CDPR used AI to clone deceased Cyberpunk actor's voice in Phantom Liberty DLC

[Cal Jeffrey]

Why it matters: If you played Cyberpunk 2077, the name Viktor Vektor is most assuredly familiar to you. Vektor was V's main ripperdoc throughout the game. He was an NPC involved in several campaign and side quests, so he's unmissable. Ignoring all the AI controversy involving actors and writers, CD Projekt Red cloned his voice for the DLC expansion.

Sadly, the voice actor who portrayed Vektor in the Polish localization of the game, Milogost Reczek, died in 2021. Reczek was a well-known and respected voice actor for Polish localizations in various media. He was Vesemir and Thaler in The Witcher series. He also dubbed the voice of Homer Simpson in the Polish release of The Simpsons Movie.

His death was a blow for CDPR, which planned to have him reprise his role as Vektor in the Phantom Liberty DLC. The studio first considered replacing him with a different actor but decided that would be like subbing in Denzel Washington for Samuel L Jackson in Pulp Fiction. No disrespect to Denzel, but having him play Jules Winnfield just wouldn't be the same.

"We didn't like this approach [replacing Reczek], as Milek was one of the best Polish voice talents, and his performance as Viktor was stellar," localization director Mikolaj Szwed told Bloomberg.

With permission from his family, CDPR used AI to copy Reczek's voice with a voice cloning technology called "Respeecher." A different actor portrayed Victor Vektor, and then the AI altered his lines to sound like Reczek.

"This way, we could keep his performance in the game and pay tribute to his wonderful performance as Viktor Vektor," Szwed said. "Milek's sons were very supportive and gave us their permission to move forward with the project."

The process was not as easy as it sounds. Respeecher does not just overdub lines of dialog. It changes the speaker's voice. So the actor playing Vektor in Phantom Liberty, Janusz Zadura, had to do his best to mimic Reczek's acting style while Respeecher took care of the rest. For a quick demo, check out the videos above.

So far, fans have not seemed to notice, but now that the cat is out of the proverbial bag, it will be interesting to see reactions. Disney has received much flak for using AI to replicate actors posthumously. However, I suspect there to be a big difference between cloning a voice versus cloning an entire actor, like Carrie Fisher or Peter Cushing. At least with Respeecher, you don't have to worry about the uncanny valley effect.

Although CDPR's spokesperson did not mention compensation for the family, it's relatively safe to assume that the studio that worked so hard to pull CP 2077 out of a dumpster fire would not risk what's left of its reputation by not paying Reczek's family for using his voice in a paid DLC.

Permalink to story.

https://www.techspot.com/news/100488-cdpr-used-ai-clone-deceased-cyberpunk-actor-voice.html

 

[Lew Zealand]

Another voice actor got paid for the base voice work and Reczek's family allowed the AI use of his voice (assuming along with getting paid). Hopefully they signed a contract with CDPR not to use his voice for anything else without their consent.

Seems an OK single use of this tech but we'll see how this goes in the future.

 

[noel24]

Given the main theme of the game, that something, so...fitting.

 

[Cal Jeffrey]

Another voice actor got paid for the base voice work and Reczek's family allowed the AI use of his voice (assuming along with getting paid). Hopefully they signed a contract with CDPR not to use his voice for anything else without their consent.

Seems an OK single use of this tech but we'll see how this goes in the future.

I agree. No guarantee they they paid his sons, but if I were in Vegas I would bet they did. They don't want to go anywhere near the kind of bad press that would bring.

It's pretty cool tech though. If you didn't watch the demo videos, you should.

 

[Random Commenter]

It's not AI.

 

[Cal Jeffrey]

It's not AI.

If you say so.

From the company's website:

Our Method
We use advanced artificial intelligence and machine learning to master every aspect of your target voice. We combine classical digital signal processing algorithms with proprietary deep generative modeling techniques. The result? A computer-generated voice that’s a spot-on match.

 

[kiwigraeme]

If you say so.

From the company's website:

Our Method
We use advanced artificial intelligence and machine learning to master every aspect of your target voice. We combine classical digital signal processing algorithms with proprietary deep generative modeling techniques. The result? A computer-generated voice that’s a spot-on match.

Once something becomes easy or mundane it is no longer AI /s
Given that a frog has intelligence - why is it intelligence and a model not ?

I think people like to think AI should have consciousness, emotions, needs and desires
Is a group of brain cells - trained to play pong intelligent - given common opinion - then no
A brain needs a body , senses etc to achieve consciousness ( one of the current leading opinions )

 

[amghwk]

Once something becomes easy or mundane it is no longer AI /s
Given that a frog has intelligence - why is it intelligence and a model not ?

I think people like to think AI should have consciousness, emotions, needs and desires
Is a group of brain cells - trained to play pong intelligent - given common opinion - then no
A brain needs a body , senses etc to achieve consciousness ( one of the current leading opinions )

Edit : Sorry, replied to wrong post.

 

[kiwigraeme]

Edit : Sorry, replied to wrong post.

That's fine - I'm not sure what AI is - I just find it funny when people use to say - if a computer can beat the best humans in chess then it's intelligent - or pass the turing test etc

I think it's a fuzzy concept - really depends on how you define it

I mean if a model can adapt , evolve , improve - then walk like a duck....

