We have a Steam curator now. You should be following it. https://store.steampowered.com/curator/44994899-RPGHQ/
[Resource] RVC voice modules
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
[Resource] RVC voice modules
-- Reserved for later.
Last edited by Metalhead33 on April 19th, 2024, 22:09, edited 6 times in total.
Tags:
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
--- Reserved for later
Last edited by Metalhead33 on April 10th, 2024, 06:58, edited 3 times in total.
- maidenhaver
- Posts: 4389
- Joined: Apr 17, '23
- Location: ROLE PLAYING GAME
- Contact:
does it has text to speech function? the rvc i mean.
Last edited by Red7 on April 3rd, 2024, 18:16, edited 1 time in total.
What's the Epoch on each of these?Metalhead33 wrote: ↑ April 3rd, 2024, 14:14These voices were trained form Medieval Total War and Shogun Total War:
- Catholic Commander & Narrator
- Orthodox Commander & Narrator
- Muslim Commander
- Muslim Narrator
- Engrish Voice
- Engrish Voice 2
- Japanese Voice
- Mongol Voice
These voices were trained from Stronghold Crusader, with the exception of the last:
- The Advisor
- The Scribe
- Stronghold 1's Original Narrator, Military Campaign
- Christian Narrator
- Christian Mission Control
- Muslim Narrator
- Muslim Mission Control
- The Caliph
- The Emir
- The Nizar
- The Sultan
- The Wazir
- The Pig
- The Rat
- The Snake
- The Wolf
- Richard Lionheart
- Saladin
- Lord Woolsack
- Sir Longarm
- Chinese Narrator from Emperor: Middle Kingdom
These voices were trained from the Hungarian localization of Stronghold Crusader:
- The Advisor
- The Scribe
- Christian Narrator & Mission Control
- Muslim Narrator & Mission Control
- The Wazir
- The Pig
- The Rat
- The Wolf
- Richard Lionheart
- Saladin
All of them can also be downloaded at this link.
I'd love to see the face on the sweet baby types as they listen to this..
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
200
No. You have to combine it with an actual TTS.
https://github.com/RVC-Project/Retrieva ... sion-WebUI
Last edited by Metalhead33 on April 4th, 2024, 07:33, edited 2 times in total.
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
Testing out the Stronghold Crusader Christian Narrator. The TTS is from NovelAI.
Last edited by Metalhead33 on April 4th, 2024, 08:20, edited 1 time in total.
- Nammu Archag
- Posts: 1070
- Joined: Nov 28, '23
- Location: Tel Uvirith
Do you have a good primer for using RVC?Metalhead33 wrote: ↑ March 29th, 2024, 11:53Here are some AI voices I trained in RVC, mostly with 60 epochs each:
Baldur's Gate 1-2 characters:
- Aerie from Baldur's Gate 2
- Ajantis Ilvastarr from Baldur's Gate 1
- Alora from Baldur's Gate 1
- Anomen Delryn from Baldur's Gate 2
- Baeloth Barrityl from Siege of Dragonspear
- Branwen from Baldur's Gate 1
- Cernd from Baldur's Gate 2
- Clara from Baldur's Gate 2
- Coran from Baldur's Gate 1
- Dorn Il-Khan from Baldur's Gate 1-2
- Dynaheir from Baldur's Gate 1
- Edwin Odesseiron from Baldur's Gate 1-2
- Eldoth Kron from Baldur's Gate 1
- Elminster Aumar from Baldur's Gate 1
- Faldorn from Baldur's Gate 1
- Garrick from Baldur's Gate 1
- Glint Gardnersonson from Siege of Dragonspear
- Haer'Dalis from Baldur's Gate 2
- Hexxat from Baldur's Gate 2
- Imoen from Baldur's Gate 1-2
- Jaheira from Baldur's Gate 1-2
- Jan Jansen from Baldur's Gate 2
- Kagain from Baldur's Gate 1
- Keldorn Firecam from Baldur's Gate 2
- Khalid from Baldur's Gate 1
- Kivan from Baldur's Gate 1
- Korgan Bloodaxe from Baldur's Gate 2
- Mazzy Fentan from Baldur's Gate 2
- Minsc from Baldur's Gate 1-2
- M'Khiin Grubdoubler from Siege of Dragonspear
- Montaron from Baldur's Gate 1
- Nalia de'Arnise from Baldur's Gate 2
- Neera from Baldur's Gate 1-2
- Quayle from Baldur's Gate 1
- Rasaad yn Bashir from Baldur's Gate 1-2
- Safana from Baldur's Gate 1
- Sarevok Anchev from Baldur's Gate 1-2
- Schael Corwin from Siege of Dragonspear
- Shar-Teel Dosan from Baldur's Gate 1
- Skie Silvershield from Baldur's Gate 1
- Tiax from Baldur's Gate 1
- Valygar Corthala
- Viconia DeVir from Baldur's Gate 1-2
- Voghiln from Baldur's Gate 1: Siege of Dragonspear
- Yeslick from Baldur's Gate 1
- Yoshimo from Baldur's Gate 2
- Xan from Baldur's Gat 1
- Xzar from Baldur's Gate 1
Others:
- Amamiya Sora, Japanese voice actress and singer
- Itou Miku, Japanese voice actress and singer
- Uesaka Sumire, Japanese voice actress and singer
- Asuka Kasen from GTA 3
- Tifa Lockhart from Final Fantasy 7 Remake
- PS1 Hagrid
- Some random dude who made a parody of my OC, very low-quality
Alternatively, the full album
No need to credit me, if you use these AI voices anywhere.
