✨│ai-help

[simple ore]

it has onnx in its name

[tight bane]

How can I look for a voice model?

[forest vector]

anyone know how to fix this ?
2025-07-03 22:02:27,643 ERROR [VoiceChangerManager] CUDA error: an illegal memory access was encountered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

running deitris' w-okada voice changer, on a gtx 1070.
happens at a random point in time, after 2-7 mins.

[simple ore]

running out of VRAM most likely

[forest vector]

I thought so too, but via task maneger it does not use more than 3/8 gb of dedicated memory

i feel like it might be inhibited by something, but cant tell why or how

[simple ore]

anything else running with hardware acceleration? discord / browser / some 3d game?

[forest vector]

discord, 1 browser with a video, and voicemeeter

[simple ore]

any overclocking/undevolting?

[forest vector]

none

[uncut eagle]

can i get help for voice changer ?

[outer frigate]

i want to thank every one who helped me it works

@simple ore @hallow thistle

[steady otter]

I used voice.ai it was good enough for what I needed

[past juniper]

do i just leave the embedder on default (hubert_base_112) i also see contentvec and whisper?

I have a 4060 and a AMD Ryzen 9 7950X3D

[kindred fern]

My mic is picking up the audio but it says "pipeline is not initliazied"

[keen coyote]

any specific settings that make a huge difference in the voice model? cant tell if the models bad or if its my settings

[smoky solstice]

Trying to run Deiteris' W Okada on an M1 Macbook Pro and getting the following error even after doing the proposed fix of using "xattr -dr com.apple.quarantine" to fix it. On Sequoia 15.2. Anyone have any ideas on what the isuse is?

[simple ore]

it is to disable quarantine for downloaded stuff

I think

"
This attribute is added so that it can ask for user confirmation the first time the downloaded program is run, to help stop malware. Upon confirmation the attribute should be removed automatically, and then the program will run normally.
"

[smoky solstice]

Sorry I'm a little confused. You're talking about the command to disable the apple quarantine right? I already did that through the "xattr -dr com.apple.quarantine" command they provided in the tutorial.

[simple ore]

okay

but you did not post anything else

[smoky solstice]

I think the issue might be an outdated MacOS version as a previous person has posted here but I'll post an update after ive updated.

[smoky solstice]

yeah that didnt work i fear 😞

[opaque scarab]

-colab

[patent trellis]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[knotty moth]

• chunk only affects the performance delay, if it's lower than what the gpu spec is capable of, it may cut off

• extra and "force fp32" in advanced settings may do fine grain quality improvement but dont expect if the model itself sounds not good

check some models in mvsep.com like some multi stem SCnet & BS roformer, then the drumsep models

[placid holly]

do i need download that 3 file (zip) for RTX 5000 W-Okada?

i only know 7z usually use 01 02 03, but dont know about Zip

[simple ore]

you need all 3 files, it is simply split to allow upload to github

[placid holly]

ok thank you 🤝

[viral mason]

it's peak

[fleet cedar]

how do i install the voice changer

i use an rtx 5070

[edgy topaz]

How do I make AI voice unnoticeable not like in real time like record in weights

[simple ore]

download the 5000 series version

[viral mason]

you don't use weights

[edgy topaz]

Oh what do I use?

[viral mason]

anything else :3

[reef yacht]

can someone tell me why is the voicechanger super delyed?

[hasty gust]

I had realtime voice changer client for 2 years now. is there a new update or a new client??

[viral mason]

what are ur settings in the voice changer, which download link did u use

[reef yacht]

watched a youtube video

[stoic bronze]

Hey, I’m using AMD and I’m wondering what I should be downloading. VCC is really acting slow on my pc and I don’t know how to fix this - it lags

Previously I downloaded a light and working VCC but I can’t remember where it’s at

[fleet cedar]

send link

[simple ore]

https://github.com/IllIlIlIllIl/voice-changer/releases/tag/b2335

Download all 3 files, then extract the .zip file, it will automatically extract ALL 3 FILES into one. Then open the MMVCServerSIO folder and run MMVCServerSIO.exe (or called MMVCServerSIO if you don't have extensions activated).

[viral mason]

send a screenshot of the settings u have

[edgy topaz]

What do I use?

[knotty moth]

bro left

[knotty moth]

what do u mean "in weights"

[viral mason]

https://tenor.com/view/nickola-viraj-gif-4864148932422811000

[edgy topaz]

So Like I use the voice in weights

[knotty moth]

then what do you expect of it?

[edgy topaz]

No, they say that I don't need to use weights

So what else do I do to not make Ai recorder or anything unnoticeable

[knotty moth]

nah I dont think that's the point

[knotty moth]

so you want the results sound like recorded in average mic?

there are some post processing effects you could do

so try searching it

[naive glen]

Hello, sorry for bothering and excuse my English but it is not my first language, but I have a question about how parrots are made in an alternative case with collab, why I tried to install kohya locally using pinokio, which did work but I don't know why an error occurs that I put all the parameters all the necessary folders to create a Lora but it tells me that the folder has not been found where one puts the images or that path does not exist I tried to do it in a thousand ways to verify that it existed And if it exists but it does not I know why it doesn't take it, so I don't know if anyone knows how to fix that error or in the worst case the truth is I don't know how to make loras in collab, why didn't I find updated links or links that currently work because at least all the ones I looked for gave me an error or something like that, so I would like to know if someone could help me or know something

[edgy topaz]

YASSSSSSSS

[earnest forge]

it could be, but now i have to figure getting that final voice into a game

[blissful lily]

Hey can you check my dm? I got a question about it

[steady otter]

Buddy atleast send the dm 😭

[blissful lily]

Sent

[pliant lily]

Can someone help me? whenever I have to mix two models, I get an error

[pliant lily]

but I tried, with two 48k models, I tested with several models

Can you send me links to some that work?

I've already tried running it locally and via Google Collab

[knotty moth]

it's because Applio treats "48k" ≠ "48000" due to models prob trained using different fork/version

so try using mainline rvc

[pliant lily]

mainline rvc?

[pliant lily]

gg

[hallow loom]

'VoiceChanger' object has no attribute 'resampler_in' what does this mean?

[wanton lion]

everytime i try to launch the start_http.bat file its crashes

[hearty dome]

guys which ai voice changer is good most ive seen are 1 year old are there any up to date ones

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

start_http.bat is apart of original wokada, ALL VIDEO TUTORIALS ARE OLD, DONT TRUST THEM, wokada deiteris fork is better

[low shard]

there's no updated video tutorial, only written guides, tell your pc gpu and what you want to do

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[glacial charm]

Hello i just want to make a pokemon song but i want to change the lyrics how can i do that

[jaunty marten]

Hello, would it be possible to make a voice like this?
https://youtu.be/JupFhvq36PA?si=ZGWF3U1_RYvYcOrq

[simple ore]

i imagine using something like ace-step with an instrumental track + new lyrics for audio2audio

[low shard]

You can search rvc ai voice models at:

• https://discord.com/channels/1159260121998827560/1175430844685484042
• In https://discord.com/channels/1159260121998827560/1163592055830880266 , Do /find with @earnest musk
• https://weights.com/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• Suggested Models for Realtime Voice Changing (Wokada)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• make it yourself with our docs guides
• Ask a free request in https://discord.com/channels/1159260121998827560/1159290139609137264
• Be aware that we don't allow any paid comms, so don't fall for any "pay me 20 dollars and i will make the model for you" dm

[earnest musk]

:wave: @low shard, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models

[edgy topaz]

Really?

