✨│ai-help

[wise fiber]

this has been running

for

several miniutes

20 now

[torn nebula]

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Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[torn nebula]

yo

which one of these the download 🥀

-help

fuckass bot

[viral mason]

I think you should do -read

[torn nebula]

-read

[viral mason]

[torn nebula]

fuck you

💔

[north falcon]

guys does anyone have tried creating realistic pictures? i tried and it wasnt realistic that much. Can somebody help me ?

[viral mason]

I meant actually read it lmao

[torn nebula]

i cant understand english

[viral mason]

you're literally typing it

[torn nebula]

gng i have no clue what the -realtime is telling me to do

js act as if

i have 3 braincells

combined

and explain what i should do

[wise fiber]

no ones helpign me with this

💔

[brazen ore]

bruh that other kid doesnt help at all he just says random bullshit

[north falcon]

guys does anyone have tried creating realistic pictures? i tried and it wasnt realistic that much. Can somebody help me ?

[viral mason]

ok so what's your gpu, open your task manager and click preformance

[wise fiber]

oi

[viral mason]

if you're on phone I'm gonna dissapear

[wise fiber]

hv any idea

why its tkaing so long

[torn nebula]

i have a 1050ti

nvidia

[viral mason]

cool

[north falcon]

can somebody help me 😂

[viral mason]

Imma send u the download and stuff and help u out on installing it and using

dms ok?

[torn nebula]

i believe i have an outdated voice changer and i want my friend to install it aswell since he got a pc upgrade, he currently has AMD though and he installed one which is old through a very old yt video so like i wanted to find the lateest one for both him and me

[viral mason]

I don't know anything abt image generation, u can keep asking tho :3

[viral mason]

I only know the link to the nvidia compatible one so he's kinda out of luck for now

I could help u tho

[torn nebula]

alr then

-amd

[fossil sundial]

😠

-nut

-manat

[viral mason]

do not nut on me

[tall vortex]

i cant hear the voice changer and dont know if its acc working or not

[viral mason]

send a screenshot of it so I can see if you have the mic settings set up correctly

[tall vortex]

oki

[wise fiber]

@viral masonbro

how tf

do i do that

thing

its STILL

been downlaoding

[torn nebula]

@low shard gurt

yo

where the thingy at

for amd

-realtie

-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

[fossil sundial]

maybe if youre lucky 🤷

[viral mason]

[wise fiber]

@viral mason

augh

its

still going

wtf do ido

[viral mason]

Idk 😭

[grand yarrow]

Guys, I'm looking for an AI that can take a rulebook with Card Outlines that I created and create hundreds of Card images designed for the strategy and rules I created because this is a tedious task for game design.

[native cedar]

what does dis do

nvm no pic perm-

s

the extra chunk thing

[low shard]

i just tried reinstalling UVR too and got no issues, you sure your windows and drivers are up to date?

also are you sure you don't just wanna use cloud which is faster than your pc and easier?

[low shard]

it has finished running.. there's the green check

also, the original wokada collab doesn't work, and the wokada deiteris fork colab does work but only in the paid tier, if you try it in the fre tier you risk your google acc to get banned

[low shard]

tell your operative system and what you want to do

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)

[low shard]

@viral mason isn't an helper, tho if he wants to help he should actually help

[low shard]

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)

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]

This is a General AI Server, AI has many fields

Elaborate:

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

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)

[native cedar]

ye nvm

[low shard]

you have to elaborate, AI is a complex and intensive task, and be sure to not use video tutorials since they are old

if you're looking for a 1 click program, open source AI isn't like that, this is a community driven program, not a product from a company

[native cedar]

i already said nvm ima just keep the voice extra chunks as u told me before 2.7

[low shard]

when did we talk?

[native cedar]

u told me how to install remember

[low shard]

@native cedar did you use an alt acc? if you send me the username i can check our old chat and help u up

[native cedar]

the browser voice changer

[upbeat prawn]

How do I que a job?

It says I’ve reached my qued job limit

But I haven’t used my weekly yet

[low shard]

that's called wokada deiteris fork

[upbeat prawn]

It says so at the top

[native cedar]

is it the best version

[low shard]

#✨│ai-help message
chunk: controls the delay, should be higher than the perf value while running
extra: controls a bit quality, higher than 2.7 can cause cutoff issues

[low shard]

wokada deiteris fork b2332 which is the one you have, yes

[native cedar]

alr ty

[low shard]

do you need any other help

also wdym u dont have access to it anymore?

[native cedar]

jus lost it

[upbeat prawn]

I’m confused, can you train one free weekly model or not?

[low shard]

if you're talking about weights.com yes

if you have any issues

ask in their server, not here

discord.gg/weights

[upbeat prawn]

Ok

[pseudo swift]

if 40k is the best sample rate for my dataset what settings should i use for exporting in audacity because they dont let you do 40k

[heavy crest]

how do you want the ss

the folder just has something that says start_https and another folder with alot of stuff in it

[viral mason]

I help with stuff I understand

But ye I'm not officially a helper

[viral mason]

I think 32k is cooler :3

[solemn harness]

guys pls i get this whenever i use an RVC voice model RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

[low shard]

start_https.bat is a thing from old original wokada, all video tutorials are outdated

elaborate the rest i asked u

[low shard]

This is a General AI Server, AI has many fields

Elaborate:

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

[heavy crest]

i have a 5070

windows 10

[flint vapor]

What error you have when you open okada?

[heavy crest]

i double click on the start thing

and it shows a window of something opening

[low shard]

what do you want to do? realtime voice changer for calls?

[heavy crest]

but insta closes

[flint vapor]

Do you have some external antivirus ?

[heavy crest]

yea the voice changer for discord and games i play

[low shard]

the version he got is outdated, i remember you're an ex helper but i don't think you're aware of the newer version

[low shard]

never follow yt tuts for that
they all use an over year old original wokada

-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 frok

don't bother using that version you got rn, it's old

[flint vapor]

Yes I didn’t never use new version I still have previous I’m stuck at 2024 hahah

I have to check all

All new stuff

[low shard]

better check the docs https://docs.aihub.gg

[pseudo swift]

already started a 40k train but i guess i'll try 32k after that

[lavish viper]

Meow I want to ask for help
I don't know, I got this answer before, if someone answers, please mark it.

[viral mason]

Ask the question

[heavy crest]

do i click link to github

and download that

[low shard]

No, you read the guide, the download is literally below

[heavy crest]

i

dont see it

only think i see it tells me to download is nvidia

where is the download

[flint vapor]

Are you on hugginface now?

[heavy crest]

im on this thing he sent

i clicked the first here

[flint vapor]

Click 5000 series

In blue

[heavy crest]

i mis typed earlier

i have 4-70

4070

also

dont i alr have nvidia

[flint vapor]

If you have 4070 click the normal windows version for nvidia

Is up

[languid granite]

hello!! im very new to using the voice changer and im just wondering if there are in guides that will work for mac?

[simple ore]

that error means one of the files with the models (.pth) is damaged - incomplete download most likely

[solemn harness]

Should i reinstall VCC CLIENT?
Or the voice models

[simple ore]

depends on where the error happens (see the console output)

[solemn harness]

I tried multiple voice models they didn’t work ill try to reinstall vcc client

Thank u for the help mate!

[simple ore]

i mean show the fking log

probably rmvpe.pt or something, just delete what you have in internal folder and it should redownload

[flint vapor]

Have you tried with some other voice model ?

