AI Voices - General Consensus Regarding LL Mods

[sixth]

What's the general moral consensus for using AI Generated Voices of Skyrim's Original VA's in Adult Mods? Is it right or wrong considering these are real people we're using the voices of to describe typically very graphic adult topics.

 

I'm of the opinion it's a little too late to care; since the community has already been using xVASynth to achieve the same result (albeit with a lot more effort being required and the quality not being nearly as good most of the time) but what do other people think of using ElevenLabs (or other AI voice generation) in the same way?

[MadMansGun]

i did voice slicing for one of my mods, the ai stuff would just be a easier way of doing what i have already done by hand.

[venomr]

We can already use their likeness. I think we know its a video game and nothing matters, so voices should be fine too.

[Heroine HoneyCrotch]

9 hours ago, sixth said:

AI Generated Voices

 

the better question is would the original voice actors be more offended that their voice is now used for sex mods or if their voice got butchered by vysnch thing. I think the auto generated voices sound like shit personally(some people do do it better,4sure)   Most of it sounds like it should require an oil change after some 2000 gaming miles/hours. I think using vanilla lines, and or slicing vanilla lines like MadMan has done is preferred. To combo all 3 would be pretty good but I dont think I have seen that one.  

[DeathApproaches]

^ THIS

 

xVASynth2 is trash. It's fun to play with, but to actually do something organic-sounding, being indistinguishable from the human voice actor? Just one sentence takes hours worth of work, especially if there's mixed emotion, pitch cadence, you know, real actual conversation? Slicing is actually what you'll be required to do, because the way voice training works can't really accommodate such things.

 

The tech is just not here yet. Go get one of those multi-tera gigabyte ones like troublesofaheroine... it's garbage. flat deadpan, shit pronounciations, cause it's from a line of text between quotes... it's literally impossible to define a human voice that way without some kind of code, intonation marks, phonic alphabet, something!

 

here's a great example, make a "perfectly trained" AI voice do an accent while quoting (or even funnier mis-quoting) a audio reference... top gun "I feel the need, the need for speed... and scientology... no just speed." You as a human with a human brain processing what you just read, know exactrly what I mean if you know the FG reference, but knowing anything about Tom Cruise is even enough to get the giggle. Get your Katy Townsend "Cait" voice to utter that with the cadence and intonations and drawl and volume of the reference, overriding her Irish brogue... what? it's just one line. What's taking so long? :-)

 

There is no AI really going on there's just code executing... the AI is the logic, of the voice building, mostly. That's just dumb assholes, people non-technical using terminology they don't grasp, like "the cloud" what all the plonker means is a network server... fucking non-technical troglodytes, bad enough we let non-degrees people live, but letting them speak and drive and interbreed? UGH. Euthanize the lot of them, I say! Because we live in a capitalist society, and it's common to intentionally keep people ignorant so you can make money off doing stuff for them. The whole reason to join the military is to learn to kill proficiently, and then to get college so you don't have to. Mandatory military service and mandatory college education, what's taking us so fucking long?! Oh right, the south... keep forgetting the south.

 

See my one little tirade. What would that take to speak aloud? 22 seconds. Just timed it. Now make an TTS system speak it in the eact same way I just did, without me controlling the voice from a voiceprint sample, just texutal input. Good fucking luck. Can't be done easily, it will take you weeks, months maybe, when I can do it in less that 30 seconds, over and over an over...

 

Seriously, the TTS software is the same shit it was 40 years go... without YOU using a voiceprint of your own cadence pitch volume, etc, the actual intonation of the voice... how can a synthetic voice, no matter how well "trained" sound anything other than flat, deadpan, a goddamned customer service autoresponder bot

 

I don't understand your request, to speak to a representative, say "representative"

 

You want the voice you have to squeal-giggle in delight like you tweaked her bum passing by? You want her to shout at you like the sexual predator you are? Scream in rage after being assaulted by you? weep-talk about her planned revenge after being gang-raped by raiders, and on and on... a VO actor can do itall in one breath. You think you'll be able, with just the voice file and the training data, to ever be able to get any emotion when TTS'ing just "speak this text in quotes"?? You're delusional.

 

I have about a terabyte of this shit installed locally, every open-source one, every "beta", even ones you're supposed to pay for, cause whoops! -- I have a couple RTX cards at my disposable and am familiar with TensorFlow and other CUDA apps... this is mostly smoke and mirrors... you've heard the samples, you've played the Sound Packs people have generated so far, and some of them are quite capable people! They are not the limitation.. People get really excited thinking it's like DALLE and image sampling or ChatGPT and a bot writing a PhD dissertation

 

It's not RIGHT or WRONG - it's that it's fucking garbage. It takes a LOT of effort to make it not garbage

 

And VO actors like Katy Townsend have nothing to fear... just yet.

 

Because it's not easy. Not yet. You can make a human sounding voice easy enough but making the voice sound human, well that's different isn't it?

 

Trying to convey emotive complex emotion, you know, speaking, one that can break out in song mid-sentence or giggle every other word flirt-talking, then switch to raging out high-speed and full-volume... do you grasp the complexity of what that requires? True AI with tens of hundreds of thousands of samples, and then hundreds of individual clips tuned to each individual emotional state volume cadence pitch everythign, merged together to make just a few seconds of natural-sounding speech, that anyone can do by speaking

 

Course if you're OK with: Press one for English, para español, oprima numero dos, appuyez le numéro trois pour le français ??

 

You should not be fine with it, is the point.