What's the name?

[viral mason]

Just download it and use it in okada 💔

[edgy topaz]

Ohh I mean on mobile sorry!

I am buying an computer soon

[viral mason]

U can dm me if u want bc idk what you're talking about

[viral mason]

Ohhh you use mobile

[edgy topaz]

Yes

[viral mason]

Yeah you're pretty limited on options until u get the computer

[edgy topaz]

Oh I see

[viral mason]

I don't know of any websites besides weights to record your voice and have it output as an ai voice model

Sorry.-.

Maybe some helpers or mods know tho

[edgy topaz]

Oh is okay

[viral mason]

Or some QC (idk what it stands for)

[analog obsidian]

qc aren't helpers

[viral mason]

They're smart tho

Like uh

Noobies

Idk if they're even a QC lol

[analog obsidian]

he's a engineer

[viral mason]

Red or Blu side

What's his load out

[fair oasis]

guys may i ask: my regular mic works fine but my virtual cable mic somehow has my computer audio bleeding into my mic. its been messing up my voice ai as a result, what can I do to fix it?

[solid sequoia]

i have installed a model from #1175430844685484042 , no matter what i do though the model won't use my rx 6600 xt instead uses my cpu which kills the performance a lot any way i can fix this?

[toxic zenith]

How an I make a custom decadriver sound?

Like the announcer

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[low shard]

what? u mean model?

what's ur pc gpu?

[solid sequoia]

rx 6600 xt
windows 11
utilizing gpu for the model to perform better. right now it only uses cpu
can't upload screen shots due to missing permissions

[low shard]

you didn't elaborate everything

what do you want to do? what tutorial link are you using?

!give-meida-perms 1h @solid sequoia

[solid sequoia]

i want to use my gpu instead of cpu for the model so it performs better

and for tutorial link are you refering to youtube links?

[low shard]

i want to use my gpu instead of cpu for the model so it performs better

how do you want to use the model? what are you planning to do? ai covers? realtime voice changer for calls?

[solid sequoia]

realtime voice changer for calls and ingame voice chat

[low shard]

theres many different ai programs

RVC = Retrieval-based-Voice-Conversion, the best Few Shots Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.

Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)

do you need wokada or rvc?

[low shard]

yes, or the download link of the program

[solid sequoia]

i came from this tutorial, used the links given in the desc
https://www.youtube.com/watch?v=SxdnGxicJOg&ab_channel=novision

[low shard]

all video tutorials are outdated

that tutorial uses an over year old original wokada lmfao

I wrote it in the comments

you just wasted time using that tutorial basically

delete the program, and delete vb audio cable too

[solid sequoia]

it seemed to work perfectly fine though, i just need it to use gpu instead of cpu

[solid sequoia]

alright

which link can i find a newer version on

[low shard]

it's outdated, dont bother using it

plus that version sucks for amd

[low shard]

-realtime

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[low shard]

read the 1st link

[solid sequoia]

alright thank you

[low shard]

wokada deiteris fork

[low shard]

let me know

[solid sequoia]

it works so much better now both low cpu and gpu usage, more responsive as well thank you, 1 last question though my "echo" "sup1" "sup2" settings seem to be disabled i couldn't find anything about it in the audio setup page you sent me

[simple ore]

they are disabled in server mode

[pine star]

anyone had issues where they followed this guide https://docs.aihub.gg/rvc-voice-changer/realism/
it broke their youtube?

youtube just keeps saying "Audio renderer error. Please restart your computer"

[crude flame]

Have you tried restarting your pc

[pine star]

yeah it works whenever i dont have voicemeeter potato on

im pretty sure that i followed every step it told me

[crude flame]

Try clicking a1 on both the things in voicemeeter

[pine star]

like this?

i clicked on a1 for voicemeeter input and aux1

[crude flame]

And in the first one if that doesn't work

[pine star]

it still dosen work

i have light host on too, idk if thats the issue

[crude flame]

Huh that usually fixes it

[pine star]

should i remove the b1 and mono in stero input 1?

[crude flame]

Try turning off the denoiser you have on in the first column

[pine star]

wait it picks up sound

but youtube is still broken lol

[crude flame]

So it's picking up sounds just not outputting them?

[pine star]

yeah

i have my system input and out put my headset

[crude flame]

In hardware out a1 you have that set to your headphones right

[pine star]

yeee

wait how does the imput work?

cuz when i talk it dosent input anythign

i dont see any instructions to put my headset mic input anywhere

[crude flame]

Not in voicemeeter

In w-okada you use your mic input

[pine star]

oh so the wokada output will be the line 1 (virtual cable)?

[crude flame]

Then the output gets put into voicemeeter then into light host then outputs into discord or whatever

[pine star]

Like this?

"Once you have completed all of the above steps you can now go into anything and set the mic input to "Voicemeeter Out B2"."

[crude flame]

No

Input is your headphones

Wait

[pine star]

oh i got confused with the instructions

[crude flame]

I'm confusing myself lol

[pine star]

this is where i got it from

[crude flame]

So input is your mic output is line 1 and monitor you can leave empty

I'm going to redo that guide when I get home

[pine star]

oh ok

[pine star]

is it possible to add more details on what changing the setting actually do

so people know what they are actually changing

maybe have a bracket next to instuctions saying what it does

[crude flame]

Yeah def

[viral mason]

are these still the best settings for the t-de-esser 2

[pine star]

@crude flame just realised that when i downloaded virtual cable lite it automatically set everything to line 1 and thats why everything is breaking lol

apprently on windows 11, there system -> sound input output

and theres also system -> sound -> volume mixer input output

[viral mason]

that's confusing

[pine star]

yeah

so much inputs and outputs

[carmine mica]

program to use the templates?

[low shard]

elaborate

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[carmine mica]

for voice models to create TTS with rvc

[swift thunder]

Does anyone know of a software or method to remove audience noise and get results like this?
https://youtu.be/0xj7UiwVOa0?si=aHo1Igg_cCklyMHp

[steady otter]

yeah xminus has a decrowd feature

[swift thunder]

I tried it but it is not very stable, there is noise in the instrumentation

like that of Mvsep

[steady otter]

cant you replace it with the oficial instrumental?

[swift thunder]

But the drums and guitar? I'm impressed by the intro, since everything is identical to the original Live instrumental. I've been trying to get such results for a year.

https://youtu.be/0r3GpS1hy_Y?si=ucNGeRqmlDU0IsDX

or this

[brittle wing]

Hey i have a question, is there an AI tool for generating subtitles? I want to show something to my friend but he doesn't understand it cuz it's in my native language and not english

[crude flame]

updated the guide also found a way to not route system audio to voicemeeter so you dont have to deal with missing audio

i just tested it on a fresh version of voicemeeter so it should work

[simple ore]

you can run the audio thru ASR like Whisper

depending on the quality of the audio and language you may get something decent... or not

you can upload the audio to youtube and let me make a transcript

[pine star]

Thank yiu!