[solemn harness]

Yeah alot

[flint vapor]

So I think instead of upload .pth you upload the zip

Extract the pth from the zip

And upload only the pth

[simple ore]

could be

unless they show the log

although with the new ipad generation @flint vapor has 80% chance to be right

[viral mason]

ipad kids scare me

[flint vapor]

Yes if he send to log we can clearly understand which is the real problem

I think he uploaded the zip

Because it was say failed reading zip archive, but I don’t know which client is he using

[simple ore]

pytorch models are .zip too

[flint vapor]

Oh yes true but I afraid maybe he uploaded directly the zip file

Because idk which client he using and how he is using

Because he said he used many voice models so I think is something about his client or he didn’t upload correct

[brazen ore]

well the thing is that i have a model for uvr, so i dont know if i would be able to load that into the cloud version

like a custom model, not one of the presets that it alr comes with

[sage sinew]

does anyone know the link for the free trial with open ai

nm

[native cedar]

@crude flame u should make a voice model of dis guy in yt "Not6ixid " he has many clips of his voice hella deep

[viral mason]

how good is it?

[crude flame]

Why me?

[native cedar]

ur insane yk

[native cedar]

ur the best ever creator

[crude flame]

I'm not that good

[native cedar]

ur insane

[crude flame]

I'm above average but I'm not that good

[native cedar]

how long does it tke to make a model

[crude flame]

Depends on my motivation

This last model I made and dumped took like a week

I couldn't get good sibilants so I gave up

But in general a model can take a day to like months

Codename's best model took him 4 months

[native cedar]

whos codename

[crude flame]

A goated model maker

But he left

[native cedar]

fr?

[crude flame]

Drama

[analog obsidian]

making a model is tricky because randomly rvc doesnt wanna learn the dataset/struggles for no apparent reason

[crude flame]

Yeah

[native cedar]

his models he posted are gone to?

[crude flame]

Yeah 😭

[crude flame]

He never posted models

[native cedar]

oh

[crude flame]

But he was good at it

[native cedar]

so ur the best creator rn

[crude flame]

No

[native cedar]

who is

[crude flame]

Idk prob lyery

[viral mason]

me

[native cedar]

what models u got

[viral mason]

I was joking-

[crude flame]

I haven't heard many models from people so IDK

[analog obsidian]

it's all subjective tbh

[crude flame]

^

[analog obsidian]

sure there is skill involved but most of the time the results are super unpredictable

[viral mason]

u can check them on my weights page

[native cedar]

whts ur page

i found a very good corpse husband model from weights

[analog obsidian]

he doesnt make egirl models

idk if razer still does realtime models

[viral mason]

bc I'm cool

[crude flame]

I do but they aren't e girl models

[viral mason]

I could copy it if the weights page wasn't shittings itself when I try copying it, just look for Claptrap on weights and find mine

[crude flame]

They are now normal girl models

[analog obsidian]

rip e-girl tag

[crude flame]

Rip all my models 😭

[native cedar]

make super hot latina ones

jokes

[viral mason]

we need more big buff oily men models

[native cedar]

we need super deep hot voice men models

[crude flame]

😭

[analog obsidian]

you can have the best dataset in the world, that you spend a week cleaning it
and rvc may still hate it lmao

[crude flame]

You are like the third person I've seen that wants guy models

I've been here since like 2023

[analog obsidian]

deep voiced man model pls realistic

[viral mason]

best I can do is ss it because weights will not let me copy 😭

[crude flame]

German deep man voice

[native cedar]

i alys wanted deep voice

[crude flame]

Real pls

[native cedar]

like corpse and eboy

u provided a eboy one so thats fire

[viral mason]

pls latina german deep voiced eboy trump model

[crude flame]

You are the second person to say that one eboy model i made was good

[opal haven]

Hey, sorry if this is a silly question... but does the Deiteris fork of W-Okada start offline after you've DLed the files?

In the mainline release, if you were to try starting it without internet, it'd say something about failing to DL models and just not work.

[analog obsidian]

could be a leftover of the original version, which for some reason, required internet connection (sussy)

[native cedar]

mak a corpse model 5 hours of the voice

[viral mason]

www.weights.com/user/your_local_worm

got it

[sharp shore]

anybody know arch linux virtual audio cable?

or really just linux

[native cedar]

how do i stop the mubling

for ai voice rvc webbrowser one

cus noice suppresion or wtver is so bad makes it super bad

[cinder canopy]

do the later versions of wokada still have the rvc quality high setting? ik the difference isnt noticeable but intonations sound beter and i have resources for it

also im on an older version of wokada, if i switched from nvidia to amd gpu, would it still work or do i have to switch versions

[analog obsidian]

im looking the code of the original w-okada and it seems what the rvc quality settings does is to repeat the inference two times, so you speak, the model produces a result, then inferences that result to give another output
that is... weird
there's no way to improve a model's quality, but what you can do in the fork is to enable fp32 inference mode, it's hard to explain in simple words what fp32 is but, in theory, it should reduce the amount of artifacts you get in the result

in the latests versions of the original w-okada i don't think you can enable fp32 without manually changing the code

in the fork u can do it in the interface/gui

besides that, increasing the extra to 5s can technically give you the highest quality realtime can offer

but if the model is already bad, it wont make it any better

[cinder canopy]

the model im using is good, i agree that the rvc quality pipeline is a bit weird but words sound more natural and theres like "more emotion" in it.

idk how else to describe it

i guess ill stick to the older version of it for now, does it work with amd gpus?

[analog obsidian]

because its easier for the model to clone its own voice

it works but ive heard its super laggy and choppy

[cinder canopy]

alright

ill just stick to watever i have now

thanks for the info tho

[cinder canopy]

link for future reference?

[analog obsidian]

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

[cinder canopy]

ty

[supple glacier]

Hey guys, what does Timbre Leakage actually look like in a spectrogram and why are whispering voices so hard to make D:

[analog obsidian]

timbre leakage is not something you see in the spectogram but something you hear
it's when the voice of the model sounds off and extremely similar to the original audio used for the inference, so let's say you make a model of your fav character, if your model has leakage everytime you try to inference something, the model is going to have a voice similar to the source audio rather than the intended voice

and whispering voices are hard to make because whispering is noise data and rvc doesn't do great with noise

it tries to average the whole dataset into a unique voice, so if u got 50% of normal speech and 50% of whispering, the model is going to try to "blend" the whispering and the normal speech together, making the model sound very weird after training

[analog obsidian]

if you trained your model with the og pretrain and contentvec you can just set the index a value to maybe 0.5~ to fix the pretrain leak

but if you used spin instead, then it cannot be fixed, its a weird inconsistent bug

[supple glacier]

@analog obsidian interesting, thank you very much, you are doing a great job!

[analog obsidian]

[supple glacier]

another question, i'm trying to understand how to interpret tensorboard graphs. i've seen you guys chatting (codename;0, Yannenou, noobies and you) about several graphs and clipping.. is there some guidance with examples? not especially about clipping, generally

[native cedar]

how do i stop the mubling
for ai voice rvc webbrowser one
cus noice suppresion or wtver is so bad makes it super bad

[violet rose]

hey

what is the best software for ia voice changer ?

okada ?

or other

[analog obsidian]

keep in mind while reading the graphs:

grad norm d = low values and going down
grad norm g = below 300 and going down, i personally noticed if grad g is over 400 the model has always some random problem
d/total = going down
fm = going up (this is the only metric that should go up)
mel = going down (very important)
kl = going down
g/total = going down

[supple glacier]

thats pin-worthy, isn't it? 🙂

[analog obsidian]

grad clipped and such were testings when grad clipping was a thing in rvc, later it was found it actually never worked as intended

xD

[supple glacier]

lol. how the turn tables

[analog obsidian]

if you're using the new exp_f0 spin branch, there is a new loss that i cant remember the name lmao, but that also goes up alongside fm i noticed

i mean its not "new" it was always there, just hidden

every of them are important but imo, keep an eye in the grad norm g since if thats high, the chances of the model having some random problem are increased

[crude flame]

adv/g

[analog obsidian]

that!

