AI to voice the additional dialogue?

[DavidW]

I take BobT's point to be that if, say, James Earl Jones (who voices Darth Vader) had died prior to the AI era, Disney would have contracted another voice actor to play Vader, and they would have chosen someone who sounded as similar as possible to Jones, and that person would have actively tried to sound as similar as possible to Jones. Similarly, cartoon-character actors are not infrequently changed but clearly those changes are done so that the new actor sounds as close as possible to the old actor. The question is how this established practice differs from use of AI, and whether it counts as impersonating someone's voice or just playing a character.

Similarly, Kevin M. Richardson voiced a number of different roles in BG; the narrator sounds pretty different to Sarevok, to the point that I actually didn't know they were the same until I googled it. An AI trained specifically on the narrator isn't going to sound like Sarevok; is it impersonating Richardson, or just a role Richardson is playing?

Then again, the narrator in BG2 is apparently John H. Mayer, not Kevin M. Richardson. He's obviously trying to sound like Richardson, and again I hadn't actually noticed they're different people until I googled it just now. If I train an AI on the narrator voicing in BG and BG2, and get back something that sounds as much like them as they do like each other, am I impersonating Richardson? Mayer? Both? Neither?

Come to that, if I have the budget to employ a human actor to narrate my mod, and I audition for a deep-voiced actor and find someone who sounds like Richardson, and give them the BG narrator files to practice on, and they do a great job, is this something I'm obliged to check with Richardson? (Presumably this is exactly what the developers of BG2 did, and I seriously doubt they got Richardson's permission to do so.)

I'm not endorsing a position here, just describing what seems to be its logic.

[Bartimaeus]

1 hour ago, DavidW said:

Then again, the narrator in BG2 is apparently John H. Mayer, not Kevin M. Richardson. He's obviously trying to sound like Richardson, and again I hadn't actually noticed they're different people until I googled it just now.

I was with you until this (note: also not endorsing a specific position here either). I'm only picking a bone here because of you saying "obviously trying to sound like Richardson": I don't think Kevin Michael Richardson and John H. Mayer...really sound even remotely alike, not in how their base voices sound or the particular voices/style that they chose (or were directed) to use for the roles - and never mind the writing of the narration as well, which is also quite different. It'd be hard for me to believe that John H. Mayer would even be asked to try to imitate Kevin Michael Richardson, as I always assumed the rather drastic departure in narration was intended to represent a change of setting as well as the tonal shift between games. It would be especially odd given that Kevin Michael Richardson literally recorded lines for Shadows of Amn, so if they wanted him or somebody that sounded like him to narrate, well...they already had him in the recording studio. Richardson's narration has a generally neutral delivery and feels rather calming (perhaps even encouraging?) in comparison to Mayer's more old-timey fantastical flair for the dramatic, not to mention all the intrigue/danger/suspense that he suggests awaits the player as you're about to visit the new locations hinted at in his narration.

However, it should be noted that I am extremely sensitive to voice changes (to the point where it's immediately obvious and bothersome to me pretty much any time a character changes voice, and I almost always recognize when someone with a distinctive voice like Kevin Michael Richardson appears in something, even for a moment...the amount of things I've seen with him where I immediately went "oh hello Sarevok" after hearing just one sentence from him is way too high), so maybe my take is in the minority.

Edited January 5 by Bartimaeus

[DavidW]

1 minute ago, Bartimaeus said:

However, it should be noted that I am extremely sensitive to voice changes, so maybe my take, even though though it may feel true and obvious to me, is in the minority.

Conversely I'm terrible at distinguishing voices, even more so from memory, and it's a while since I actually listened to the narrator!

[Bartimaeus]

12 minutes ago, DavidW said:

Conversely I'm terrible at distinguishing voices, even more so from memory, and it's a while since I actually listened to the narrator!

I edited in a little more for that paragraph to give more context for that statement: for me, it can be both a blessing and a curse, especially as characters I'm fond of lose their voices. A number of my favorite characters from longer-running games and animated series lost much of their charm for me when they decided to replace the original voice actors/actresses...I just went through this with literally Mario and Luigi of all things with the latest Super Mario game, Super Mario Wonder, when I watched the trailer and I was like "...was Charles Martinet sick while he was recording for this game, what's up with their voices?". Nope, after voicing them for literally 30 years straight, Nintendo finally replaced him. Now I fear the day they replace Princess Daisy...the lady that has played her for the last 20 years straight up made that character for me with the absolute loony-bin way she performs the voice, and I don't think any replacement is going to be able to equal her...but I just know it's going to happen eventually.