[fleet cedar]

i kept getting this error help

[simple ore]

norton antivirus?

[fleet cedar]

no

i dont have any antiviru

[simple ore]

weird

[brisk grove]

@simple ore what do u recommend for me to use for the ai voice in games?

[fleet cedar]

i already uninstalled a antivirus like 2 days ago

bitdefender

help

[fleet cedar]

when i click this "ok" it would say this error

[simple ore]

why are you copying?

[fleet cedar]

im not copying..

[simple ore]

you need to download all 3 files, use 7-zip to unzip

dont use windows BS

it is the worst

[fleet cedar]

not winrar?

can i use winrar

[brisk grove]

what do i download for the voice changer?

[simple ore]

winrar should be able to handle the split archive too

or winzip

[fleet cedar]

ill get winrar

or 7zip

@simple ore now it works after i installed 7zip

[brittle wing]

how do i fix

this

[brisk grove]

@fleet cedarwhat do u have that u put the voice to?

[fleet cedar]

wym

[brisk grove]

like the soundboard

to be able to use the voice

like voicemod

[fleet cedar]

@simple ore can u give me the right settings for my GPU

[simple ore]

https://rentry.co/forkvoicechangerguide

[brisk grove]

@fleet cedarwhats that?

[fleet cedar]

is this correct

idk

[fleet cedar]

bro cant u clearly see

its obviously the AI rvc voice

[brittle wing]

how to fdix

[brisk grove]

can u send a link for the download

@fleet cedar

[simple ore]

read the guide

[ionic oak]

Hello, I don’t know if this is the right place to ask, but I don’t know how to change my voice in real time. Which app should I use? Because I have several models that people sent me, but I don’t know how or where to use them?

[simple ore]

-rt

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[brisk grove]

@fleet cedaralr i got it setup now how can i get it to actually work in game and in discord?

[amber verge]

some reason it doesn't open

[dapper rain]

Is this channel for dedicated RVC assistance?

[low shard]

it's about any general ai help, rvc included

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[dapper rain]

Gotcha but you guys will still keep up RVC troubleshooting support still?

[low shard]

yeah ofc, we just merged the channels to have less channels

if you need help, pls elaborate

[plain pumice]

Yo

My voice changer is stuttering

It sounds so bad

Its like tweaking

My words are transfering but its static

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[plain pumice]

Oh

[low shard]

voice changer is too generic, there could be over 100 programs classified like that lol

you need to elaborate more to get help, else we dunno even how to help

[plain pumice]

XD

I cant screenshot here

Can I just dm?

[low shard]

!give-media-perms 1h @plain pumice

[low shard]

now u can

also it would be better u elaborate everything

all the infos i asked are crucial

[plain pumice]

So basically

[low shard]

lemme guess, youtube tutorial?

[plain pumice]

Im trying to use a voice changer on

Yes

[low shard]

you're using an over year old version of original wokada

its the same as using windows xp in 2050 basically

[plain pumice]

LOL

[low shard]

also vb audio cable has been reported to use issues on windows

[low shard]

all video tuts are outdated

[plain pumice]

How do I still use it

[low shard]

you can simply uninstall everything and forget you even watched it basically

[low shard]

simply, you can't, its a shitty version

you need a better one

[plain pumice]

Like?

Whats the new wakada

[low shard]

what's your pc gpu? what's your operative system?

[plain pumice]

Gpu?

[low shard]

that's crucial

[plain pumice]

Intel core i5

[low shard]

thats a cpu

cpu = central processing unit

[plain pumice]

Oh

[low shard]

gpu = graphics processing unit

this isn't chatgpt, it runs on your hardware and its way more intensive and complex

[plain pumice]

Oh

[low shard]

gpu does all the complex tasks like gaming, 3d, and AI

[low shard]

You can check your pc gpu on Windows via:
ctrl+shift+esc (task manager) -> Performance tab -> GPU

so yeah, don't expect a 1 click experience, AI is not really user friendly

chatgpt runs on everything bc it runs on cloud, remote good pc

while this is a program that runs on your hardware, not a rich hardware by someone else

[plain pumice]

Intel (r) uhd graphics

[low shard]

do you have any other gpus?

check gpu 1 and gpu 0

the one you mentioned is integrated graphics, it's literally too weak to do any type of AI and to even get recognized as a GPU lol

[plain pumice]

Oh

[low shard]

don't expect AI to run on bad hardware, it's more intensive than gaming

[low shard]

soo, you got another GPU?

[plain pumice]

How do I check?

[low shard]

just send a screenshot of your task manager

you should have gpu 0 and gpu 1 maybe

[plain pumice]

Alr

[low shard]

the whole task manager

[plain pumice]

Ok

\

[low shard]

the performance tab

[plain pumice]

I know I may be wasting ur time

(I definetly am but not on purpose)

Im just if you call it.. A caveman when it comes to checking ur pc and allat stuff

@low shard

[low shard]

yep you don't got got any other GPUs

You got 3 options:

• Buy a better pc
• Run it locally (on ur pc) using the CPU mode of the wokada fork which has better performance https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/ (but this isn't suggested as it could be unstable)
• Use **cloud **(remote good pc):

About Cloud, there are different services:

• Google Colabs (4 hours daily of free T4 gpu, easy to use, require only a google account) :
  • W-Okada's Deiteris' Fork Voice Changer Google Colab (currently works only on google colab PAID tier)
  • How to use Original W-Okada's Voice Changer Google Colab (has a Guide) (currently broken)
• Kaggles (30 hours weekly of better GPUs, T4x2 & P100, harder to use, requires an account and a phone number):
  • W-Okada's Deiteris' Fork Voice Changer Kaggle (the best and only working one currently for free)
  • Original W-Okada's Voice Changer Kaggle (currently broken)

so yeah that's why it was laggy, other than being an old version with worse performance, you also got bad hardware, so not a good combo

the best options are just either using cloud or buying a better pc

[plain pumice]

How do I use cloud?

[low shard]

click this link and read the guide

reminder that it's more complex and you got limited free time btw

and you also need to give your phone number, as it's a google service and they dont want you to use alt accs

[plain pumice]

Uj

Ik

Google is always like that

Like a really secure software

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

rvc and wokada do 2 different things

RVC = Retrieval-based-Voice-Conversion, the best Few Shots Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.

Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)

they do not want u to use alt accs lol

[exotic elm]

hey guys, how i use a model from the #1175430844685484042 ?

[low shard]

Elaborate:

• your PC GPU
• your operative system
• what you want to do

-realtime

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[low shard]

1st link, wokada deiteris fork

[long obsidian]

Anyone know realistic ai photo generator which can use your face to make a picture

[toxic zenith]

yeah, I mean custom text to speech, I see decade driver models here but i don't know how to convert text to speech into my style

[simple ore]

https://github.com/VectorSpaceLab/OmniGen2?tab=readme-ov-file

[long obsidian]

Thank youu

[long obsidian]

Also when creating images of people do they look genuinely

And can i feed it data to create pictures from it

[runic thunder]

hii

how can I create an ai voice?

