The beginning of a new Era or the final defiliment - The project to give voice all Morrowind's dialogues with AI has begun

[Tuco Benedicto Pacifico]

They should have gone with Vivec Voice from the in game file.

Sounds like shit tbh

Yep.

 

[KörangarTheMighty]

Just since it's vaguely pertinent... I've crossed a guy on the Larian forum that was experimenting a bit with this stuff and voiced several lines of Viconia in BG2 using AI:

Still doesn't sound right tbh. Grey DeLisle has a certain coquettishness to her voice when she's acting seductive. Especially apparent in number 48 after the "lend me your ear" line.
"That is too bad..." and onward sounds flat, certain words that the actress would draw out are neglected by the A.I. The tech isn't quite there yet.

 

[Tuco Benedicto Pacifico]

Still doesn't sound right tbh.

Eh, it still sounds leagues better than a lot of "amateur voice acting" seen in plenty of fan mods (the recent Final Fantasy VII abomination comes to mind) and of a lot of bad voice acting seen in games, in general.
Oblivion, for one, is filled to the brim of lines recited with worse delivery than a lot of AI-based stuff in this thread.

 

[KörangarTheMighty]

Whatever the case, it's clear to me that in a single character the twain should never meet. Switching between the real actress and the A.I. for subsequent lines is somehow more distracting than going from authentic voiceover to text. For unimportant NPCs like innkeepers, merchants, random soldiers etc. you can probably get away with A.I., but I'd always choose a real actor, even a bad one, for the important roles. At the very least, I'll have something to chuckle at (in Oblivion's case especially).

 

[Baron Dupek]

They should have gone with Vivec Voice from the in game file.

I can't imagine listening his dialogues, with this voice, without pulling the cable sooner or later.

 

[Jarmaro]

Kezyma has released an updated demo showcase of his mod. Caisus Cosades, M'aiq the Liar, etc.
At the moment 17.28% of the original voice-lines are finished.

 

[skaraher]

 

[dreughjiggers]

Just since it's vaguely pertinent... I've crossed a guy on the Larian forum that was experimenting a bit with this stuff and voiced several lines of Viconia in BG2 using AI:

Still doesn't sound right tbh. Grey DeLisle has a certain coquettishness to her voice when she's acting seductive. Especially apparent in number 48 after the "lend me your ear" line.
"That is too bad..." and onward sounds flat, certain words that the actress would draw out are neglected by the A.I. The tech isn't quite there yet.

Its all about how much text you feed the labs. Too much, and as you say the delivery goes flat. Too little, you don't hear enough context, and neither does the algorithm.

 

[Tuco Benedicto Pacifico]

Its all about how much text you feed the labs. Too much, and as you say the delivery goes flat. Too little, you don't hear enough context, and neither does the algorithm.

At this point in time in this specific field there are SEVERAL competing solutions that can vary drastically in ease of use, amount of data required to "train" them and verisimilitude/"naturalness" of the result..
For instance this Microsoft version can clone a vocal timbre with 3 seconds of audio feed:



But generally speaking, the more training data the better. There's no such a thing as "too much".

 

[Lemming42]

Can't wait for someone to come up with a proper open source solution for this, preferably that can be run locally. It'll be a big gamechanger when you can make your own audiobooks in a matter of minutes.

Elevenlabs is a highly impressive piece of tech but they're taking the absolute piss with their character limits and payment plans.

 

[Jarmaro]

levenlabs is a highly impressive piece of tech but they're taking the absolute piss with their character limits and payment plans.

Are they, though? It's dirt cheap all things considered. You can make an audibook for an average book for as much as 25$. Pennies.

 

[Lemming42]

I'd be willing to pay a monthly fee for unlimited access, but the problem with their quotas is that you sometimes get results that just go wrong - I've had a couple where certain segments of the prompt got pronounced in totally bizarre ways, meaning the entire prompt needs to be redone, and you thus get "charged" twice for it, since there's no way to refund a bad result.

And then factor in how, if you're looking for serious high-quality results, you'll be messing around with the very sensitive intensity sliders to try and get the exact result you want, meaning you might end up re-entering a prompt four or five times. The quotas which seem generous at first glance end up being pretty tight.

I wasted a good chunk of my free monthly quota just trying to fine-tune a very short joke message for my friends.

 

[Norfleet]

Whatever the case, it's clear to me that in a single character the twain should never meet. Switching between the real actress and the A.I. for subsequent lines is somehow more distracting than going from authentic voiceover to text. For unimportant NPCs like innkeepers, merchants, random soldiers etc. you can probably get away with A.I., but I'd always choose a real actor, even a bad one, for the important roles. At the very least, I'll have something to chuckle at (in Oblivion's case especially).

For unimportant NPCs, AI would probably be preferrable, because otherwise every single unimportant NPC sounds the same unless you have an enormous pool of voice actors to do unimportant roles. With AI-generated, every NPC can sound different when being unimportant.

 

[dreughjiggers]

Its all about how much text you feed the labs. Too much, and as you say the delivery goes flat. Too little, you don't hear enough context, and neither does the algorithm.

At this point in time in this specific field there are SEVERAL competing solutions that can vary drastically in ease of use, amount of data required to "train" them and verisimilitude/"naturalness" of the result..
For instance this Microsoft version can clone a vocal timbre with 3 seconds of audio feed:



But generally speaking, the more training data the better. There's no such a thing as "too much".

In the case of 11labs, yes there's such a thing as too much. I don't know about the others.

