We have a Steam curator now. You should be following it. https://store.steampowered.com/curator/44994899-RPGHQ/
Chat client updated, if you have issues using chat press CTRL + SHIFT + R to force a hard refresh.

[Resource] RVC voice modules

Game development hub. Projects, modding, and resources.

Moderator: Mod Janitor

Ignore Topic
User avatar
Metalhead33
Posts: 333
Joined: Feb 26, '24

Geolocation

Post by Metalhead33 »

rusty_shackleford wrote: ↑ April 4th, 2024, 15:48
This reminds me of how people are unaware pixel art for consoles was designed around the limitations of the medium which is why it looks like **** without scanlines.
Not just the CRT scanlines, but also the NTSC distortion of the signal, which blended in the dithered pixels.

Image
rusty_shackleford wrote: ↑ April 4th, 2024, 15:48
Yes, they knew the audio recording devices had limitations, and they were making voices that sounded best within that limitation, not something to be cleaned up later.
Except when we're training an AI voice clone, we aren't making it for a fake 1930s movie - we're making it for a mod for a contemporary video game, where high-quality audio is expected.

I like Lotte.

Tags:
User avatar
rusty_shackleford
Site Admin
Posts: 45443
Joined: Feb 2, '23
Gender: Watermelon

Geolocation

Adventurer's Guild

Post by rusty_shackleford »

Metalhead33 wrote: ↑ April 4th, 2024, 16:03
Except when we're training an AI voice clone, we aren't making it for a fake 1930s movie - we're making it for a mod for a contemporary video game, where high-quality audio is expected.
But it still sounds wrong.
Thank you for your attention to this matter!
Steam friend code: 40552640 https://steamcommunity.com/friends/add | email: [email protected]
Having trouble running an old Windows game?
Rusty's Stuff Collection
User avatar
Metalhead33
Posts: 333
Joined: Feb 26, '24

Geolocation

Post by Metalhead33 »

rusty_shackleford wrote: ↑ April 4th, 2024, 16:05
Metalhead33 wrote: ↑ April 4th, 2024, 16:03
Except when we're training an AI voice clone, we aren't making it for a fake 1930s movie - we're making it for a mod for a contemporary video game, where high-quality audio is expected.
But it still sounds wrong.
Would you rather have a mixture of voices (in the same game or same mod), where some have heavy static (and zero bass) while others sound ultra-HD?

I like Lotte.
User avatar
Metalhead33
Posts: 333
Joined: Feb 26, '24

Geolocation

Post by Metalhead33 »

orinEsque wrote: ↑ April 4th, 2024, 13:53
You definitely need more epochs. The conversions are too tinny.
I am re-training my modules at 500 epochs.

I wonder how many epochs would be needed to approach ElevenLabs' quality?
Metalhead33 wrote: ↑ April 5th, 2024, 05:10
I forgot to unsubscribe from ElevenLabs, so they charged my credit card $5 yesterday. So I decided to use up all the monthly credits as soon as possible, and then unsubscribe.

So, here is some stuff from my fictional world's lore:

https://waysofdarkness.miraheze.org/wik ... War_of_718
https://voca.ro/11fRT1Lgm2uZ

https://waysofdarkness.miraheze.org/wik ... Revolution
"The Rape War" is NOT a reference to @Hubumba
https://voca.ro/1ir4QiEIletY

Sypnosis of my future book, that I am co-writing with someone
https://voca.ro/1khV3gP4xpcH

The aforementioned co-writer's OC
https://voca.ro/1cVV6yesNaZq

One of the villains, who initially pretends to be an ally
https://voca.ro/1oKPJrYFNbrJ

Yes, I used the Stronghold Crusader narrator voices and Stronghold Crusader soundtrack.
Last edited by Metalhead33 on April 5th, 2024, 05:19, edited 1 time in total.

I like Lotte.
User avatar
orinEsque
Posts: 4917
Joined: Oct 9, '23
Location: Dubai
Gender: Potato

Geolocation

Post by orinEsque »

Metalhead33 wrote: ↑ April 5th, 2024, 05:17
orinEsque wrote: ↑ April 4th, 2024, 13:53
You definitely need more epochs. The conversions are too tinny.
I am re-training my modules at 500 epochs.