[long obsidian]

@simple ore and also which out of these do u think is the best omnigen2/fluc/SDXL

[raven barn]

literally have no idea how to train with a 5070

it just doesnt wanna work

[toxic zenith]

how can I make my own text to speech?

[knotty moth]

make sure you're using the latest Applio

and do manual install with latest pytorch (2.7) and cuda 12.8

the original rvc and even mangio won't work at all

[raven barn]

i did that

like so many times

i dont know what i did wrong

[knotty moth]

a bit correction

if not the latest release

clone the Applio repo itself https://github.com/IAHispano/Applio

then just double click run-install.bat

it should include torch 2.7.1 which is needed for RTX 50-series

[raven barn]

okay ill try this

[long obsidian]

@knotty moth can u check dms for a sec i asked u a question there i can send it here too if u want

[knotty moth]

no just ask here

[long obsidian]

Do you know a good ai photo generator which can use a face to generate pictures

I saw flux and onnigen2 are good ones but which would you recommend me

[raven barn]

do u mind sending links for those, i dont want download the wrong thing

just incase i may have

[raven barn]

no for the cuda and pytorch

i think i found it

[toxic zenith]

where can I train AI voice using google colab

[long obsidian]

@knotty moth so what do you think

[raven barn]

currently trying it rn

downloaded pytorch 2.7 and cuda 12.8

didnt do anything

proof

idk but im tempted on just getting out my 4070 and using that

cuz ik it will work

is it more or less the same thing?

ill look into it when i wake up

been trying to get this to work for ab 10 hours

rvc doesnt work

appolio doesnt work

i hope but thx

[simple ore]

to do image edits, like replacing characters/merging images, both omnigen2 and flux kontext

[simple ore]

full screenshot how you installed it

[long obsidian]

For creating people or changing small features (making ai influencer) which would u suggest me to use

[simple ore]

it also requires a small fix to add "50" to "infer-web.py"

[simple ore]

omnigen2 then

all in one

[simple ore]

you're messing something up then

[raven barn]

Can I dm u the screenshots tmr I just got off

[simple ore]

https://github.com/IAHispano/Applio/issues/1020

[raven barn]

That’s exactly what I did

I got that line straight from their website

Even installed cuda 12.8 or wtv

[simple ore]

i require a screenshot of how you done it

because there are 20+ who said it works fine

[raven barn]

The only version of pyhton I could get to work with that version of PyTorch was 13.11.9

[silver lynx]

hey is there still like a list of what settings to use with which gpus

[simple ore]

https://rentry.co/forkvoicechangerguide#known-working-settings-for-chunk-and-extra

[raven barn]

I’ll send it to u tmr

[silver lynx]

seems like a bad website, for amd 6xxx XT gpu's it says its MAX settings are 128 + 2.7s and then like a few sentences below the tabel it says the 6650 XT can do 60-80 ms

[simple ore]

read the section above

[knotty moth]

torch 2.7.1 exactly

you were trying on 2.7.0

[raven barn]

Oh

[knotty moth]

if you run through run_install.bat it should have installed torch 2.7.1

[raven barn]

I found the hugging face version

I’ll look at it again when I wake up later

[simple ore]

does not matter, both work

RVC1006Nvidia requires a small manual fix

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[hybrid flame]

Hello, can you please advise if there is an up to date guide to install and use RVC on a pc with AMD graphics card (7800xt) for real time voice changing? Thanks

[limber siren]

Hi, did Weights remove the option to log in with github?

[hybrid flame]

I don't understand, since I'm not a developer, I'm just here to use it.

[knotty moth]

get the AMD one here

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#download-amd-intel-and-cpu-on-windows

[hybrid flame]

thanks

[willow cipher]

why does echo and suppresion can't be turned on?

nvm figured out

[long obsidian]

@simple ore hi i got a question about the installation my gpu is nvidia 5060ti and my cuda version is 12.9 on the pytorch the newest one is 12.8 but the flash attention doesnt support the pytorch version of 2.6.0 with this cuda version what should i do? Should I download the 12.4 one and the standard flash attention? (this is about the omnigen2), will it work like that despite my gpu being newer version

[simple ore]

you dont need cuda toolkit

as long as you have torch cu128 either 2.7.0 or 2.7.1 it is fine

you can find flash attention 2 wheels here https://huggingface.co/lldacing/flash-attention-windows-wheel/tree/main

download cp version that matches your python install (3.10, 3.11, 3.12)

[long obsidian]

@simple ore so i need to install with this: pip install torch==2.7.1+cu128 --index-url https://download.pytorch.org/whl/cu128 and install one of the flash attn versions corresponding to my python (3.11) uve listed?

[simple ore]

you install torch and torchvision pip install torch torchvision --upgrade --index-url https://download.pytorch.org/whl/cu128

then you download cp311 flash wheel and use pip install the_name_of_the_downloaded_file.whl

[long obsidian]

i just found out that the python version ive downloaded is 3.13 if i download 3.11 will it work?

[simple ore]

nope

3.13 is not recommended for anything, too many incompatible libraries out there

[long obsidian]

how can i swap it to 3.11, just delete it?

[simple ore]

install 3.11, make sure you select 'add to path' checkbox on the install screen

[long obsidian]

alr imma do this rn thanks

[simple ore]

then remaking the virtual envrionment using 3.11

py -3.11 -m venv venv

[long obsidian]

there are multiple versions of 3.11 (after the 11 which version should i get)

[simple ore]

https://www.python.org/ftp/python/3.11.9/python-3.11.9-amd64.exe

[long obsidian]

ty im downloading the stuff rn

[long obsidian]

is it normal to be this slow for a simple image

also @simple ore how can i train models for it (can i actually do it)

im seeing this for the past 5 min

[simple ore]

did you make sure you have cuda torch installed?

check device manager/performance / memory and vram use

[sinful mango]

Hello all !
I am a newbie here and not a developper at all, I dabble a bit and mostly just surf, read, and do lots of trials & errors.
I am currently working on mods for Cyberpunk 2077 for my private usage (not for sharing on nexusmods or else for copyright issues).
I saw a lot of "voice ai" swaps for the main character and wanted to create my own with a voice actor I really appreciate (a french dubber for a character in a tv show).
I recently tried Zonos to create TTS audio with a sample of the french dubber and the result is quite good.
But that is just the beginning, now I am in front of the hardest part :
Take all the audio files of the character in the game , and create modified audio files of those source files but with the cloned voice I get on Zonos.
And so I have two options :
Either I get all the text of those audio files and script something with python to batch generate the audio files using Zonos.
Either I find a tool allowing audio-to-audio by using a cloned audio reader (is that even possible and does it exist ?)
My configuration is : i9 14900 / 64gb ram / RTX 4090 / Windows 11 Pro

Any help/pointers would be deeply appreciated ^^ (and I repeat : I am not a developer, I can dabble and am willing to learn but consider me an utter noob)