Edited February 26, 2023 by DeathApproaches
typos when you type 230wpm and ranting, it happens

[belegost]

On 2/26/2023 at 10:36 AM, DeathApproaches said:

Euthanize the lot of them, I say!

...

?

 

Hitler approves.

Edited February 27, 2023 by belegost

[SmedleyDButler]

Y'all need more Dagoth Ur memes in your lives.

 

 

[Mars1991]

On 2/26/2023 at 9:36 AM, DeathApproaches said:

^ THIS

 

xVASynth2 is trash. It's fun to play with, but to actually do something organic-sounding, being indistinguishable from the human voice actor? Just one sentence takes hours worth of work, especially if there's mixed emotion, pitch cadence, you know, real actual conversation? Slicing is actually what you'll be required to do, because the way voice training works can't really accommodate such things.

 

The tech is just not here yet. Go get one of those multi-tera gigabyte ones like troublesofaheroine... it's garbage. flat deadpan, shit pronounciations, cause it's from a line of text between quotes... it's literally impossible to define a human voice that way without some kind of code, intonation marks, phonic alphabet, something!

 

here's a great example, make a "perfectly trained" AI voice do an accent while quoting (or even funnier mis-quoting) a audio reference... top gun "I feel the need, the need for speed... and scientology... no just speed." You as a human with a human brain processing what you just read, know exactrly what I mean if you know the FG reference, but knowing anything about Tom Cruise is even enough to get the giggle. Get your Katy Townsend "Cait" voice to utter that with the cadence and intonations and drawl and volume of the reference, overriding her Irish brogue... what? it's just one line. What's taking so long? ?

 

There is no AI really going on there's just code executing... the AI is the logic, of the voice building, mostly. That's just dumb assholes, people non-technical using terminology they don't grasp, like "the cloud" what all the plonker means is a network server... fucking non-technical troglodytes, bad enough we let non-degrees people live, but letting them speak and drive and interbreed? UGH. Euthanize the lot of them, I say! Because we live in a capitalist society, and it's common to intentionally keep people ignorant so you can make money off doing stuff for them. The whole reason to join the military is to learn to kill proficiently, and then to get college so you don't have to. Mandatory military service and mandatory college education, what's taking us so fucking long?! Oh right, the south... keep forgetting the south.

 

See my one little tirade. What would that take to speak aloud? 22 seconds. Just timed it. Now make an TTS system speak it in the eact same way I just did, without me controlling the voice from a voiceprint sample, just texutal input. Good fucking luck. Can't be done easily, it will take you weeks, months maybe, when I can do it in less that 30 seconds, over and over an over...

 

Seriously, the TTS software is the same shit it was 40 years go... without YOU using a voiceprint of your own cadence pitch volume, etc, the actual intonation of the voice... how can a synthetic voice, no matter how well "trained" sound anything other than flat, deadpan, a goddamned customer service autoresponder bot

 

I don't understand your request, to speak to a representative, say "representative"

 

You want the voice you have to squeal-giggle in delight like you tweaked her bum passing by? You want her to shout at you like the sexual predator you are? Scream in rage after being assaulted by you? weep-talk about her planned revenge after being gang-raped by raiders, and on and on... a VO actor can do itall in one breath. You think you'll be able, with just the voice file and the training data, to ever be able to get any emotion when TTS'ing just "speak this text in quotes"?? You're delusional.

 

I have about a terabyte of this shit installed locally, every open-source one, every "beta", even ones you're supposed to pay for, cause whoops! -- I have a couple RTX cards at my disposable and am familiar with TensorFlow and other CUDA apps... this is mostly smoke and mirrors... you've heard the samples, you've played the Sound Packs people have generated so far, and some of them are quite capable people! They are not the limitation.. People get really excited thinking it's like DALLE and image sampling or ChatGPT and a bot writing a PhD dissertation

 

It's not RIGHT or WRONG - it's that it's fucking garbage. It takes a LOT of effort to make it not garbage

 

And VO actors like Katy Townsend have nothing to fear... just yet.

 

Because it's not easy. Not yet. You can make a human sounding voice easy enough but making the voice sound human, well that's different isn't it?

 

Trying to convey emotive complex emotion, you know, speaking, one that can break out in song mid-sentence or giggle every other word flirt-talking, then switch to raging out high-speed and full-volume... do you grasp the complexity of what that requires? True AI with tens of hundreds of thousands of samples, and then hundreds of individual clips tuned to each individual emotional state volume cadence pitch everythign, merged together to make just a few seconds of natural-sounding speech, that anyone can do by speaking

 

Course if you're OK with: Press one for English, para español, oprima numero dos, appuyez le numéro trois pour le français ??

 

You should not be fine with it, is the point.

Main problem right now, with ElevenLabs is that you can't control the speed of words, nor where the AI puts the accent on. Just being able to control those two things would allow you to expand the range of emotion incredibley, and do a lot to get rid of the whole deadpan robot issue. It still won't be able to do very emotional stuff, but there is allegedly software that does that too, it was just never released to the public because of safety concerns. If someone can perfectly mymic your old grandma crying on the phone how she got stuck outside of town and needs 20$ for an Uber, well, i think you can see how that can lead to fraud.

[Guest]

On 2/23/2023 at 8:19 PM, sixth said:

Is it right or wrong...

Ultimately I think it comes down to "informed consent".

For instance, Kate Beckinsdale was some flattered that a modder put her likeness in Skyrim...others in other circumstances might not feel the same way.

Real vs fake? Even voice actors are faking it, right?

Thing is, this is "the new economy" here and you can bet ElevenLabs is recording your efforts...forever. So...there's that.