I wonder how many epochs would be needed to approach ElevenLabs' quality?
Metalhead33 wrote: ↑ April 5th, 2024, 05:10
I forgot to unsubscribe from ElevenLabs, so they charged my credit card $5 yesterday. So I decided to use up all the monthly credits as soon as possible, and then unsubscribe.

So, here is some stuff from my fictional world's lore:

https://waysofdarkness.miraheze.org/wik ... War_of_718
https://voca.ro/11fRT1Lgm2uZ

https://waysofdarkness.miraheze.org/wik ... Revolution
"The Rape War" is NOT a reference to @Hubumba
https://voca.ro/1ir4QiEIletY

Sypnosis of my future book, that I am co-writing with someone
https://voca.ro/1khV3gP4xpcH

The aforementioned co-writer's OC
https://voca.ro/1cVV6yesNaZq

One of the villains, who initially pretends to be an ally
https://voca.ro/1oKPJrYFNbrJ

Yes, I used the Stronghold Crusader narrator voices and Stronghold Crusader soundtrack.
Probably a lot. And more varied input in some cases. Sometimes when you don't have enough samples you can train an eleven lana sample, have it read a book chapter or something and use the output as further input for RVC
Victors clap when others succeed; Losers feel every spotlight as a personal bleed.
User avatar
Metalhead33
Posts: 333
Joined: Feb 26, '24

Geolocation

Post by Metalhead33 »

orinEsque wrote: ↑ April 5th, 2024, 08:37
Probably a lot. And more varied input in some cases. Sometimes when you don't have enough samples you can train an eleven lana sample, have it read a book chapter or something and use the output as further input for RVC
If only I knew this earlier.

(I generated a ton of conversations between Baldur's Gate 1-2 characters for the lulz using ElevenLabs. I still have them saved on my HDD.)

I like Lotte.
User avatar
orinEsque
Posts: 4917
Joined: Oct 9, '23
Location: Dubai
Gender: Potato

Geolocation

Post by orinEsque »

Metalhead33 wrote: ↑ April 5th, 2024, 08:44
orinEsque wrote: ↑ April 5th, 2024, 08:37
Probably a lot. And more varied input in some cases. Sometimes when you don't have enough samples you can train an eleven lana sample, have it read a book chapter or something and use the output as further input for RVC
If only I knew this earlier.

(I generated a ton of conversations between Baldur's Gate 1-2 characters for the lulz using ElevenLabs. I still have them saved on my HDD.)
Ah but i think you can continue training an already trained model.
Victors clap when others succeed; Losers feel every spotlight as a personal bleed.
User avatar
Metalhead33
Posts: 333
Joined: Feb 26, '24

Geolocation

Post by Metalhead33 »

orinEsque wrote: ↑ April 5th, 2024, 10:39
Ah but i think you can continue training an already trained model.
And that is exactly what i am doing.
Turns out, it's taking longer than I expected though. I can guarantee that I will post the 500-epoch retrained versions this week though. Just not today.

EDIT: @orinEsque It is done.
Last edited by Metalhead33 on April 10th, 2024, 06:58, edited 1 time in total.

I like Lotte.
User avatar
Vergil
Posts: 15670
Joined: Sep 6, '23

Geolocation

Post by Vergil »

I'm confused where are the mods
I'm just stating the facts.
Question is are you going to gargle the truth or swallow?
User avatar
Norfleet
Posts: 2762
Joined: Jun 3, '23

Geolocation

Post by Norfleet »

The way I see it, if you're making an actual product, AI voice or art needs purely to achieve the level of badness that will absolutely infuriate any professionals that hear it. At which point you challenge them to do it better. Because one of them had the gall to say that hiring an actual voice actor WASN'T some kind of arcane ritual.

Naturally, he provided absolutely no hints as to how such a thing was to be accomplished. It absolutely is an arcane ritual, requiring skillsets from at least two unrelated disciplines that aren't even tangentially connected to the skillset of being a computer programmer. You'd also have to simultaneously be a lawyer and a talent agent. No surprise, given the arcane rituals involved in getting an actual voice actor, that developers are leveraging the skillset they already have, that of making computers do things, and using that to get the job done instead.
User avatar
Metalhead33
Posts: 333
Joined: Feb 26, '24

Geolocation

Post by Metalhead33 »

Norfleet wrote: ↑ May 6th, 2024, 05:39
The way I see it, if you're making an actual product, AI voice or art needs purely to achieve the level of badness that will absolutely infuriate any professionals that hear it. At which point you challenge them to do it better. Because one of them had the gall to say that hiring an actual voice actor WASN'T some kind of arcane ritual.