 

[Zombra]

I'm pleased that the technology has reached a point where it sounds realistic enough for games to actually use. I've been complaining for 30 years about game devs using voice actors instead of speech synthesis. (The fucking Commodore 64 SID chip could synthesize voices with all kinds of inflections and parameters.) It's exciting as hell to see this happen in my lifetime and I'm looking forward to populated game worlds filled with characters who each have their own voice. Once AI conversation engines can generate content the insanity will be complete.

SO. Now the big challenge is going to be EDITING. I observed 4 years ago that editors will be the next great heroes in game writing, and this is even more true with the advent of AI speech. Even with normal voice acted games, designs make players spend WAY too much time sitting around listening to actors talk. Game writers lovvvve for players to read pages on end of their brilliant work. Imagine how much worse it will be when we have to listen to some computer talk instead at 1/10 the speed we can read. The most prestigious games of the future will be the ones that get to the fucking point.

 

[Baldanders]

I can't wait until this is good enough to replace voice actors - insufferable f*gs.

 

[Baldanders]

I'd be willing to pay a monthly fee for unlimited access, but the problem with their quotas is that you sometimes get results that just go wrong - I've had a couple where certain segments of the prompt got pronounced in totally bizarre ways, meaning the entire prompt needs to be redone, and you thus get "charged" twice for it, since there's no way to refund a bad result.

And then factor in how, if you're looking for serious high-quality results, you'll be messing around with the very sensitive intensity sliders to try and get the exact result you want, meaning you might end up re-entering a prompt four or five times. The quotas which seem generous at first glance end up being pretty tight.

I wasted a good chunk of my free monthly quota just trying to fine-tune a very short joke message for my friends.

Sometimes the errors are funny, like Joe Rogan mispronouncing Automaton https://www.youtube.com/watch?v=-0QxuL7StOM

 

[Seethe]

Still doesn't sound right tbh.

Eh, it still sounds leagues better than a lot of "amateur voice acting" seen in plenty of fan mods (the recent Final Fantasy VII abomination comes to mind) and of a lot of bad voice acting seen in games, in general.
Oblivion, for one, is filled to the brim of lines recited with worse delivery than a lot of AI-based stuff in this thread.

This reply shows how you basically desire the degenerate future where machines are pushed in front of humans. Kinda like how this retard below instantly betrays it

I can't wait until this is good enough to replace voice actors - insufferable f*gs.

I'm trying to understand where this is coming from, because this is not the first time I've heard it. Are you seething that your voice dries pussies in real life? A hint of envy perhaps, towards the genetically superior. You realize that "AI" (lmao) won't make you sound less like a retarded nerd. Or maybe you requested a voice actor on Twitter to be your monkey and say sentences on command, and you got BTFO/ignored.


More shockingly, I didn't expect RPGcodex to shill this much for voice acting in RPGs, let alone "voice acting" done through identity theft software.

I'm pleased that the technology has reached a point where it sounds realistic enough for games to actually use. I've been complaining for 30 years about game devs using voice actors instead of speech synthesis. (The fucking Commodore 64 SID chip could synthesize voices with all kinds of inflections and parameters.) It's exciting as hell to see this happen in my lifetime and I'm looking forward to populated game worlds filled with characters who each have their own voice. Once AI conversation engines can generate content the insanity will be complete.

So you've been complaining for 30 years about devs using voice actors instead of speech synthesis, but now that you have supposedly synthesis that can mimic humans perfectly that is exciting? Isn't this redundant, if the speech synthesizer sounds exactly like a human?

 

[MWaser]

It's hard for me to exactly point out the absurdity of going "yeah I wish more games used shitty sound chip voice synthesis instead of voice acting", you know, if you don't like voice acting then how about just... read the text?

Are you morons using some kind of voice synthesizer to read these web pages or something?

 

[Seethe]

It's hard for me to exactly point out the absurdity of going "yeah I wish more games used shitty sound chip voice synthesis instead of voice acting", you know, if you don't like voice acting then how about just... read the text?

Are you morons using some kind of voice synthesizer to read these web pages or something?

Average codexer lacks an internal voice I guess. Best form of voice acting in CRPGs/RPGs is hearing snippets of voices of various characters, and then read text in your mind with their voices. Or not, if you don't like their voices. Not looking towards the future of even more plastic shit.

 

[Mr. Magniloquent]

This is very interesting. Dev teams will probably be able to hire 1 or 2 actors with decent range/breadth to train "personality" patterns for unique characters. Devs will then be able to mix voice samples of prior art to get distinct voices. Eventually libraries will be so big, it will be like selecting a soundset for your MC on character creation, except you'll be able to give it whatever bark you want. Shovelware is about to get much more talkative.

 

[Tuco Benedicto Pacifico]

This reply shows how you basically desire the degenerate future where machines are pushed in front of humans. Kinda like how this retard below instantly betrays it

Yours shows me you are a retarded with a bias that makes a load of moronic assumptions.

 

[Harthwain]

So you've been complaining for 30 years about devs using voice actors instead of speech synthesis, but now that you have supposedly synthesis that can mimic humans perfectly that is exciting? Isn't this redundant, if the speech synthesizer sounds exactly like a human?

The problem with humans is that most of them are bad at voice acting. So if you can get the technology to where a synthetized voice sounds exactly like a human, and has the added benefit of voice acting well, then this is something to be excited for, considering how weak voice acting for games is in general. But I find that to be a tad optimistic.

 

[Tweed]

Still waiting for someone to bring the Gothic 1 English VA into Gothic 2 and 3.

 

[Tuco Benedicto Pacifico]

Has science gone too far?