The Edwin voice was already tested out atDynaheir Test:Metalhead33 wrote: ↑ March 26th, 2024, 18:05So, today, I set up RVC and went ahead to alter my voice into Edwin's voice.
My real voice:
Edwin voice transformations:
Depending on how many Tav lines are there, I could make an attempt to be the "voice double" for wannabe-Edwin Tavs.
And yes, this is my natural accent.
Another test run was performed at:
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
I used the default settings for everything, except for GPU batches and epochs (60 epochs for the Baldur's Gate 1-2 characters, 200 for everyone else).
Otherwise, I'm not sure what you mean by "primer". If you are looking for a good TTS, I recommend NovelAI - it allows you to download the generated speech, which can then be transformed with RVC.
Last edited by Metalhead33 on April 4th, 2024, 08:22, edited 1 time in total.
- Nammu Archag
- Posts: 1070
- Joined: Nov 28, '23
- Location: Tel Uvirith
RVC is voice-to-voice out of the box right?Metalhead33 wrote: ↑ April 4th, 2024, 08:21I used the default settings for everything, except for GPU batches and epochs (60 epochs for the Baldur's Gate 1-2 characters, 200 for everyone else).
Otherwise, I'm not sure what you mean by "primer". If you are looking for a good TTS, I recommend NovelAI - it allows you to download the generated speech, which can then be transformed with RVC.
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
Yep. It is. You need to provide it with a voice input: either your own voice recordings, or something generated by a TTS.Nammu Archag wrote: ↑ April 4th, 2024, 08:29RVC is voice-to-voice out of the box right?Metalhead33 wrote: ↑ April 4th, 2024, 08:21I used the default settings for everything, except for GPU batches and epochs (60 epochs for the Baldur's Gate 1-2 characters, 200 for everyone else).
Otherwise, I'm not sure what you mean by "primer". If you are looking for a good TTS, I recommend NovelAI - it allows you to download the generated speech, which can then be transformed with RVC.
- Nammu Archag
- Posts: 1070
- Joined: Nov 28, '23
- Location: Tel Uvirith
Thanks, I'll just try it out again. For whatever reason I failed to get it to work the last time I attempted a couple months back, though it may have just been due to my own error.Metalhead33 wrote: ↑ April 4th, 2024, 08:43Yep. It is. You need to provide it with a voice input: either your own voice recordings, or something generated by a TTS.Nammu Archag wrote: ↑ April 4th, 2024, 08:29RVC is voice-to-voice out of the box right?Metalhead33 wrote: ↑ April 4th, 2024, 08:21
I used the default settings for everything, except for GPU batches and epochs (60 epochs for the Baldur's Gate 1-2 characters, 200 for everyone else).
Otherwise, I'm not sure what you mean by "primer". If you are looking for a good TTS, I recommend NovelAI - it allows you to download the generated speech, which can then be transformed with RVC.
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
Oh, and the .pth files go into the assets/weights folder, while the other file goes into the assets/indices folder.Nammu Archag wrote: ↑ April 4th, 2024, 08:50Thanks, I'll just try it out again. For whatever reason I failed to get it to work the last time I attempted a couple months back, though it may have just been due to my own error.
I suggest you go 500 on bg1 and 2 voice lines. There's too few lines.
Could be why your Edwin sounds a bit off
Could be why your Edwin sounds a bit off
Last edited by orinEsque on April 4th, 2024, 10:02, edited 1 time in total.
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
I think that might have more to do with either my microphone, or the way I speak. I noticed that even when I use a module trained with 200 epochs, some weird things happen, like my Ms turning into Bs, or the Ss being muted, regardless of the AI module I use. The same issue does not happen when I feed a TTS's output to the RVC AI.