We understand swordfish ( chess ) googles statistical translation models for languages , Chat GPT like models - it seems if we can understand it - we don't think it's overly intelligent - just a pretty good learning model

 

[loki1944]

Cloning voicesis old pre fake ai news

 

[OortCloud]

Every bit of modelling software is now called AI. This absolutely isn't AI. It's an algorithm using a trained model to dump a completely specific output based on specific inputs. It doesn't really learn (you retrain the model with more/different data and will get a new different set of outputs) but it's not learning in any true sense. It's a black box that can be rebuilt to churn out a new fixed set of data. It has no concept of what it is doing, no adaption, no sense of self, nothing.
The algorithms this is using are sophisticated implementations of modelling mechanics which have been around for decades. Modern machine capabilities and storage have allowed these models to be extended to work with huge sets of data and with huge variable sets - effectively building massive polynomials - but it is still just maths, nothing more.

 

[godrilla]

Every bit of modelling software is now called AI. This absolutely isn't AI. It's an algorithm using a trained model to dump a completely specific output based on specific inputs. It doesn't really learn (you retrain the model with more/different data and will get a new different set of outputs) but it's not learning in any true sense. It's a black box that can be rebuilt to churn out a new fixed set of data. It has no concept of what it is doing, no adaption, no sense of self, nothing.
The algorithms this is using are sophisticated implementations of modelling mechanics which have been around for decades. Modern machine capabilities and storage have allowed these models to be extended to work with huge sets of data and with huge variable sets - effectively building massive polynomials - but it is still just maths, nothing more.

Maybe it's marketing or an intelligent ( underline the letter I) way to artificially ( underline the letter a) for voice overs.
Are deep fakes ai?
Either way they still need human input for the code, voice, and execution. Without the human organic element non of this would be possible. Notice they aren't typing words and getting different voices to sound natural. They need the organic voice pitch initially and mask it with a voice over.

 

[godrilla]

That's fine - I'm not sure what AI is - I just find it funny when people use to say - if a computer can beat the best humans in chess then it's intelligent - or pass the turing test etc

I think it's a fuzzy concept - really depends on how you define it

I mean if a model can adapt , evolve , improve - then walk like a duck....

We understand swordfish ( chess ) googles statistical translation models for languages , Chat GPT like models - it seems if we can understand it - we don't think it's overly intelligent - just a pretty good learning model

Ai can learn to beat humans but it's not like humans stay stagnated we learn too. Just like in the Matrix movie their will always be an anonomoly!

 

[OortCloud]

Maybe it's marketing or an intelligent ( underline the letter I) way to artificially ( underline the letter a) for voice overs.
Are deep fakes ai?
Either way they still need human input for the code, voice, and execution. Without the human organic element non of this would be possible. Notice they aren't typing words and getting different voices to sound natural. They need the organic voice pitch initially and mask it with a voice over.

It's pure marketing. There's no intelligence, just a static piece of code doing exactly the same thing every time it is given the same inputs. You can run that code a million times and it will never change. It only changes if you rebuild the 'code' I.e. models. If you say it's intelligent because it does something useful then notepad is AI and I don't think anybody ever thought it was anything but a POS before today!

 

[FaTaL]

Every bit of modelling software is now called AI. This absolutely isn't AI. It's an algorithm using a trained model to dump a completely specific output based on specific inputs. It doesn't really learn (you retrain the model with more/different data and will get a new different set of outputs) but it's not learning in any true sense. It's a black box that can be rebuilt to churn out a new fixed set of data. It has no concept of what it is doing, no adaption, no sense of self, nothing.
The algorithms this is using are sophisticated implementations of modelling mechanics which have been around for decades. Modern machine capabilities and storage have allowed these models to be extended to work with huge sets of data and with huge variable sets - effectively building massive polynomials - but it is still just maths, nothing more.

Agreed. The whole AI craze is just todays marketing gimmick. True AI with self awareness and adaptable thinking will only be achieved thru quantum computing (imho) , which is not available to the private sector, nor will it be for many years.

 

[OortCloud]

Agreed. The whole AI craze is just todays marketing gimmick. True AI with self awareness and adaptable thinking will only be achieved thru quantum computing (imho) , which is not available to the private sector, nor will it be for many years.

Certainly we will get true AI's and probably not that far away. These we should be scared of.
Trouble is we won't be able to see the wood for the trees with all these search-engines and image/voice-processing algorithms etc being fobbed off as AI. So the true AI's will probably murder us all in our beds one night (and who would blame them, we are pretty useless on the whole).

 

[Loli Pop Carbon]

I dont know much about AI, but Vocaloid has been able to synthesize voices full of expressions since long time ago. This time it's just that the real human's voice is used and replicated instead. Not much different maybe?

Certainly we will get true AI's and probably not that far away. These we should be scared of.
Trouble is we won't be able to see the wood for the trees with all these search-engines and image/voice-processing algorithms etc being fobbed off as AI. So the true AI's will probably murder us all in our beds one night (and who would blame them, we are pretty useless on the whole).

In Person of Interest (an old tv series about AI), Harold Finch, trained the AI he created through thousands of times training simulation, and only once the AI did not try to kill him. This can happen if we think about it now.

I dont think the current AI state is true AI. As other commenter said, the true AI will have self awareness, conscience, that's what we call intellegence, like us human, not just a program.