[long obsidian]

ughh i followed ur steps

from here i have done the steps till 2

then i deleted the pip files and followed ur instructions

i didnt do the 3.2 tho

i will try doing that rn

[long obsidian]

should i change --upgrade here or leave it like this

also this is the requirments txt file

torch==2.6.0
torchvision==0.21.0
timm
einops
accelerate
transformers==4.51.3
diffusers
opencv-python-headless
scipy
wandb
matplotlib
Pillow
tqdm
omegaconf
python-dotenv
ninja
ipykernel
wheel
triton-windows; sys_platform == "win32"

[simple ore]

install the requirements, then upgrade torch

[viral mason]

Hey there, I can help you create a voice model of the person you're talking about ^^

[long obsidian]

i changed the text file to

torch==2.7.1
torchvision==0.22.1
timm
einops
accelerate
transformers==4.51.3
diffusers
opencv-python-headless
scipy
wandb
matplotlib
Pillow
tqdm
omegaconf
python-dotenv
ninja
ipykernel
wheel
triton-windows; sys_platform == "win32"

and i will check if it works now

i copied the torch and torchvision from the cmd from the installing step before it

its still sitting on 0/50

and doesnt move

can u type step by step what i should do to fix this (sorry if im being to annoying

[simple ore]

You likely have cpu torch

reinstall cu128

[long obsidian]

with this command? pip install torch torchvision --upgrade --index-url https://download.pytorch.org/whl/cu128

[simple ore]

activate the environment

then run that

[long obsidian]

okey

it says this

[simple ore]

or venv\scripts\python -m pip install torch torchvision --upgrade --index-url https://download.pytorch.org/whl/cu128

i dunno why you're using conda

[long obsidian]

should i delete it

and start anew

[simple ore]

regular python venv

[long obsidian]

and if i delete conda and the file itself (conda create -n omnigen2 python=3.11
conda activate omnigen2) i change the conda part here to venv?

[long obsidian]

oh okey

and after making this enviroment do i have to activate it or just go on?

[simple ore]

after that is done, pip install torch torchvision --upgrade --index-url https://download.pytorch.org/whl/cu128

and pip install flash.whl

[long obsidian]

i will try that rn

[sinful mango]

Would love that ! I'm all ears !

[viral mason]

Alrighty I can help out in dms since these two are doing nerd stuff 👍

[toxic zenith]

how can I make custom text to speech

I mean using custom models, not pretrain models

[viral mason]

I think they're talking about something like uberduck

Like how it had TF2 tts

[long obsidian]

i did what u told me and it still stays like this

[toxic zenith]

yeah, for text to speech

i want to make my own sound on decade driver

[viral mason]

I dunno how the rvc works I just know how to clean datasets and how to read graph

Trust

Whoever is typing RN your name is all rectangles lmao

It's scary

I am proof that anyone can make a voice model as long as they try

[analog obsidian]

TTS are zero shot, no dataset needed, only a few seconds
rvc requires training, 10 mins minimum for increased consistency

fun fact: rvc core component (vits) is 'hacked' in order to do speech to speech conversion instead of tts

yup actually rvc first "name" was just vits

[toxic zenith]

ok

so how can I create an AI voice model?

rvc pls

[analog obsidian]

https://docs.aihub.gg/essentials/how-to-make-voice-models/

[long obsidian]

@simple ore i tried reinstalling and following your order again but it stays the same just 0/50 0%

[hallow thistle]

You mean RVC voice model and not TTS? Alright, there are ways to train a voice model.

[analog obsidian]

https://www.bilibili.com/video/BV1A14y1a75R/ this is probably the very first rvc model ever made, when it was internally just named "vits" (since rvc is a modified vits)

[simple ore]

check venv/lib/site-packages folder and see what torch is installed

[long obsidian]

torch-2.7.1+cu128.dist-info its this

rn im trying running it without the flash-attn to see if it will work

[simple ore]

would it be easier for you to just use comfyUI?

[tribal cosmos]

for live voice changers, is w okada still the best or is there something new?

[simple ore]

-rt

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[long obsidian]

idk how to set it up and also still doesnt move without the file

i will try rq to reboot pc and see if it will work

[tribal cosmos]

??

[simple ore]

read the guide above

[tribal cosmos]

oh thank you

[simple ore]

https://docs.comfy.org/tutorials/image/omnigen/omnigen2

[long obsidian]

can u also send a guide for comfyui

[tribal cosmos]

whats the difference bteween the top one and the 2nd one?

[simple ore]

Wokada Deiteris Fork

use that

[tribal cosmos]

okay thx

[hallow thistle]

Deiteris' fork W-Okada and "original" W-Okada are both versions of W-Okada, but they are developed separately by different authors. Deiteris' W-Okada is better. What are differences? Performance and some bug fixes, especially.

[tribal cosmos]

oh i see thanks

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

[tribal cosmos]

does anyone know where the download button is i cant find it 😭

[toxic zenith]

ok i know

how about tts

[hallow thistle]

Don't.

[hallow thistle]

What is your PC GPU? NVIDIA GeForce or AMD Radeon RX?

[tribal cosmos]

nvidia 3080

[hallow thistle]

https://cdn.discordapp.com/attachments/1159290139609137264/1371371778181431328/image.png

[tribal cosmos]

thank you

[toxic zenith]

ok

may be i try to tts in another app

[tribal cosmos]

should i uninstall my old w okada??

[toxic zenith]

then use rvc to emulate decadriver

[hallow thistle]

Yes.

[toxic zenith]

I think none of u all know about kr decade

[hallow thistle]

Yes, you guessed it right. I have no idea what Kr Decade even is.

[toxic zenith]

I tried to recreate custom decade driver sound for my decade driver

[hallow thistle]

While Applio has TTS feature built-in, it's edge-tts, the RVC itself isn't TTS.

[toxic zenith]

by training models using sounds from csm toys

and tv series using UVR5

[hallow thistle]

Sound driver? I only know Intel/Realtek HD Audio and Creative Sound Blaster as sound card drivers.

[toxic zenith]

no

"toy" sound ok

[hallow thistle]

[toxic zenith]

i don't mean audio driver fr

i mean changing sound in my decade driver bootleg toy model

[tribal cosmos]

do i have to run any of these bat files?

or do i just straight do the exe

[viral mason]

The exe

[tribal cosmos]

oh ok thx

also

how do i uninstall my old one

i had w okada

like 2 years ago

[viral mason]

Just delete all files related to it

[tribal cosmos]

thats it??

[viral mason]

Should just be in that folder unless they got replaced by the new stuff

[viral mason]

Yup

[tribal cosmos]

when i run the new exe is it js gonna replace the old one?

[viral mason]

I don't think so

[tribal cosmos]

oh wtf

[viral mason]

Pretty sure the files would've already been replaced by default unless you chose to skip or smth or didn't get that option after extraction

[tribal cosmos]

oh i mean like the acutal w okada

[viral mason]

Wdym

I'm slow

:3

[tribal cosmos]

nahh its okay lmaooo

yo i have another quesiton

[viral mason]

Yah?