There are certain lines from characters I'm particular to that ring out in my head randomly, even years or decades since last hearing them. So I guess you could say that voices are pretty important to me, .

Edited January 5 by Bartimaeus

[Incrementis]

This is a somewhat exciting topic for me, so I looked for an answer and found a source(google/search: "Are AI Voices Legal?").

Two topics seem to be relevant(search result: blog post from voices website "Are AI Voices Legal?" ).

• "Legal Aspects of AI Voice Usage": 
• "Factors Affecting AI Voice Legality"

These topics basically say: Generating an AI voice from a (prominent) person without their permission can lead to legal problems!

 

If you are interested in this topic in general then google/search "Generative Artificial Intelligence and Copyright Law, Updated September 29, 2023". It is relatively short and easy to understand.

 

Do not take my words for granted as I am not a lawyer. It's best to take a look around yourself.

Personally, I would love it if the celebrities allowed us to use AI generated voices for modding.

Cheers!

Edited January 5 by Incrementis

[BobT]

18 hours ago, Morgoth said:

I don't know how things will be sorted out concerning AI but I have the following thoughts:

- generally AI does a very subpar voice compared to original voice-actors. I'm a fan of no voice acting; when it's added to games or mod it should be perfect and I can assure you, we are very far from perfection - unless you are fine with Dagoth Ur not having any inflexion or emotion in his voice when talking to you.

- The Ai does an imitation. Should the original voice actor of duke nukem sue Gianni Matragnano for doing Duke nukem's imitation? 

 

Depends on the AI, I think. It does struggle with accents and emotion, if using a TTS version. However if using the version where you record it with your own voice (putting on the accent / emotion), then use the AI to apply the target voice, I think it does pretty well. It's not perfect, but good enough that can't easily "tell" that it was inserted in, from the bits I've heard. Of course that version is more work though. 

Exactly, regarding the imitation lol. It's transformative work. 

[BobT]

13 hours ago, Thacobell said:

"Yet it's the character "Edwin Odysserion's sound though isn't it, as fictional as the writing for it. No-one's making the "actor" say anything or attributing anything to them. Extending their voice work (FOR FREE) is exactly the same as extending the writing or programming. VA's aren't special here."
There is NO Edwin Odesseiron, he is a fictional character and does not have a voice. That is Jim Meskimen's voice. An actual person.

And the both of you are STILL disregarding the fact that voice actors have already voiced their opinon on the matter.

And? It's still not "making him say things" etc. is it, that could somehow bring disrepute to the VA (when just extending the existing work and character? Stupid argument anyway). It's the EDWIN voice for example, most people wouldn't even know who the VA for that is, unless they bother to look it up. 

 

SOME voice actors may have voiced their opinion. And fine! They're entitled to their opinion. That can still only extend as far as the law reaches, though. 

[BobT]

13 hours ago, DavidW said:

I take BobT's point to be that if, say, James Earl Jones (who voices Darth Vader) had died prior to the AI era, Disney would have contracted another voice actor to play Vader, and they would have chosen someone who sounded as similar as possible to Jones, and that person would have actively tried to sound as similar as possible to Jones. Similarly, cartoon-character actors are not infrequently changed but clearly those changes are done so that the new actor sounds as close as possible to the old actor. The question is how this established practice differs from use of AI, and whether it counts as impersonating someone's voice or just playing a character.

Similarly, Kevin M. Richardson voiced a number of different roles in BG; the narrator sounds pretty different to Sarevok, to the point that I actually didn't know they were the same until I googled it. An AI trained specifically on the narrator isn't going to sound like Sarevok; is it impersonating Richardson, or just a role Richardson is playing?

Then again, the narrator in BG2 is apparently John H. Mayer, not Kevin M. Richardson. He's obviously trying to sound like Richardson, and again I hadn't actually noticed they're different people until I googled it just now. If I train an AI on the narrator voicing in BG and BG2, and get back something that sounds as much like them as they do like each other, am I impersonating Richardson? Mayer? Both? Neither?

Come to that, if I have the budget to employ a human actor to narrate my mod, and I audition for a deep-voiced actor and find someone who sounds like Richardson, and give them the BG narrator files to practice on, and they do a great job, is this something I'm obliged to check with Richardson? (Presumably this is exactly what the developers of BG2 did, and I seriously doubt they got Richardson's permission to do so.)