[crude flame]

Adversarial generator

[analog obsidian]

then dont use the noise suppresor

[native cedar]

then it mubles like crazy

[analog obsidian]

rip

[analog obsidian]

yes

[supple glacier]

i have literally no idea what you are talking about, the whole terminology is still new to me, f0 okay, ground frequency - pitch related, grad i guess gradient - norm, normalization? the learning curve is kinda steep i was playing around first of all and now i start digging

[analog obsidian]

these are your grad norms

gradients

they're in the tensorboard

so ideally you want them to go down and chill

with low values like that

if they're always going up and having crazy high values, something is wrong with the dataset, or the batch size may be too low

[analog obsidian]

they're noisy and fluctuate (moving up and going down constantly), thats normal

as long they're not super high everythings gonna be fine

[supple glacier]

this doesnt tell me anything about the learningprogress itself, right?

[analog obsidian]

they do

[supple glacier]

i mean yeah, if something is off

[analog obsidian]

none of the metrics can tell u when your model overtrained exactly

or tell u which one is the best epoch

[supple glacier]

how do you encounter that, going for the top x out of y lowestest values at the given epoch and running several examples through UTMOS?

[analog obsidian]

ow that is not implemented in rvc

ive asked for some validation losses before

there are other stuff more precise than val loss as well that vits used in their paper

[crude flame]

Codename's fork has some 😎

[supple glacier]

i've implemented val/loss in seed-vc, pytorch makes it easy

[analog obsidian]

yea there is already val loss in code's fork and noobies also did his own val loss train.py

it's better than nothing imo

the regular losses are kinda a meme since they dont help at all

[supple glacier]

@supple glacier that was the intention behind the question

[analog obsidian]

i see i understand now

hmm yea obviously adding validation stuff would be awesome

they're far superior to rvc's current metrics

the rvc metrics dont really tell u anything lmao

[crude flame]

Chat is this overtraining?

[analog obsidian]

YES

[crude flame]

Y

Line go down

[analog obsidian]

peak

train more

[crude flame]

I train 100 epoch

When overtrain

[analog obsidian]

btw @supple glacier we have tested batch sizes in rvc, we've found that you can use bigger batches in datasets just fine

batch 64 works fine in a 10 min set for example

[crude flame]

Why my 30 second e girl dataset sounds bad?

[analog obsidian]

gives u a smoother convergence and the model see less noise through the training

[supple glacier]

if i save every 100 epochs a checkpoint, does rvc use exactly the trained parameters at the specific epoch, or does it look -+ epochs to the left and right and averages that?

[analog obsidian]

if u got more than 10 mins, yeah use the highest batch ur gpu can handle

thats a good question i wanna know too lol

ik rvc saves the exact parameters at that epoch, allowing to stop the training, and continue training later

about the other thing, 0 clue, noobies probably knows xD

[crude flame]

It uses that epochs parameters

Imo doesn't make much sense to avg the epochs before and after

By avg those you can be making the model worse

[analog obsidian]

yea that makes sense to me aswell

[supple glacier]

lines = checkpoints, circle the epoch i would like to check out the most, right?

i cant, since no checkpoint

[analog obsidian]

lowest point in the graph is usually noise

[crude flame]

Save every epoch

[supple glacier]

your ssd controller and cells hates that trick

[analog obsidian]

i actually got the best epoch in the moment g/total started to rise forever

the pain truth is
the only way to know which epoch is the best is by hearing them

[supple glacier]

hah, interesting

[crude flame]

Also sometimes for no good reason RVC will just hate the voice you're trying to train and it will never sound good

[analog obsidian]

yeah the loss metrics are really that bad haha, we really need some validation added

[crude flame]

Ik from a lot of personal experience

[crude flame]

Codename's fork

Have you tried them yet?

[supple glacier]

due to the source-speaker maybe or in general?

[crude flame]

I have no idea why but it will randomly just not like a voice and no matter what you change it won't be good

I wasted a week trying to get a model RVC didn't like to sound good

[analog obsidian]

im gonna check them real quick

[supple glacier]

i've encountered a similar issue with a voice, the person itself sounded like a robot..not in a electric way/distorted way, just monotonous, no dynamic/depth.. i mean how do you make a robot make sound more natural haha

[crude flame]

This model was the model I was going to try hard on 😭

[analog obsidian]

interesting, i got this exact problem when i tried training a model using the old crepe with custom hop

[crude flame]

But nooo

[crude flame]

😭

[analog obsidian]

the model was not able to do any pitch changes

that happens when you get out of vram while doing feature extraction

it generates a f0 file with 0 information in there

so weird

[supple glacier]

indeed

alright, gotta go, thanks for the chat.. just starting with this whole stuff.. lets see where the journey goes... starting work in 4 hours 😦

[analog obsidian]

bye

[crude flame]

8gb of vram problems 😎

[crude flame]

You seem promising, I think you'll do great and be great!

[gusty cedar]

hey guys, so I followed the github steps to get OPENVINO with A1111 Stable diffusion, and even watched a video from Intel themselves and... I am getting an issue of my SD using the CPU instead of the GPU

[pine star]

is 56min of dataset too long?

[crude flame]

thats a good amount

some would dare say its not enough for realistic results

[pine star]

another question, if your dataset is monotone, will the voice model also be monotone no matter how you say things?

[crude flame]

yea

pretty much

it wont be able to generalize well

[pine star]

oh so even if you try not to sound monotone, it will just sound weird?

[crude flame]

yea

[pine star]

gotcha

[native cedar]

why is ma ping so high

[pseudo egret]

What is the best real-time AI voice program currently available that supports custom model uploads?

[young halo]

Does anyone have a tutorial for making a model with applio colab?

[stoic ember]

chat, does anybody know a community where i could get help w llamaindex

[candid terrace]

does anyone have the E-women voice file??

[flint vapor]

I have a 3060 and I handle good rvc and okada

[knotty moth]

I think there should be an option to only save model weights but G and D files are only saved in the target epoch, which is what I've tweaked

the downside is that it could lose the progress if interrupted

[knotty moth]

no idea but the testing results still seem unpredictable for me

I'd consider some factors like amount of noise, glitches, ability to handle edge cases, etc.

[dusty zodiac]

hey there , where can i find flux kontext dev kaggle notebook ?

[true ravine]

@red kayak fix my voice changer

it sounds bad

[red kayak]

i cant simply fix it without any context. tell me whats up

[true ravine]

sounds robotic

i made the model in weights

is that normal

[red kayak]

yeah it could work

have you tried looking into our guides?

[true ravine]

no

where is it

[dusty zodiac]

hello

[true ravine]

is ur profile picture ai

[dusty zodiac]

Can i have Flux kontext dev kaggle notebook ?

and is there a kaggle nb collection ?