[tribal cosmos]

i havent doen any ai voice stuff in years

but when i was making models

i used to use appolio

is there a new one ppl are using now or is it still apollio

becuase i feel like alot has probbaly changed

[viral mason]

There's applio and mainline

And local rvc stuff which idk anything about

But applio still exists ya

Nothing new I know about

[long obsidian]

@simple ore i tried doing the comfyui one u suggested but im always getting this error

[simple ore]

read the link, download the files, place them into right places

[long obsidian]

i did that

[tribal cosmos]

ohhh ok thanks

[long obsidian]

here for an example

[tribal cosmos]

is this bad?

[viral mason]

Uhh

[simple ore]

sweeet mother of god

[viral mason]

Try running it again

[tribal cosmos]

okk imma try

[long obsidian]

what 😭

[simple ore]

works fine

[simple ore]

with the files placed into right places

[long obsidian]

i have same settings same stuff

[silver lynx]

oh right for this, im kinda stupid but the highest amd card on it is 7xxx XT

im getting a 9070 XT, so im wondering what applies to that

[tribal cosmos]

okay i think it finished, is it supposed ot open in ur broswer right?

[viral mason]

Yup!

Btw if u wanted we could move the convo to dms

[tribal cosmos]

ohh okayyy

[viral mason]

Ye

[long obsidian]

@simple ore is it possible for me to screenshare for us to fix the omnigen2 local version in vc here?

pls i wanna kym at this point 😭

[simple ore]

cd OmniGen2

py -3.11 -m venv venv

venv/scripts/activate

pip install -r requirements.txt

pip install torch torchvision --upgrade --index-url https://download.pytorch.org/whl/cu128

pip install https://huggingface.co/lldacing/flash-attention-windows-wheel/resolve/main/flash_attn-2.7.4.post1%2Bcu128torch2.7.0cxx11abiFALSE-cp311-cp311-win_amd64.whl

python inference.py --model_path "OmniGen2/OmniGen2" --num_inference_step 50  --height 1024 --width 1024 --text_guidance_scale 4.0 --instruction "The sun rises slightly, the dew on the rose petals in the garden is clear, a crystal ladybug is crawling to the dew, the background is the early morning garden, macro lens." --output_image_path outputs/output_t2i.png --num_images_per_prompt 1```

[long obsidian]

i will try it now

[long obsidian]

its still sitting on 0

[simple ore]

show vram use

add --enable_model_cpu_offload parameter to inference call

[simple ore]

that parameter should help

[long obsidian]

let me try this, so i add it at the back of this python inference.py --model_path "OmniGen2/OmniGen2" --num_inference_step 50 --height 1024 --width 1024 --text_guidance_scale 4.0 --instruction "The sun rises slightly, the dew on the rose petals in the garden is clear, a crystal ladybug is crawling to the dew, the background is the early morning garden, macro lens." --output_image_path outputs/output_t2i.png --num_images_per_prompt 1

[simple ore]

at the end

[long obsidian]

alr

[long obsidian]

its working now, will it work if i do python app.py --enable_model_cpu_offload

[simple ore]

yes

can try the other too

[long obsidian]

can u send it

as code

[shadow orbit]

help

mic is working but i can't hear myself

on okada voice changer

:((

[forest vector]

for some reason, the "2025-07-03 22:02:27,643 ERROR [VoiceChangerManager] CUDA error: an illegal memory access was encountered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.
"- error is fixed when I run a game of sorts on the same PC.

[forest vector]

you mean, you cannot hear yourself, as in your real voice or your ai-converted voice ?

[simple ore]

python app.py --enable_model_cpu_offload

and/or --enable_sequential_cpu_offload

[shadow orbit]

both

no sound at all

when i hit passthru i hear myself tho

[forest vector]

probably the voice isnt converting

[shadow orbit]

how do i fix that 😭

i sent a screenshot of my options in #1192011222023950368

[long obsidian]

thanks it works perfect now

[hushed thicket]

how do i add a voice model

[low shard]

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

This is a general AI server, we can't know which program you're talking about, so that's why we need more information on the questions I asked you please

[hushed thicket]

oh i downloaded the realtime voice changer and a voice that I want and im confused on how to add it

is there a tutorial video?

[low shard]

please reply to the questions I asked you, there's thousands of different programs

and your pc gpu and operative system is crucial too

[low shard]

also, there's no updated video tutorial for realtime voice changing, they mostly use old programs, did you follow one?

[hushed thicket]

yea i followed a old one

[low shard]

if you followed a video tutorial, you can just delete everything you got off it, you probably got original wokada like version 1.5.3.8 and vb audio cable

AI runs at sonic speed, youtube tuts aren't the best for ai programs, they get outdated easily

[hushed thicket]

4070, windows, use a voice, https://www.youtube.com/watch?v=We5oYpCR3WQ, I cant upload images

[low shard]

it's best you also forget everything they tell you in it, they also tell outdated info like using "crepe"

[low shard]

yeah, that youtube tutorial is outdated asf lol

also, we don't endorse anything that duckus does

duckus is an horrible person

[hushed thicket]

ooh i didnt know that

i just wanted a simple tut

[low shard]

duckus describes himself as a "certified catfisher"

he makes money off catfishing people for the pure fun of it

AI should be used for good and fun, not for catfishing

[hushed thicket]

yeah i agree

I want to use it to troll my friends and some randoms

[low shard]

-realtime

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[low shard]

read up the 1st link, wokada deiteris fork

this is the only updated tutorial

wokada deiteris fork got various improvements in performance and quality

[hushed thicket]

ok i installed it

and the virtual cable

[low shard]

nice, let me know for any issues

[hushed thicket]

how do i input a voice

[low shard]

like add a model?

[hushed thicket]

yeah

[low shard]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#adding-models

also, if you share a screenshot of your wokada, i can help you with settings

[shadow flax]

can someone give me a simple video on how to install RVC :,) ?
(NVIDIA, WIN11)

[low shard]

please tell me your pc gpu and what you want to do

[shadow flax]

5070ti, use realtime voice changer (RVC), the part as to how i can use it in games/calls i know myself i just cant find the right RVC to download

[hushed thicket]

everytime I download on the thing it says to create a covere

[low shard]

realtime voice changer (RVC),
Yeah that's why I asked what you want to do, RVC doesn't mean that, it means Retrieval-based-Voice-Conversion

RVC = Retrieval-based-Voice-Conversion, the best Few Shots Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.

Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)

I think what you want is actually wokada deiteris fork, right?

[low shard]

share the link of the model you're trying to download

[hushed thicket]

https://www.weights.com/models/cmc8m0p4p00uon915fz7qukh3

[shadow flax]

oh sorry seems i was mistaken then, i just remember having used it on AMD like a year ago from "okada".
whichever you think fits best ill get then, heres a picture of the UI i remembered

[low shard]

that one is a pretty outdated version of original wokada, wouldn't be suggested anymore

-realtime

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[low shard]

read up the 1st link, wokada deiteris fork

it has better quality and performance

and it supports the rtx 50 serie

[shadow flax]

oh thats cool, never seen it in tutorials like you said. thanks

[low shard]

because all youtube tutorials are outdated lol

AI moves at sonic speed, dont trust yt tuts for everything

[hushed thicket]

https://www.weights.com/models/cmc8m0p4p00uon915fz7qukh3

[low shard]

make a weights.com account, click the 3 dots, then download

[hushed thicket]

it still says select a model first after I uploaded the pth

[golden walrus]

Sir Nick. Can i ask if there is a way to compress my voice effectively? Cuz when i use Wokada or Vonovox, discord will "bonk" it and make it pretty robotic and unnatural somehow

[low shard]

show a screenshot of your wokada

[golden walrus]

I have EQ band and compressor enable

[low shard]

click close
click the model slot
click start

[golden walrus]

[shadow flax]

do i need to download all 3?