Maybe I have a cold without knowing it?
Last edited by Metalhead33 on April 4th, 2024, 10:59, edited 1 time in total.
I've had issues getting some enunciations to work even with input lines from Native English speakers as well.Metalhead33 wrote: ↑ April 4th, 2024, 10:59I think that might have more to do with either my microphone, or the way I speak. I noticed that even when I use a module trained with 200 epochs, some weird things happen, like my Ms turning into Bs, or the Ss being muted, regardless of the AI module I use. The same issue does not happen when I feed a TTS's output to the RVC AI.
Maybe I have a cold without knowing it?
Had to rerecord multiple times to get the right alphabet sound sometimes as well T.T
Last edited by orinEsque on April 4th, 2024, 11:34, edited 2 times in total.
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
I think that might have more to do with either my microphone, or the way I speak. I noticed that even when I use a module trained with 200 epochs, some weird things happen, like my Ms turning into Bs, or the Ss being muted, regardless of the AI module I use. The same issue does not happen when I feed a TTS's output to the RVC AI.
Maybe I have a cold without knowing it?
[/quote]
I've had issues getting some enunciations to work even with input lines from Native English speakers as well.
Had to rerecord multiple times to get the right alphabet sound sometimes as well T.T
[/quote]
I noticed that - regardless of the AI module I use (60 epoch BG1-2 vs 200 epoch TW/SH) - it works well with clean recordings taken from video games, TTS outputs, as well as.... this.
Versus using my own voice as an input:
Tfw my English was Engrish all along.
Last edited by Metalhead33 on April 4th, 2024, 11:43, edited 1 time in total.
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
Semi-related, but when training for AI, the audio could be pre-cleaned with this: https://podcast.adobe.com/enhance
Haven't used it before, but in a video, the results are impressive:
Haven't used it before, but in a video, the results are impressive:
You definitely need more epochs. The conversions are too tinny.
Last edited by orinEsque on April 4th, 2024, 13:54, edited 1 time in total.
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
I'll re-train some of the BG 1-2 ones tomorrow.orinEsque wrote: ↑ April 4th, 2024, 13:53You definitely need more epochs. The conversions are too tinny.
- rusty_shackleford
- Site Admin
- Posts: 10741
- Joined: Feb 2, '23
- Gender: Watermelon
- Contact:
"enhanced"??Metalhead33 wrote: ↑ April 4th, 2024, 12:23Semi-related, but when training for AI, the audio could be pre-cleaned with this: https://podcast.adobe.com/enhance
Haven't used it before, but in a video, the results are impressive:
- Metalhead33
- Posts: 319
- Joined: Feb 26, '24
Cleaned up.
- rusty_shackleford
- Site Admin
- Posts: 10741
- Joined: Feb 2, '23
- Gender: Watermelon
- Contact:
"cleaned up"???
It sounds far worse.
- Oyster Sauce
- Turtle
- Posts: 2229
- Joined: Jun 2, '23
You going to train AI on a voice sample smothered with static?rusty_shackleford wrote: ↑ April 4th, 2024, 15:36"cleaned up"???
It sounds far worse.
- rusty_shackleford
- Site Admin
- Posts: 10741
- Joined: Feb 2, '23
- Gender: Watermelon
- Contact:
If I wanted it to sound like how those actually sounded, yes.Oyster Sauce wrote: ↑ April 4th, 2024, 15:40You going to train AI on a voice sample smothered with static?
- Oyster Sauce
- Turtle
- Posts: 2229
- Joined: Jun 2, '23
rusty_shackleford wrote: ↑ April 4th, 2024, 15:45If I wanted it to sound like a fan-voiced Skyrim mod, yes.Oyster Sauce wrote: ↑ April 4th, 2024, 15:40You going to train AI on a voice sample smothered with static?
- rusty_shackleford
- Site Admin
- Posts: 10741
- Joined: Feb 2, '23
- Gender: Watermelon
- Contact:
Did you listen to the video? It doesn't even sound like the same person.Oyster Sauce wrote: ↑ April 4th, 2024, 15:46rusty_shackleford wrote: ↑ April 4th, 2024, 15:45If I wanted it to sound like a fan-voiced Skyrim mod, yes.Oyster Sauce wrote: ↑ April 4th, 2024, 15:40
You going to train AI on a voice sample smothered with static?
- rusty_shackleford
- Site Admin
- Posts: 10741
- Joined: Feb 2, '23
- Gender: Watermelon
- Contact:
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 shit without scanlines.
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.
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.