[red kayak]

https://discord.com/channels/1159260121998827560/1297207135469305866

have a look

[true ravine]

is it my model

that sounds bad

[red kayak]

it could be yes

quality also matters

and prounounciation does

[true ravine]

what is the best model

ever

to exist

[red kayak]

theres no such thing

[true ravine]

so like

no model

thats trained over and over again

like 1 month of training

[red kayak]

nope, thats not really how machine learning works

[true ravine]

oh

[balmy plover]

How do you use AI cover now? Use the simplest method on colab

[ruby cairn]

is it normal for my voice changer have like 1-2 sec delay in dc voice chat?

using the server audio will reduce the delay but it will have echo issues

[flint vapor]

Yes is normal 😁

[thin summit]

hey guys where can i download okada??

[flint vapor]

-wokada

[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

[flint vapor]

Here there is a guide! https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

[ruby cairn]

well that sucks, it not suitable for gaming,
at least i could use it for chatting

[latent kettle]

yeah thats czz it process on your local machine hardware rather than a server or cloud computer

[zenith mountain]

Question, is the "v.1.5.3.18a onnxgpu-cuda" the old version? Trying to use it for a combine soldier voice model but it is at times a bit choppy and not always smooth so I feel like it's the old original version. Which one is the newest stable original version? Is it the 2.0.78 beta or 2.1.4 alpha?
I got a 2060 RTX, and Windows 11, trying to use it in G'mod.

[ruby cairn]

So theoretically, IF someone have like high end Hardware (like, GPU not sure Which factor, I'm a tech boom💀 ),does it work even better and as fast as real time?

[latent kettle]

Probably, I don't have a decent GPU. But yes, with a good gpu like RTX 4090 or RTX 5090 you can run games and voice changer at same time.

[knotty moth]

I suppose RTX 4060 Ti 16 GB could be good enough for marvel rivals with voice changer and some tweaked down settings

[viral mason]

I think I'm the best model

[low shard]

You had an old version of the program, click on the Nvidia windows download

[low shard]

Don't use anything from YouTube, all videos are old

What's your Mac m chip?

[low shard]

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)

This is a General AI Server, AI has many fields

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]

You should try checking in https://docs.aihub.gg/rvc/resources/dataset-isolation/#uvr-zero-gpu

[low shard]

This is a General AI Server, AI has many fields

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]

Elaborate:

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

[viral mason]

why do people say ai backwards

[low shard]

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)

Elaborate:

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

[low shard]

Whats your PC GPU and operative system first?

[low shard]

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)

This is a General AI Server, AI has many fields

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]

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)

Elaborate:

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

[low shard]

There's no such thing as 0 delay

[low shard]

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)

Elaborate:

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

[low shard]

Yes that's the over year old original wokada, the best one Is wokada deiteris fork b2332, not original wokada

[zenith mountain]

But why is it the best? And what is the difference between the deiteris and latest original wokada?

I noticed the originals are considerably bigger in gigabytes than the deiteris

[viral mason]

preformance I believe, as well as a cool formant shifting

[zenith mountain]

I care about quality over performance. I do have good enough of a GPU to run either.

[viral mason]

nick prob knows more and I just woke up I'm sleepy as hell

[low shard]

Performance and quality, fork means modified version

[zenith mountain]

Alright I will give it a go. Thanks both of you @viral mason.

[viral mason]

I tried

[low shard]

Even the latest original wokada just has UI changes, not anything related to performance and quality, except some performance for Beatrice, but Beatrice models are lightweight and have worse quality than rvc, which is why nobody uses them

[viral mason]

what is beatrice

lmao

[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]

1st link wokada deiteris fork btw

[zenith mountain]

Aye cheers. One last thing actually- would I have to uninstall the original before installing this one or not?

[flint vapor]

Is a model integrated in default okada 😄 but is not so much customizable

[low shard]

Beatrice V2 is a type of Speech To Speech model made by Wok, they are the preset ones too for example, they are lightweight and can run fine even on CPU, but sacrifices quality

[viral mason]

quick reccomendation before you do that, take a screenshot or screenrecording of what models you have in case u forget

[zenith mountain]

Only got one haha, from Half Life 2.

[viral mason]

I usually have over a hundred models I put into mine if I enjoy the software hehe

[zenith mountain]

Eh personally I just need it for the cool metro cop voice.

[viral mason]

oh you're about to be in luck bc I'm making half-life models soon

more like remaking

[zenith mountain]

Oooh, well I think a classic combine soldier would be cool.

I was only able to find a metro cop one, and some Alyx one's that aren't really the same as the classic combine soldiers.

If you remember to do tag me! Especially if you make a pretrain model for it that I can use.

[viral mason]

the one that sounds like this right?

[low shard]

I mean you don't have to, tho at least uninstall vb audio cable if you use that, vac lite is better, many users reported that vb audio cable gives issues randomly on windows

[flint vapor]

Yes true, to me vb audio cable was crack a lot

And vac fixed it

[zenith mountain]

Oh yeah. Where is this? I couldn't find it.

[viral mason]

I tried vac lite and couldn't get it to work with voicemod which is why I use that still

[viral mason]

just look up combine in the voice models section

[zenith mountain]

I did but I only saw metro cops and the alyx one's.

Let me try again though!

[flint vapor]

With VB cable is working good to you?

[viral mason]

most recent of mine of the hl2 one is that dancing combine soldier

[viral mason]

yup

[viral mason]

https://discord.com/channels/1159260121998827560/1303926690657079317

[flint vapor]

Nice! Sometime i think depends and is not for all same error, to me while using okada it was a bit crackling, but only when i was in a voice chat

[viral mason]

interesting

[zenith mountain]

Voicemod is that subscription app right? Is it actually better than wokada? Though the only thing I'd be using it for is the Half Life stuff which I might as well use wokada for, but I am still curious.

[viral mason]

I only use voicemod because it allows me to use custom fx without being complicated

I use it with Okada tho

[zenith mountain]

Doesn't that make the delay worse? Or is the Voicemod fx pretty much a pre-applied filter aka an instant conversion to where only the wokada translation speed is noticeable.

[low shard]

Voicemod isn't even ai, it's just vocal effects

And yes it's paid

It gives random free effects iirc

[viral mason]

Voicemod doesn't affect the delay on okada from what I have seen

it just puts like a filter over your voice like reverb or a deeper pitch ect

[zenith mountain]

Fair.

[zenith mountain]

Got an issue, note I did not uninstall the original maybe that's why? Here is the error log:

2025-07-09 15:25:01,973 INFO [main] Python: 3.12.7 (tags/v3.12.7:0b05ead, Oct 1 2024, 03:06:41) [MSC v.1941 64 bit (AMD64)]
2025-07-09 15:25:01,974 INFO [main] Voice changer version: b2332 NVIDIA-CUDA
2025-07-09 15:25:01,974 INFO [WeightDownloader] Loading weights.
2025-07-09 15:25:02,423 INFO [Downloader] Verified pretrain/crepe_onnx_tiny.onnx
2025-07-09 15:25:02,429 INFO [Downloader] Verified pretrain/crepe_tiny.pth
2025-07-09 15:25:02,493 INFO [Downloader] Verified pretrain/fcpe.onnx
2025-07-09 15:30:07,088 ERROR [WeightDownloader] Failed to download or verify pretrain/crepe_full.pth
2025-07-09 15:30:07,088 ERROR [WeightDownloader]
NoneType: None
2025-07-09 15:30:07,088 ERROR [WeightDownloader] Failed to download or verify pretrain/content_vec_500.onnx
2025-07-09 15:30:07,088 ERROR [WeightDownloader]
NoneType: None
2025-07-09 15:30:07,089 ERROR [WeightDownloader] Failed to download or verify pretrain/rmvpe.pt
2025-07-09 15:30:07,089 ERROR [WeightDownloader]
NoneType: None
2025-07-09 15:30:07,089 ERROR [WeightDownloader] Failed to download or verify pretrain/rmvpe.onnx
2025-07-09 15:30:07,089 ERROR [WeightDownloader]
NoneType: None
2025-07-09 15:30:07,089 ERROR [WeightDownloader] Failed to download or verify pretrain/fcpe.pt
2025-07-09 15:30:07,089 ERROR [WeightDownloader]
NoneType: None
Traceback (most recent call last):
File "client.py", line 22, in <module>
File "asyncio\runners.py", line 194, in run
File "asyncio\runners.py", line 118, in run
File "asyncio\base_events.py", line 687, in run_until_complete
File "main.py", line 90, in main
File "downloader\WeightDownloader.py", line 88, in downloadWeight
Exceptions.PretrainDownloadException: 'Failed to download pretrain models.'

[low shard]

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

it happens when u have a slow connection

check that to fix it

[zenith mountain]

Will do.

[zenith mountain]

Clicked on the link (https://drive.google.com/drive/folders/1OFfM9rmxCZkiYjxoK_yzhRbcXpt0TiJ0?usp=drive_link) to download the files manually got this:

Google
404. That’s an error.

The requested URL was not found on this server. That’s all we know.

So maybe the problem is that the site where the prompt downloads them from is down?

[low shard]

@analog obsidian did deiteris delete those files from his drive?

[low shard]

the google drive is just a backup

it might be better you retry with a better connection till he reuploads it on his drivew

because the issue is your wifi is that slow that it goes in timeout for taking too long downloading them

[zenith mountain]

Would you be able to maybe send me a copy from you? I know it's a bit much to ask.

Like a copy of the pretrain models. If you have the default one's that should come with the download.

[low shard]

I'm not on PC rn 😭

[zenith mountain]

Rip. Well guess I'll keep re-trying and waiting for a new back up link to appear.

[knotty moth]

lighthost is a free alternative

[viral mason]

never heard of it

[quasi condor]

yo its been a long time since ive used applio on kaggle so what the heck is this o - o

[viral mason]

no clue

[quasi condor]

why do i need to log in

[quasi condor]

[viral mason]

you need an account on kaggle to make things lol

[quasi condor]

yea i know

[viral mason]

just use your google acc

[quasi condor]

dude its from this

[viral mason]

idk what that is

applio on kaggle doesn't need all that

[quasi condor]

its from shirou's

[viral mason]

what's that

[quasi condor]

@warm shell <----- THIS GUY

yk what nvm ill just figure it out

[simple ore]

this is a file browser, just add login/password in the cell that makes it

[viral mason]

yeah but like

what kaggle space is that

applio or smth else

[quasi condor]

i just told you the dude's name :

[viral mason]

you said you got it from that dude

I'm confused

[zenith mountain]

Small update, I think I got it to install but it only opens up to the web client, I don't see a bat file anywhere to open it locally. I downloaded the 2.74 GB Deiteris':

Download NVIDIA on Windows
The lastest version as of December 7th 2024 is: nvidia-b2332 (click here to download)

[low shard]

Yeah that's completely normal

It opens locally

On a web user interface

[zenith mountain]

Ooh. Is there no way to make it a custom interface?

I don't like it opening on Opera since I have like a billion tabs open.

[low shard]

Nope, it's better you read https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#why-does-it-run-in-a-browser-and-not-its-own-window for more info

[low shard]

You can just copy the url and paste it in another browser

[zenith mountain]

Aye, thanks for the link.

[low shard]

Opera gx can cause issues as some reported, but not sure I don't use it

[zenith mountain]

Fair, not my cup of tea, but I guess it's not a huge deal.

[nocturne helm]

anyone have the voice changer download link?

[thin summit]

whats the settings for gtx 1660s graphics card

for the voice changer of course

[viral mason]

do u have nvidia gpu?

[nocturne helm]

yeah

gtx 1650

[viral mason]

cool! gimme a sec and I'll send u it

I'll hekp u set it up and stuff

[knotty moth]

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

[viral mason]

nvm it's right there

[analog obsidian]

lol yea the link also doesnt work for me

[viral mason]

wait what

[ornate dagger]

how to setup voice changer program?

[viral mason]

look at that link up there

[low shard]

hows it going btw?

[low shard]

can u do a reupload or dm him?

[analog obsidian]

i dont have contact with him lol

[low shard]

windows and you want a realtime voice changer right?

[low shard]

share a screenshot of ur program settings

!give-media-perms 1h @thin summit

[low shard]

what's your pc gpu? operative system? what do you want to do?

[zenith mountain]

Seems to work all fine for now from preview. I haven't tested the Half Life voice model in-game yet though.

[nocturne helm]

yeah

[violet rose]

RTX 3070
Windows 11
I want a voice that sounds ultra realist

[viral mason]

wouldn't you wanna sound like Homer

[violet rose]

Well, I want it to be realistic that we can laugh/cough without it bugging, you know?

[viral mason]

fair

if the dataset had coughing, laughing, ect it would be able to if trained correctly

there's enough simpson to make a reallllllllllllllllllllllly good model I'd think

[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]

1st link, wokada deiteris fork

[low shard]

if u share a screenshot of ur settings i can check them

[low shard]

RVC has a limitation on that

[low shard]

so realtime voice changer for calls right?

[violet rose]

What should be done in this case?

[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]

1st link

wokada deiteris fork

that's if you want a realtime voice changer for calls

it's for sure a great gpu, what do u want to do btw

[violet rose]

this is the last version "voice-changer-windows-nvidia-b2332"

?

[low shard]

yes

depends which deepseek? deepseek 671b? you'd need an entire server to run that

[analog obsidian]

save more and buy 5090

[low shard]

sure, u just need over 1k gb of vram lmao https://apxml.com/posts/system-requirements-deepseek-models

i mean ofc the 5090 would be better

for local rvc you don't have to worry at all

you can also do wokada deiteris fork with intensive games without issues

about 3d modelling and image gen, you should be fine, tho i haven't tried the new omnigen2 and chroma that @simple ore talked about

the most intensive thing that you could be worried about is LLMs

you should be able to run fine models with like 70b max iirc parameters, which is great tbh

https://www.databasemart.com/blog/choosing-the-right-gpu-for-popluar-llms-on-ollama#:~:text=Small Models (2B%20%2D%2010B%20parameters,with%2024%2D48GB%20of%20VRAM.

[ornate dagger]

I have finished downloading but I don't see the program. How do I use it?

[low shard]

nevermind about that bit, i confused with 24gb for a sec 😭

yeah google gave me that info when i tried googling th ertx 5080 vram as i forgot 😭

[zenith mountain]

It should be all good 👍

I will test it out tonight and let you know if I have any other issues though.

