology, unfortunately.
It doesn’t seems like going to have much use in game in particular. Games uses only voices and the lines aren’t very long either. You can just have better, faster and cheaper results just by asking the va to do it again on the next session since you will need the va anyway.
In a year you’ll forget this conversation even happened. These conversations have been happening for 20 years. Also in that timeframe, there have been approximately ten million “new battery technologies” that were going to change the mobile device world. Aaaand we’re still using lithium.
Really, AI can’t improvise a human conversation better than the original VA can record it. I’m sure companies will try, and it’ll sound like hammered horseshlt, but people will pretend it isn’t. I’m reminded of a promo I saw for an LOTR game back in like 2003 where the developers were gushing over how it was “just like the movie” while the demo showed clearly that it did not look anything like the movie.
Even if it’s work-for-hire, it’s still good money. The VA might even get paid more upfront for buying out future reuse, depending on the agreement, but even if they do get residuals, it won’t be much. Residual payments are a big issue in the industry. A lot of actors survive off of reuse of old material. But licensing is getting cheaper, the cuts are getting smaller, and studios are almost obsolete due to the rise in home recording. VAs are pretty expendable, though. For a sad story, look up what happened to the original voice of Solid Snake.
And then there are the soundalikes, which are cheaper and more available, and sometimes you can’t tell the difference.