[low shard]

also, you didn't set up the settings, show an entire screenshot

[low shard]

like, there's cracking?

[golden walrus]

ye, sharp decline in quality

[low shard]

yes

[analog obsidian]

bitcrush

[golden walrus]

i know it has something to do with the bitrate

[hushed thicket]

what settings do u recomend?

[golden walrus]

on it, sire

[low shard]

chunk: 80ms
extra: 2.7
f0: rmvpe
input: microphone
output: line 1
monitor: headphones, optional to hear urself

[golden walrus]

wait, is it the same with bad mic effect?

[mystic sky]

What is the most reliable or update version of this AI in real-time?

[analog obsidian]

yea, bitcrushing simulates how the audio would sound if it were of low quality

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[golden walrus]

do voicemeeter have it built-in? or do you have any recommendation for the external app? i really hope to get a cleaner voice in low quality

[analog obsidian]

idk i dont use filters in realtime

[golden walrus]

[analog obsidian]

but the docs recommends one app

1 sec

[shadow orbit]

whaaa

[analog obsidian]

https://docs.aihub.gg/rvc-voice-changer/realism/
this

[golden walrus]

ah Kiloheart

[analog obsidian]

yea just get kiloheart bitcrusher

[golden walrus]

hmmm, i hope i can listen playback so i can adjust the stat to get cleaner

[shadow orbit]

can someone help with models upload..

[shadow orbit]

that thing comes up when i upload anything from #1175430844685484042

[shadow flax]

its not letting me extract the zip 😭

[golden walrus]

damn, the playback sound so good but discord just bonk me hard

[shadow flax]

whats this even meaaan 😭

im actually gonna crashout

why can i unzip like everything but not this particular zip

dw i got it extracting using a different tool

[viral mason]

that simple

no need to read

just

do that

in fact

that video should be pinned

it's easy

videos like these should be in the server for simple people like me

[raven barn]

i already have it installed

but it still doesnt work

[analog obsidian]

how many models u have lmao

[viral mason]

uhhh

[viral mason]

lemme look

[shadow flax]

whys nordvpn telling me this aint safe 😔 the WebUI

[simple ore]

looks good

[viral mason]

you replied to a bot lol

[simple ore]

altough I dont like your ( ) folder

[shadow flax]

i knoe

[raven barn]

whats wrong with it

[viral mason]

currently 115 models

[shadow flax]

where can i find the command line?

[analog obsidian]

[viral mason]

https://cdn.discordapp.com/attachments/936087908631343105/1278441206451273748/what_did_bro_expect.gif

[raven barn]

when i try to launch appolio it says this

ive never had this issue

[shadow flax]

whys everything not worky

idk if i leak my IP by sending that or not lol

[viral mason]

let's go to this guy's house and fix it for him personally

[raven barn]

okay i got it to work

[simple ore]

it tell you what the problem is

[raven barn]

i fixed it

[simple ore]

the other window with a bunch of text

[raven barn]

is this even doing anything wtf

[simple ore]

show the preprocess step screenshot

[simple ore]

you did not add models or something

[shadow flax]

hmm i thought i did

[raven barn]

all my wav and audio files i just put in there

[simple ore]

remove "

[raven barn]

both of them?

or just hte last one

[simple ore]

or just use a different path c:\training\data

put your files there

[shadow flax]

why does it not want to take the model? (does it have to be in the actual WebUI/app directory?)

it literally just does not want to take the model

its even in the model -> 0 something directory

[simple ore]

close the app, find the model folder, and nuke it

[viral mason]

why does it say diddy

[simple ore]

illegal combination is using WASAPI input and MME output

use both WASAPI

[shadow flax]

delete?

[raven barn]

Heh

[shadow flax]

how do i know whats what

[simple ore]

input and output devices

use both WASAPI type

[shadow flax]

btw what rest of the settings should i use?

also how can i delete the "saved" audio

[long obsidian]

if i have 32k sample voice can i use text to speach and save the wav file to extract the voice and train it to higher samples?

[simple ore]

tts are usually 24k at best

[long obsidian]

is there a way i can increase the samples of a voice then

[long obsidian]

if i upload a wav file to applio which has a little bit of background sounds from games lets say will it do the extraction good

[simple ore]

just use 32k audio

no, use a proper denoise / vocal extraction

[long obsidian]

whats the best app that can isolate podcast sounds and make it pure voice audio

and whats the best app to download youtube videos as audio files

[brittle wing]

how can I fuse 2 models in applio ?

[narrow sun]

I have so much ms can someone help me

[neat grove]

ok so will gtx 1650 sup + ryzen 3 3100 will be good on w-Okada cause i hear a lot of background even if i put steelseries gg mic or when i talk on my native language i don't think it's trained on my language cause like there is some word that ai can't say it and it will be obvious to anyone that it's ai

[forest vector]

voice changers only fool a very small portion of the active gaming population at best

maybe if you do male to male conversion, and u already sound a bit like the model itself

[pine star]

where do you guys usually get sources from to train ai for rcv?

[neat grove]

when you have like a good device you can change small thing and no one will notice, ofc the best will be like the creative sound blaster SB0490 but these getting like hella expensive

[neat grove]

even if i do i will have some weird ai sound like crashing lol

[forest vector]

what setting do you do your crossfade at

[torpid turtle]

hii all

just found a perfect live walpaper but its not well looped

you can clearly see that the vid replays each time

is there an ai tha can help with this?

[shadow flax]

could someone share with me what the optimal settings are for Wokada Deiteris Fork?

[viral mason]

kaggle applio keeps doing this

it's pissing me off

[neat grove]

like ?

[forest vector]

id say 0,12-0,15s

depending on the results u get

[umbral frigate]

You sure?

[forest vector]

pretty sure, or everything I heard till now wasnt the best there is

[umbral frigate]

Yeah, i've trolled many times and if I give them a heads up that its a voice changer they will pick up on it but 90% of the time it goes undetected lol

Tbf i do think the model sounds realistic

[forest vector]

are breathing, and other misc sounds also realistic ?

[umbral frigate]

Yeah normal speech is very realistic you can kinda hear the breath too ig?