[low shard]

on rtx 4060 ti 16gb vram with 32gb ram i can run max like 32b parameter models, but they are kinda slow

[low shard]

nope
extra to 2.7, over that that can cause cutoff issues on some models
f0: rmvpe without onnx
output: uninstall vb audio cable it causes issues on windows, get vac lite from th eguide instead
input: microphone
chunk: 128ms

on wokada deiteris fork, you can **optionally **use more advanced settings for benefits:

• Force FP32 mode: on (THIS IS OFF BY DEFAULT! Turning this on improves stability. Increases VRAM usage by 200 MB)
• Disable JIT compilation: off for faster loading speed of the program, on for slightly better performance (10-15 ms) for Nvidia only)
• Reduce the delay via: https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#reduce-more-delay

ofc ai is vram hungry

[zenith mountain]

Aye, I will try those settings, I do have vac lite installed as well.

[low shard]

@timber crow btw not sure how the rtx 50 serie is rn, i hope they fixed the bugs with drivers tho, i heard some complain while others not
about programs, most programs should support the 50 serie, and if they don't you can simply update their pytorch version

[zenith mountain]

I know the reduce delay but how do I disable JIT compilation and Force FP32 mode? Probably gonna find it instructions on the website yes? Or is in the advanced settings tab?

Yhup found it.

[low shard]

elaborate what i asked, also what program? what did u download?

[low shard]

advanced settings tab

[zenith mountain]

Check out?

[ornate dagger]

The link you gave me

[native cedar]

why ur chunk so high

[low shard]

set disable jit compilation on, if you want it to have slightly better performance but slightly takes more on startup

[zenith mountain]

Better quality.

Takes longer to process audio and I have tested it, sounds better.

For me it takes 2 seconds.

[native cedar]

but very long time dam'

[zenith mountain]

At said chun.

Nah 2 seconds.

[native cedar]

2 secs dam

prob the same wit me

[zenith mountain]

Cheers.

[low shard]

what link? i just asked ur pc gpu and operative system

you should also need to tell me that

[zenith mountain]

None of these settings actually noticably lower the audio quality though right?

[low shard]

yeah unfortunately the rtx 50 serie didn't start so much well

[low shard]

nope

[native cedar]

what should ma advance settings be for rvc website version

[ornate dagger]

what is index?

[low shard]

it's not rvc nor a website

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)

It's wokada deiteris fork

anyways,
on wokada deiteris fork, you can **optionally **use more advanced settings for benefits:

• Force FP32 mode: on (THIS IS OFF BY DEFAULT! Turning this on improves stability. Increases VRAM usage by 200 MB)
• Disable JIT compilation: off for faster loading speed of the program, on for slightly better performance (10-15 ms) for Nvidia only)
• Reduce the delay via: https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#reduce-more-delay

[native cedar]

wdym the voice changer the advance setings

[low shard]

those advanced settings are suggested for everyone

[low shard]

i explained you the program above, read it, rvc doesn't mean realtime voice changer

[young halo]

"Preprocess completed in 0.00 seconds on 00:00:00 seconds of audio." what does that mean? i'm trying to train a model on applio colab

[low shard]

the trained accent

[low shard]

what's ur dataset file format? how did u make it?

[native cedar]

so dis isnt a voice changer?

[ornate dagger]

Should I open it?

[young halo]

its a zip file, the audio itself it's wav with like 11 minutes

[low shard]

it is, but it's not called like that, it's called wokada deiteris fork, RVC has a different meaning, please read it above

voice changer is a too broad term

there's thousands of programs i could simply call "voice changer"

[zenith mountain]

It is.

Like the admin said.

[native cedar]

asked the admin

[viral mason]

you have to remember people aren't smart

just look at me

https://tenor.com/view/zhouzhou-orange-cat-cat-with-orange-sleepy-cat-gif-14937184825975080647

[young halo]

it's not letting me generate index also

[low shard]

What program did you use to make the dataset? have you tried re-exporting the wav via audacity and using applio's dataset maker tool?

[zenith mountain]

https://tenor.com/view/neuro-head-spinning-gif-12552456361102904795

[low shard]

I was just teaching them

[young halo]

i used audacity actually, and then put the files in a folder and made it a zip file

i've always done it like this on rvc disconnected and i've never had any problem

[analog obsidian]

rvc = ai voice cloning
w-okada = app that runs rvc models in realtime

[young halo]

but i guess it's different with applio colab

[low shard]

rvc = ai voice cloning
STS ai voice cloning btw

[viral mason]

https://tenor.com/view/epic-embed-fail-ryan-gosling-cereal-embed-failure-laugh-at-this-user-gif-20627924

/j

[low shard]

you can do it without zipping https://docs.aihub.gg/rvc/cloud/applio-colab/#b-dataset-path

[native cedar]

for the w okada web browser version is there a way to stop all the backround air cus it mubles and sup1 and sup2 makes it really bad or do i have to use sup2 and 1

[young halo]

so just the audio itself?

[hasty condor]

What would you guys say are the best models out of #1175430844685484042
Best as in most realistic sounding ones xD

[zenith mountain]

rough no gif perms 😭

[young halo]

i did so and i got the same message, 0 seconds of audio

what am i doing wrong?

[nocturne helm]

whats the best settings for smooth realistic voice for Deiteris' W Okada

gtx 1650

[young halo]

when making the model, which one of these index do i have to zip with the pth?

[keen coyote]

anyone know the setting solution for a robotic sounding voice?

[odd shale]

There is no setting solution for robotic-sounding models, if the model sounds robotic, the fault is on your dataset

Reasons can range from uncleaned noise, sfx, harsh sibilants, overtraining and etc

[keen coyote]

alr thanks leo ill keep that in mind

[craggy idol]

hi everyone, can yall help me with something? I'm trying to download applio but it keeps coming without the run_instal.bat

what am I doing wrong? I made sure my antivirus was off

[flint vapor]

You have to install it from huggingface i think you have installed from github

[flint vapor]

https://huggingface.co/IAHispano/Applio/resolve/main/Compiled/Windows/ApplioV3.2.9.zip

[craggy idol]

oh yes I did

thank you, I'll try it now!!

[flint vapor]

You welcome! This will have the .bat file 100%

run-applio.bat

[low shard]

it's called wokada deiteris fork

use just sup2 and echo

[low shard]

yes

[craggy idol]

oh wait, mine has this

but it does not has the instal one

[flint vapor]

Don't worry if you do run applio it will install

[craggy idol]

I already applio downloaded, but its not working anymore

[low shard]

use applio's dataset maker @young halo

[low shard]

share a screenshot of ur wokada

!give-media-perms 1h @nocturne helm

[flint vapor]

Try do again run-applio.bat is the command for open the interface

[craggy idol]

okii

[young halo]

I did now, thank you!

[craggy idol]

ill try it on the new one u sent me tho

[flint vapor]

Sure! Let me know 😄

[nocturne helm]

K 1 sec

[low shard]

its to check ur settings and suggest better ones

[craggy idol]

@flint vapor it worked! thank you ml ❤️

[flint vapor]

You welcome! If you need more help you can always text here!

[low shard]

f0: rmvpe without onnx
extra: 2.7
chunk: 200
output: line 1

[nocturne helm]

k

[low shard]

on wokada deiteris fork, you can **optionally **use more advanced settings for benefits:

• Force FP32 mode: on (THIS IS OFF BY DEFAULT! Turning this on improves stability. Increases VRAM usage by 200 MB)
• Disable JIT compilation: off for faster loading speed of the program, on for slightly better performance (10-15 ms) for Nvidia only)
• Reduce the delay via: https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#reduce-more-delay

[analog obsidian]

16xx series can't do fp16 inference, so he has fp32 enabled by default

[viral mason]

chunk at 360 is pretty good tbh

[lament remnant]

Hello, is there anyone here who knows about TTS?