I'm not endorsing a position here, just describing what seems to be its logic.

This exactly. 
And I still giggle a bit at the prospect of hobby modders somehow getting funding for professional VAs lol, especially when the hobby modder is the one doing the work, for free. 
Same with like I say too, are modders meant to ask the programmers before tweaking and extending their code? The writers? The asset artists or level designers? At that point the hobby modders may as well be expected to fund their own studios lol.. Completely contrary to the point of modding. 

Again, I think with AI a decent set of practices should still be followed, even for hobby modders:

- Credit the original staff of who's work they've tweaked or used for reference.

- Be clear and transparent with what's original and what's the modder's creation.

- If using AI, do not distribute any models, only the output.

- Do not paywall any content that's been generated based on reference to someone else's work. (Ideally none at all).

And for companies:

- Some sort of royalty scheme for where a VA or whomever has contributed to a model, each time a sale is made where the output has been used.

- Clarification on ownership of the models.

- Independent advice for creative staff to aid them in negotiating their contracts with companies regarding models etc.
Regardless, this bit applies to professional companies and is irrelevant for free hobby modding. 

Standard "Impersonation" or "harassment" laws etc. already apply, I don't think anything new is needed there. 

 

  

11 hours ago, Bartimaeus said:

The amount of things I've seen with him where I immediately went "oh hello Sarevok" after hearing just one sentence from him is way too high), so maybe my take is in the minority.

lol, I recently heard Thaedus voice when watching Invincible S2 and I just couldn't get "My name is Optimus Prime" out of my head.. 

Edited January 6 by BobT

[BobT]

6 hours ago, Incrementis said:

These topics basically say: Generating an AI voice from a (prominent) person without their permission can lead to legal problems!

That's so basic and vague it's just a standard @rse covering statement lol. Of course it could, in specific circumstances. Such as say, using it for malicious impersonation of Joe Biden or someone, well you'd have the exact same legal trouble as if you altered something by traditional means or photoshopped them into something compromising. Nothing new there.
Also I imagine, if you were making money off it. I think the transformative work aspect could take it far, but whenever money gets involved, things get complicated, as it should. 

For a free hobby mod though? Much different situation. The "stealing jobs" argument is stupid as with hobby mods, there was never anything to steal. The MODDER is doing the work.

I was talking about this with a friend and she had a go with a tool called "Mangio RVC", check out these examples: 

https://voca.ro/19y8OfnWt6zt

https://voca.ro/13wFCtt97vYZ
Of course it's not perfect, that's just a 5-min half-a-job and without any final polishing in audacity or whatevs, but it's a good start. 

[Thacobell]

 @DavidW The thing is, a human impersonator and ai generation are fundamentally different. The impersonator is at least a person using their own voice of their volition. You aren't loading up a computer program to take that voice from another person. They aren't equivalent.

@BobT "Its not stealing, its fine!"
The actual owners of those voices disagree, as has been noted multiple times. Actors do NOT want their voices taken without consent. But you clearly do not care if it hurts anyone, because you keep ignoring this.
Also, you have been presented with legal issues that can arise from it, as it IS potentially stealing. Which is WHY actors are pushing against this so hard. But again, any reassurance you have to give that its "fine" is just a pretense. You brush everything aside saying its a bad idea, because you don't actually care. You only care what you can get out of it. It doesn't matter to you that vocie actors themselves are saying its stealing, it doesn't matter to you that there is potential legal issues with it. Anykind of justification you give is nothing but pretense.

[morpheus562]

The Crucible mod is going to be wading into the use of AI Voices discussion, and I've paid to use ai software from Replica Studios. First time I've personally put money down to generate a free mod, but I think it will better the mod. Of note, I have already generated 115 fully voiced lines of dialog for various net new characters the player will interact with. I estimate there will be close to 200 fully voiced lines of dialog when it is completed. Any existing character that already has voice work (e.g. Sarevok, Imoen, etc.) will be mute in their interjections as I do view it as an ethical concern to mirror the work of an existing voice actor to voice a character they are behind. I did tweet out to the original voice actor for Sarevok to see what it would take, but I doubt I will get a response as I have no people to reach out to his people to set something up.

[morpheus562]

If through the grace of some divine intervention I'm able to get some of the OG cast to respond, I'd have to resort to some form of crowdfunding like Kickstarter to compensate them.