Naturally, he provided absolutely no hints as to how such a thing was to be accomplished. It absolutely is an arcane ritual, requiring skillsets from at least two unrelated disciplines that aren't even tangentially connected to the skillset of being a computer programmer. You'd also have to simultaneously be a lawyer and a talent agent. No surprise, given the arcane rituals involved in getting an actual voice actor, that developers are leveraging the skillset they already have, that of making computers do things, and using that to get the job done instead.
The problem is, that AI voices cannot really be legally used without the original voice actor's consent. Now, sure ,for mods and anything that will fly under the radar, it can work, but not so much for commercial works.

See the California Legal Code:
California Legal Code wrote:
(a) Any person who knowingly uses another's name, voice, signature, photograph, or likeness, in any manner, on or in products, merchandise, or goods, or for purposes of advertising or selling, or soliciting purchases of, products, merchandise, goods or services, without such person's prior consent, or, in the case of a minor, the prior consent of his parent or legal guardian, shall be liable for any damages sustained by the person or persons injured as a result thereof. In addition, in any action brought under this section, the person who violated the section shall be liable to the injured party or parties in an amount equal to the greater of seven hundred fifty dollars ($750) or the actual damages suffered by him or her as a result of the unauthorized use, and any profits from the unauthorized use that are attributable to the use and are not taken into account in computing the actual damages. In establishing such profits, the injured party or parties are required to present proof only of the gross revenue attributable to such use, and the person who violated this section is required to prove his or her deductible expenses. Punitive damages may also be awarded to the injured party or parties. The prevailing party in any action under this section shall also be entitled to attorney's fees and costs.

(e) The use of a name, voice, signature, photograph, or likeness in a commercial medium shall not constitute a use for which consent is required under subdivision (a) solely because the material containing such use is commercially sponsored or contains paid advertising. Rather it shall be a question of fact whether or not the use of the person's name, voice, signature, photograph, or likeness was so directly connected with the commercial sponsorship or with the paid advertising as to constitute a use for which consent is required under subdivision (a).

(f) Nothing in this section shall apply to the owners or employees of any medium used for advertising, including, but not limited to, newspapers, magazines, radio and television networks and stations, cable television systems, billboards, and transit ads, by whom any advertisement or solicitation in violation of this section is published or disseminated, unless it is established that such owners or employees had knowledge of the unauthorized use of the person's name, voice, signature, photograph, or likeness as prohibited by this section.
I'm not a lawyer, but I don't think that "Btw, all voices in this game are artificially synthesised, any similarity with a real-life voice actor's voice is purely a coincidence" is a valid defense if Cosmo Jarvis comes to sue me for using an AI voice clone of his to voice my character.

Anyway, I'll probably restore the voices, eventually. I'll have to train the existing modules a bit more.
Last edited by Metalhead33 on May 6th, 2024, 06:31, edited 1 time in total.

I like Lotte.
User avatar
Norfleet
Posts: 2762
Joined: Jun 3, '23

Geolocation

Post by Norfleet »

Metalhead33 wrote: ↑ May 6th, 2024, 06:31
The problem is, that AI voices cannot really be legally used without the original voice actor's consent. Now, sure ,for mods and anything that will fly under the radar, it can work, but not so much for commercial works.
I'm not talking about using AI voice to emulate a real voice actor, though. I'm talking about a generic, non-specific-person AI voice. And intentionally doing it half-assedly, so that people will immediately notice what a ****** AI voice it is and begin bitching about it and why you are using this instead of just hiring a real voice actor.

Because seriously, how the **** would I even go about doing that?
User avatar
Metalhead33
Posts: 333
Joined: Feb 26, '24

Geolocation

Post by Metalhead33 »

I decided to re-train most existing modules from higher-quality sources.... that is, the original voices ran through Adobe Voice Enhancer, with some manual bass boost added by me. It will probably take over a week to train the new, HQ voices, but it'll be worth it.

I like Lotte.