[flint vapor]

For TTS i recommend you GPT-SoVits

The new version with V2ProPlus

[lament remnant]

I’m getting an error while fine-tuning with VITS together with Coqui. I wanted to ask for help to solve the error.

[simple ore]

what language do you need?

[lament remnant]

I am currently trying to fine tune Turkish with coqui VITS

[simple ore]

Coqui is dead

VITS is morally outdated

XTTS-v2 has Turkish

how good, I can not tell

most new models use some kind of GPT/LLM to generate speech tokens so the output does not sound monotone and robotic

[lament remnant]

Unfortunately, I cannot use these models because they are paid, but I will take a look at the model you mentioned, thank you very much.

[flint vapor]

There is a new veersion released in June, V2ProPlus sounds good but the only problem is that languages are very few

If someone need more language yes XTTS is good

[simple ore]

xtts is free for personal use

gpt sovits without finetuning was ass and it only supports en, cn, jp

maybe the new model is better, but anyway

[flint vapor]

Oh yes the new model work very good without fine tuning i saw

[chilly grove]

so im using a voice rn but its picking up everyones audio and idk how to fix it

[knotty moth]

use headphones as the system audio to listen and turn down the volume

if you mean irl ppl, tell them to be quiet

[chilly grove]

no in game lol

so by turning down volume would it be the monitor

or input or out

cause im using voicemod to try input in, but its just picking up all its audio

[knotty moth]

turn down the ppl's audio volume and anything on the headphones

[chilly grove]

ohhhh ok

[viral mason]

he fixed it

btw

[urban osprey]

quick question, is the RVC available for Linux?

(cuz im planning to switch to Linux).

[snow inlet]

what is the best voice changer to create a girl's voice for a video?

it doesn't have to be in real time

[flint vapor]

Yes it is but you need to install all manually

Which distro you want move on?

[urban osprey]

hmmm....
Linux Mint ig?

[sick venture]

Anyone know how to retrain a voice model I lwky forgot

[crystal gate]

how do i download w okada for mac?

[hidden latch]

Why does it not work in discord

[native cedar]

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

@duckus

where is he

[urban osprey]

Does it work on its own?
then it is on Discord's side.
Have you check if Discor is using the default mice or the voice changer?

[simple yoke]

Can somebody help me identify what version of RVC I had? I can't seem to post images in here to give a reference, but I downloaded it somtime around december as a recent release. However if I look at any github links here nothing seems to match the timeline

[warm cedar]

im using w okada and i just am looking for some fun ideas for voices to try out and hoping someone could send me the link to some models

like a girl voice model or anime girl voice model

im very new to it all and was just wondering what you guys considered must haves for doing a bit like fortnite trolling lol

[ionic fossil]

What's up with the tag removal in the voice models search?

Is this to do with the recent Weights changes?

[viral mason]

@low shard he stole your joke

[viral mason]

https://cdn.discordapp.com/attachments/1369012167621214240/1392509350706282556/blink-gman.gif

[native cedar]

bro in voice model what happen to e girl tag

[viral mason]

glad it's gone, it attracted yucky pests

[brittle wing]

Hello, How do I get Ai model training chats?

[wise sand]

What does my error mean does anyone know?

Exceptions.VoiceChangerIsNotSelectedException: 'Voice Changer is not selected.'
2025-07-10 00:05:58,057 ERROR [VoiceChangerManager] 'Voice Changer is not selected.'
Traceback (most recent call last):
File "voice_changer\VoiceChangerManager.py", line 212, in change_voice
audio, vol, perf = self.vc.on_request(receivedData)
File "torch\utils_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
File "voice_changer\VoiceChangerV2.py", line 159, in on_request
raise VoiceChangerIsNotSelectedException("Voice Changer is not selected.")
Exceptions.VoiceChangerIsNotSelectedException: 'Voice Changer is not selected.'

Pls help

[proven mortar]

Verification Required

If you wish to continue playing Dank Memer, you need to verify your account.

To Verify:

• Click the button that says "Pass Verification" to visit the Dank Memer Website.
• Read and accept the rules.
• Complete the Captcha

After verifying, you'll be able to play again as normal. You will also receive 100 for passing the verification.

If you have trouble verifying, visit the Support Server for help.

There is no time limit.

[proven mortar]

@wise sand

You have an unread alert!

Use </alert:1011560370864930852> to read it!

[edgy minnow]

haiiii why is OKADA lowing all the volumes of EVERY other program without permission

[simple ore]

when you're using an input or output that is flagged ad 'default communications device' on either playback or recording tab

[grizzled sapphire]

how do i download voice changer

and use

[flint vapor]

There are different UI and software that make you use RVC, Can i see a screenshot of the UI?

[flint vapor]

-ùokada

[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

[flint vapor]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/ Here there is the guide 😄

[simple yoke]

I cant add pictures in here for some reason

[flint vapor]

Oh you can send me in DM 😄

[latent cypress]

is there any ai to extend a song locally?

[limber shell]

hi guys, been out of the loop for 2 years, where do I find the newest models recommended for vocal separation as suggested on the aihub website?

[pseudo swift]

https://huggingface.co/spaces/TheStinger/UVR5_UI

[simple ore]

fv4

[limber shell]

ohh we using this rather than the local app now?

[latent cypress]

not really

[pseudo swift]

i was recommended that by other people in this server but for local app i think the equivalent is https://github.com/Eddycrack864/UVR5-UI

[latent cypress]

https://ultimatevocalremover.com/ this one

[limber shell]

Yea I have the beta version of that

Was wondering why I couldn’t find the recommended models anywhere

[limber shell]

Turns out it’s like a separate fork?

[pseudo swift]

im not sure but it does have the recommended models

[latent cypress]

because it's on the creators huggingface page

https://huggingface.co/GaboxR67/MelBandRoformers/tree/main/melbandroformers/vocals

[latent cypress]

voc fv4 ckpt and yaml file is here

[limber shell]

oh yes I remember finding that last night, no idea how to install custom models like that though

[pseudo swift]

what would be the difference between this and the one by eddycrack?

[latent cypress]

👨‍💻 HOW TO DOWNLOAD AND PUT THE MODEL MANUALLY INTO UVR GUI?

• download the .ckpt and .yaml files of the model you want on huggingface
• open uvr gui
• click settings icon (on the bottom left, right next to the 'start processing' button)
• click "choose advanced menu"
• click "advanced mdx net option"
• click "open models folder"
• drop the .ckpt file into the MDX_NET_Models folder (when you click "open models folder," you’re already inside the MDX_NET_Models folder.).
• once that’s done, go to the model_data folder, open the mdx_c_configs folder, and place the .yaml file there.

[limber shell]

thanks a bunch, will try

[latent cypress]

tell me if the uvr system ask you to set the parameters

[knotty moth]

alternatively you can clone and install this repository

https://github.com/ZFTurbo/Music-Source-Separation-Training/

[latent cypress]

not sure, i’ve never really compared them (in terms of output quality). i usually just use eddie’s uvr hf when my pc’s running heavy processes

[pseudo swift]

ahh i see that makes sense

btw would you cut out talking from a dataset if you want a purely singing model?

[low shard]

Any updates?

[nocturne helm]

good but chunk was too low

it sounds weirdly roblotic

[low shard]

I'm guessing you're talking about either mainline or Applio and yes

[low shard]

Elaborate more

Are you trying to catfish?