But anything other than that is cooked LOL

[forest vector]

ahh ... okay thought I missed something

[umbral frigate]

So in my experience ive learned to just adapt to it

[forest vector]

that helps alot ye

[umbral frigate]

Bc ive been using the same one for a while i kinda know how to speak with it

[neat grove]

ill send you photo so i can know whats your meaning, i'm not the best in eng but i can understand atleast very well

[forest vector]

if you dont mind, id like to hear a short sample as im curious

[umbral frigate]

No problem, but im trying to get it to work rn for some reason my whole thing stopped working

[forest vector]

stage fright huh

[umbral frigate]

Nope im fine with sharing it but it wont go through voicemeeter

[crude flame]

retruthed

[forest vector]

I was referring to the voice model itself lmao

[umbral frigate]

Oh wdym

[forest vector]

ehh ... still dont think that it is true, it sounds obvious to me up until now

[crude flame]

rvc is very easy to spot

[analog obsidian]

obviously people that dont know about ai will not immediately fall for this, but give them a couple of minutes talking with u and they're gonna spot its rvc quickly

[pastel oak]

gtx 1650 is ok, not great but ok, depends on your use case
dont bother trying to fool anyone in any language other than english its too obvious
"i hear a lot of background" if you mean noise then its microphone related, your mic is picking up every sound you make or that can be heard in the background

[brittle wing]

does anyone know if this is a good guide for rx that I should follow for my models https://rentry.co/RVC-dataset-RX11#spectral-denoising-the-audio

[analog obsidian]

that was literally how i learned about rvc, some random dude was talking with me and i started to notice his voice was weird

[pastel oak]

i can giggle and moan with ai u guys got nothing on me

[umbral frigate]

Idk why but my microphone works fine but it wont convert my voice everythings silent

[torpid turtle]

chat

help please

[umbral frigate]

Like wdym giggle

[torpid turtle]

just found a perfect live walpaper but its not well looped
you can clearly see that the vid replays each time
is there an ai tha can help with this?

[umbral frigate]

like hahaha?

[neat grove]

yeah, make sense for the language i think

[crude flame]

vewy easy

[forest vector]

if its rvc id reckon it sounds really obvious

[low shard]

This is a General AI Server, we won't be focused on voices anymore

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[umbral frigate]

an indicator is if the mic is too good ngl

[analog obsidian]

both sounds very ai to me

[umbral frigate]

ive been told if you make the mic shittier it sounds much more believable

[low shard]

yt-dlp

[analog obsidian]

younger me would also notice it's ai

[crude flame]

yea

but you can just bitcrush and that solves the problem

[crude flame]

well one is real

[umbral frigate]

or that yea

[low shard]

https://docs.aihub.gg/rvc/resources/dataset-isolation/

[umbral frigate]

Its just placebo dude

[analog obsidian]

i mean the first one yeah

[umbral frigate]

Shes speaking like a voice actor

[analog obsidian]

i only heard the second one and i assumed the first one was also rvc but unedited lmao

[crude flame]

the one with the dash at the end?

[brittle wing]

is this the best guide to rx I should follow https://rentry.co/RVC-dataset-RX11#spectral-denoising-the-audio

[forest vector]

the breath gave it away on the first one

[neat grove]

for better result for no one to notice just make the base low

[analog obsidian]

never mind, both are rvc

[crude flame]

no

[umbral frigate]

LOL

[crude flame]

one of them is real

and dont skip to the laugh

that will make it to easy

[forest vector]

if the 2nd one was rvc id say we actually should gib up trying to recoqnize

[pastel oak]

Most likely yes

[analog obsidian]

okay the second one is real

sorry i cheated

XD

[brittle wing]

ok ty

[crude flame]

😡

smhhh

[analog obsidian]

just look the spectograms lmaooo

[crude flame]

wrong

second one is real

[pastel oak]

dunno if you pay attention to it like i do but can tell the 2nd is the real one based on the pronounciation of "I have" in the beginning

and if the first one is real then i question my existence

[analog obsidian]

people have to learn rvc can't be realistic no matter what u do

[forest vector]

if one needs to take one of these out, then id say we can fool 95% of the population

[analog obsidian]

it didnt fool me, the thing is i didnt bother in hearing them closely

razer always share that audio

[crude flame]

skip to 36 sec and listen

[forest vector]

as with most conversations

[analog obsidian]

rvc is 2023 tech

its not a voice cloning ai

what the ai is trying to do is literally reproducing mel specs and pitches

is not trying to clone expressions or shit because its not meant for that

all results are flat asf

[crude flame]

vits2 rvc when?

[analog obsidian]

good rvc models can fool the more casual side of the internet

but they will find out it's ai, rvc will glitch in any moment

[forest vector]

If I were you, I wouldnt bet on that

[analog obsidian]

i have been training models since 2023

[low shard]

when they make a good discord competitor

[analog obsidian]

i know whats inside rvc, i know how it does the conversion

and i know it cant do everything

[forest vector]

maybe not that specific way u have in mind it cant, I agree.

[crude flame]

pretty much anything non verbal rvc cant do

[analog obsidian]

the embedder is trash

[forest vector]

but somewhere down the line I believe it will become hard to recoqnize

[crude flame]

eventually yes voice cloning will be that good but now its not

[forest vector]

now its ... meh

[analog obsidian]

i agree with this, flat inferences are usually the most realistic sounding because those don't have expressions, so rvc doesnt struggle

[umbral frigate]

How would they even go abt that

[forest vector]

but really impressive compared to what we had back in the day

[crude flame]

no one cares about ai audio so nothing is happening to it 😔

[forest vector]

but when I hear real voice actors im like "what am I doing 💀"

[analog obsidian]

we do have real voice cloning ai

[narrow sun]

Yes i have bad pc.. i dont have gpu

[analog obsidian]

but they're tts

[crude flame]

better embedder, vits2, better pretrains, better GAN/vocoder

[umbral frigate]

elevenlabs?

[analog obsidian]

eleven, chatterbox, yeah they're real voice cloning ai because the ai is actually learning and reproducing expressions

rvc doesn't learn expressions

[umbral frigate]

So you mean if elevenlabs became like sts?

Would be pretty nuts

[analog obsidian]

when you give rvc an audio, it'll extract the mel spec and the pitch data alongisde the features of it

then it'll try to reproduce them afterwards

[analog obsidian]

no idea

[umbral frigate]

Is there even an estimate for how far in the future until we get updates on rvc

[analog obsidian]

rvc is SOTA

[umbral frigate]

SOTA?

[analog obsidian]

like the best sts

[umbral frigate]

like no updates

[analog obsidian]

the reason why it doesn't learn emotions is because is using a pretty old architecture named vits

[crude flame]

the og dev team left rvc to do tts

[umbral frigate]

Why tf

[analog obsidian]

because tts are superior

arch wise

they can learn emotions and non verbal sounds better

[umbral frigate]

yeah but that means sts is neglected

if everyone on rvc went to tts

[crude flame]

no one cares enough about sts to update it

[umbral frigate]

i think sts is cool

[analog obsidian]

updating sts it's a really hard task

[analog obsidian]

me too

rvc interally it's a hack of a tts architecture actually

rvc-boss took vits, and did a couple of changes in order to "convert" it to sts

if we remove those changes, it'll be regular tts vits

[umbral frigate]

So if there were to be strides in sts would they have to build it from the ground up?

Or can rvc as it is rn be improved

and its just shitty architecture

[analog obsidian]

yea no the arch is too shit to be updated

[umbral frigate]

yeah damn

[analog obsidian]

some devs of here have tried