[DavidW]

2 hours ago, Thacobell said:

The thing is, a human impersonator and ai generation are fundamentally different. The impersonator is at least a person using their own voice of their volition. You aren't loading up a computer program to take that voice from another person. They aren't equivalent.

It's never been entirely clear to me *why* they're different, though. (That's not to say that there isn't a relevant difference, only that I've not seen a clear statement of it.)

[Bartimaeus]

9 hours ago, DavidW said:

It's never been entirely clear to me *why* they're different, though. (That's not to say that there isn't a relevant difference, only that I've not seen a clear statement of it.)

I think there are ethical concerns when we're talking about using someone's exact likeness. When someone draws a picture of someone, nobody gets confused about whether the drawing is actually the person or not, and the drawing can be attributed to the artist; when someone does an audio cut-up where they'll take and re-arrange words that someone says to make something new (typically to the effect of something silly), it's usually pretty obvious because of how unnatural it'll sound/look, even when it's a matter of just cutting out a single word (and also, in the vast majority of cases where this is done, at least some other people will be familiar with the original and know that it was edited and be able to relay that to others who aren't); when someone does a voice impersonation of someone else...well, it's just literally not that person's voice - even if it's really close, there are people who will be able to discern that one person's vocal chords and the unique timbre associated with them (and never mind all their exact little speaking mannerisms, be it word/phrase choice, accents, inflections and cadences, or even just the exact length that they hold or roll their tongue to pronounce certain sounds) simply isn't exactly the same as another person's. And fundamentally, if you can produce it with your own mouth, it can be attributed to you and not anyone else. The difference with this AI crap is that it's not anyone's new unique work because these are just programs that are using the exact likeness of someone else or their work as input in order to output in minutes/hours new content that could pass as being them/theirs, and that likeness can be made to dance to whatever tune the puppeteer likes - without another human being having ever contributed anything new to make it, without being able to attribute the work to anyone else but the person who was used as input. Well, with the exception of generated art taking from so many different sources that it can't be clearly attributed to any one artist, but all those hack "AI artists" effectively stealing work from others and presenting it as their own is a whole other stupid can of worms we don't need to get into. I won't deny that it would be great fun if I could import a recording of someone's voice into a program like Audacity and be able to mess around with some dials and buttons in order to transform that little bit of voice recording to say anything I like, but no matter how I edit or transform it, that voice recording is still of that person's voice and no-one else's.

And that's not even to mention the issue of whether of whether the AI makes convincing transformations of someone else's work, which I think is entirely besides the point but certainly adds to the squickiness. I'm not going to go out of my way to get too bent out of shape about it when it comes to examples where its use is obvious and not really harmful/exploitative (for example, if someone takes David Warner's recordings of Jon Irenicus from BG2 to make new and free content involving him that's more or less in line with the character he portrayed - if no money is made off of it, if it's clear it's just a character portrayal set in the appropriate circumstances, if everyone who plays that content is dutifully informed of the use of AI before downloading it), but I won't lie in that there are some very obvious avenues for gross abuse with this technology that give me pause as to whether it should be available to anyone for any purpose. Even within just the sphere of Baldur's Gate...wasn't there an Imoen romance mod around at some point? You could easily use an AI to take recordings of Melissa Disney's portrayal of Imoen to make some VERY sketchy mod content around that idea, with Melissa having no ability to protect her likeness in being used for that purpose. If it were someone else impersonating her, then fine - that person would've agreed to have done it (whether for compensation or because they just wanted to do it because they liked the idea or thought it would be fun or...it doesn't really matter), it would be their voice and not Melissa's, and the work could be fairly attributed to them. But when it's an AI, well, it would literally be Melissa's voice, it can really only be attributed to her, even with the qualifier that it was AI-generated. Yuck! And this technology is still in its infancy, relatively speaking...even if you think it's not always convincing now (and I would say that it certainly can be when skilfully used and in the appropriate circumstances), who can say what the future holds?

Edited January 7 by Bartimaeus

[Guest Jesse]

The performance information has to be encoded somewhere.  So at some fundamental level, whether that's in electrical circuits in machine memory or in brains seems to come down to a preference of where power or capital is allocated.  iPods did not decimate the live-musician market, as many feared, but the technology to record-once-play-everywhere transformed that industry, and not broadly in favor to the musicians.