[low shard]

That's pretty not bad for Linux beginners

[urban osprey]

uhhh..... idk....
i just want to know if the RealTime Voice Changer works in Linux

[low shard]

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)

Elaborate:

• your PC GPU
• your operative system

[urban osprey]

idk, just wanna try it

[low shard]

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)

Elaborate:

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

[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

[low shard]

Lmao

[low shard]

🔥🔥

Jokes aside it's pretty hard for people to help without info, that's why I always ask them

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]

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)

What's your PC GPU and operative system? Which program are you really looking to use ?

[low shard]

You mean the removed e girl tag? It wasn't needed

I removed it myself

[low shard]

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)

Elaborate:

• your PC GPU
• your operative system

[low shard]

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)

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]

Try other models, not all models are good

[brittle wing]

AI chatbots: ChatGPT, Gemini, Copilot, Perplexity, Grok, and Claude. Out of these seven, which three are the best AI chatbots for research?

[flint vapor]

Perplexity

[brittle wing]

May I ask you, why you recommended Perplexity?

[flint vapor]

Because they introduced also lab mode, and it also create you a scheme or an entire list of the research that perplexity do 😄 , but perplexity is mostly recommended if you have premium version

[brittle wing]

What about the free version?

[flint vapor]

Also the free is good but mostly only for normal research

Then in perplexity you can set which ai you prefer to use

[brittle wing]

Could you recommend another two AI chatbots for research purposes?

[flint vapor]

ChatGPT and gemini or Claude

[latent kettle]

Is there any possible way to do inference on mobile (locally) not cloud

[flint vapor]

Unfortunately local at the moment on phone is not possible

[brittle wing]

Appreciate the recommendations brother but I only need another two.

[latent kettle]

Inference too ? I think I have listed about that

[low shard]

Yes

Via termux

Is it possible? Yes, should you do it? No

[latent kettle]

Crazy I knew that

[low shard]

I'm the creator of the guide lol

[latent kettle]

Why

[low shard]

I'm actually the first that ran it on phone btw

[latent kettle]

Yeah I know that but I was not sure

[flint vapor]

Sorry i thought he mean the entire gui like on pc

[low shard]

It can take over than 60 secs for inferencing 8 secs

[flint vapor]

If is for inference yes

[low shard]

Yeah, you can do that

It shows the gui

[flint vapor]

Oh nice i didn't know it can be done, but i think just inference training for now is impossible

[latent kettle]

Reasonable, I want to try that

[flint vapor]

I think ChatGPT and Claude are more precise

[latent kettle]

I got snapdragon 855 and 6gb RAM using samsung s10e

[brittle wing]

Thanks for the information brother. May God bless you.

[flint vapor]

Thank you so much! You too 😄

[low shard]

Training is impossible, at least from the power of the phones they were tested on, it would overhear shortly after

https://discord.com/channels/1159260121998827560/1289538710307602554 I haven't tested it in a while tho

I would really not suggest it unless you want to do it just for fun or hare cloud

I did it just for fun lol

[flint vapor]

Yes training require more power and phones atm don't have enough, also inference i saw you said is possible but i think is always recommend to do it on cloud if someone doesn't have on pc, is very slower local on phone i think

[low shard]

Yeah ofc, did it just for the memes

[latent kettle]

I want to do it for fun

So it's basically installing ubantu distribution in termux and than installing RVC on it ?

[low shard]

Kinda

[warm cedar]

is a 4060 good enough for w okada?

also when i open w okada how do i close the system directory thing without closing okada (the black box that shows what’s happening)

[covert valve]

can someone help me to setup the AI voice changer?

[fleet cedar]

yes

[fleet cedar]

depends on ur gpu

and cpu

[flint vapor]

Yes it is

[warm cedar]

does anyone know how to fix this issue though? it’s really annoying

i don’t trust myself to mess with things either 😭

[fleet cedar]

yo wtf bro why cant i hear my voice model but it works when i talk in discord

mon isnt working

[covert valve]

i got amd

idk how to setup can you help?

[fleet cedar]

idk too

[uncut haven]

How to reduce the chunk without losing quality? 192 ís about 512ms

[covert valve]

whats ur gpu

[uncut haven]

3070 8gb

[covert valve]

idk

[uncut haven]

I've 192 (512.9ms, 24576) and EXTRA: 131072

[covert valve]

what is this error

[low shard]

yes

[low shard]

you don't, you can't, that thing is the command prompt, it's what is actually running wokada, the wokada in oyur bwoser is just the web user interface

[flint vapor]

vccclient is a text to speech made by the same author of wokada

[low shard]

I hope you're using wokada deiteris fork and not youtube tutorials, right?

[covert valve]

what to do?

i need the voice changer but my gpu is amd

[flint vapor]

-okada

-wokada

[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]

that's old original wokada from yt tuts, all yt tuts are outdated

uninstall everything

[covert valve]

and?

[warm cedar]

thank you; even if it’s annoying it’s nice to know i’m at least doing it correctly

[flint vapor]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/ there is here a tutorial 😄

[low shard]

what's your pc gpu and operative system?

[flint vapor]

Is a guide

[covert valve]

for amd gpu?

[low shard]

let's check first if the gpu is good enough

[flint vapor]

Yes sure 😄

[low shard]

what's your pc gpu first? there's many AMD ones

[covert valve]

win 11 5700XT

radeon rx 5700xt

[flint vapor]

Yes there is also an amd version but is recommended to use only if the gpu is ok

[covert valve]

what to do now?

[low shard]

alr good enough

[low shard]

read the wokada deiteris fork guide

[covert valve]

nick tell me what to do mate

[uncut haven]

So I've a 3070 8GB
I've Windows 11
I want to use different Girl voices to troll people (Found the settings just the m/s is annoying)
https://www.youtube.com/watch?v=SxdnGxicJOg
https://i.imgur.com/09Pgl8a.png

[flint vapor]

There is a download link in the guide for amd

[covert valve]

im dumb i cant

[low shard]

guessing that you want a realtime voice changer for calls

[flint vapor]

https://github.com/deiteris/voice-changer/releases/download/b2332/voice-changer-windows-amd64-dml.zip

[low shard]

you need to read it

[covert valve]

BUT IM DUMB

[low shard]

that video is outdated, it uses an over year old version of original wokada, and vb audio cable causes issues on windows..

[flint vapor]

@covert valve

[covert valve]

its not openning anything

the link

[low shard]

uninstall everything you got

-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, wokada deiteris fork

[flint vapor]

If you click it will download amd version

[covert valve]

nick could you call me?

[uncut haven]

How to fix it?

[low shard]

you have to read it to understand it, it's open source ai, it's not as easy as using chatgpt, the program is complex and intensive

you have to read it to atleast understand

if you dont read it, you will never understand it

[covert valve]

well

im not english so ...

[low shard]

uninstall everything, get wokada deiteris fork from the written guide i sent you

[low shard]

is there a specific part you don't understand well in the guide ?

[flint vapor]

Where are you from? I can translate it for you and send you translation

[low shard]

sorry but this server is english only

[covert valve]

are you AI?

[low shard]

we are not ai 💔

[covert valve]

bruh

[flint vapor]

I will use Translator and send you in DM

[uncut haven]

I didn't notice the link can you send again?

[covert valve]

look i just want to do a setup for realtime voice changer and my gpu is 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]

1st link, wokada deiteris fork

[covert valve]

here is the link you sent

[low shard]

first of all, uninstall vb audio cable from the windows app settings

delete the old original wokada you got