✨│ai-help

[simple ore]

instead of C:\training

[analog obsidian]

lol

[bleak nymph]

you're honor i am innocent

[analog obsidian]

lmaooo

[bleak nymph]

i didn't

[analog obsidian]

uhm

what if u just
do the preprocess steps again

[simple ore]

that's pretty obvious it did not load anything

[bleak nymph]

yeah it is

[simple ore]

remove "

[bleak nymph]

i just don't read the cmd

[bleak nymph]

oh 💀

why windows like this

[analog obsidian]

[bleak nymph]

is that the reason??

[simple ore]

and for fks sake, move the whole thing into C:\Applio

[analog obsidian]

[bleak nymph]

i'm preprocessing it

[bleak nymph]

wym

sorry i didn't mean to be rude

my fingers genuinely typed that faster than my brain

i'm not even lying

[simple ore]

I mean move the applio folder to the root

[bleak nymph]

root of what

[simple ore]

of any drive

C:\Applio

[bleak nymph]

oh

why?

what difference would that make

[simple ore]

because keeping software under Desktop is not right

[bleak nymph]

well i don't really consider it like that tbh

idk why

also i do need to access it quite a bit

for model files

so having to search for it everytime is a bit more annoying then just clicking straight to it

also holy moly this preprocess is taking ages

[analog obsidian]

i have my trainer in my desktop too

[bleak nymph]

yay

[crude apex]

Is there any alternative to vac lite (have to pay for a license) or VB audio (doesn't seem to work on win 10)

[simple ore]

you dont have to pay for vac v4.7.0 lite

[left ore]

Any clue why I can't get any inferencing models to show up in RVC WebUI? I got everything working as far as I can tell, but putting downloaded .pth and .index files in their respective locations (weights and logs) does nothing.

[crude apex]

vac lite will keep spamming my ears with that automatic voice about the trial, it is annoying

[bleak nymph]

hi so it's been over half an hour preprocessing aint done

should i be concerned

[simple ore]

you did not download the lite version

[analog obsidian]

yes

[bleak nymph]

oh

what do i do then

[simple ore]

show the screen and the terminal

how many files you got?

[bleak nymph]

1

the cmd hasn't updated yet

oh wtf

i accidentally hit a key on my keyboard in cmd

[simple ore]

did you accidentally click on the terminal window and selected a text

[bleak nymph]

and it started preprocessing

[simple ore]

and it froze everything

yeah

[bleak nymph]

no

i hit enter

and it started it

are you FUCKING KJIDDING ME

[simple ore]

yes, you gradded a mouse

selecting a text

and it froze everything until you press enter and copy the text

[bleak nymph]

bruhhh

that's crazy

yeah it's done

[simple ore]

https://software.muzychenko.net/freeware/vac470lite.zip

[bleak nymph]

insane

[simple ore]

use this, uninstall the other thing

[prime dagger]

guys how to use the RVCAICoverMakerUI

I can't send images

[unique snow]

That to bad.

[prime dagger]

I already installed it im using it in google colab

[unique snow]

Oh.

[bleak nymph]

um so i extracted the features and i scrolled down and i see stop training

instead of start training

i ain't even training anything 🥀

[simple ore]

because UI thinks you're still training

f5

then make sure you re-select model name and sample rate

[crude apex]

Okay I'm using this now, voice works fine. But when paired with discord and testing the voice, I cannot hear myself. Is it only supposed to work inside a VC?

[simple ore]

you just need to route the audio properly

and use push to talk in discord / turn off Krisp voice or whatever it is

[prime dagger]

guys how to use applio in google colab for ai covers

[vast elm]

google colab tends to disconnect users when using gpu instances on an external webui

i'd recommend not to go that way if thats not your only choice

[prime dagger]

oooh ok

so which one is better

[vast elm]

have you even tried weights yet?

[prime dagger]

no

[brittle wing]

does anyone know where i can find a good real time voice changer software?

i heard okada wasn’t bad but i’m not sure if there’s better that’s available

[bleak nymph]

diabolical, it doesn't even appear in the dropdown

[vast elm]

check it out, you can create ai covers in the weights app, no need to go through all that trouble unless you specifically want full customization over what your doing

[prime dagger]

ok i'll try thankss

do you have to pay for any of the options tho?

[simple ore]

click refresh button

[bleak nymph]

i did

[vast elm]

https://www.weights.com/pricing

[bleak nymph]

weights ai model training sucks imo

[vast elm]

no need to pay @prime dagger you can create 20 for free per day

[bleak nymph]

sounds terrible when i tried it

which is why im tryna use rvc

[knotty moth]

24 threads seem crazy, must be an intel i7 cpu

but anyway using more cores in preprocessing & feature doesnt quite give benefit over 1 thread

[vast elm]

well ngl, you probably either didnt give it a good dataset, or just messed something up

cuz its perfectly fine imo

[bleak nymph]

hell nah i hate intel

[bleak nymph]

🤷‍♂️ maybe

[bleak nymph]

ryzen 9 3900x

cpu

[knotty moth]

ah mb but ik

as well as 7900/9900X/X3D

[bleak nymph]

i redid it

it took 8.45 seconds this time lmao

[knotty moth]

depend on the dataset length

and how the dataset is prepared

[bleak nymph]

lets goooo

[analog obsidian]

dont click the cmd next time

[bleak nymph]

finally starting training

[bleak nymph]

yes i learned my lesson

fucking hell my pc made my room so warm now

me when heatwave inside my room

[analog obsidian]

[bleak nymph]

lolll

[bleak nymph]

what does this random spike mean?

suddenly went up and then went back down

[analog obsidian]

noise

[bleak nymph]

oh tf

i'm assuming that's not good

[analog obsidian]

it's normal

[bleak nymph]

oh ok

anyways how's the graph looking

[analog obsidian]

looks fine to me

are your gradients below 1k?

[bleak nymph]

what does that mean

[sacred lichen]

why? is it better voice conversion quality than rvc?

[analog obsidian]

check grad norm g and d

grad norm G should be less than 1k

should be around the 500 range

rvc = ai voice cloning software, retrieval based voice conversion
w-okada = rvc models in realtime
vonovox = rvc models in realtime

[bleak nymph]

uh this?

i don't really know what i'm looking at

[analog obsidian]

disable ignore outliers

smoothing 0.6

[analog obsidian]

imo loss graphs aren't worth enough to watch, rvc finetune barely/never diverges and they lack the issues that are present in training from scratch

[analog obsidian]

if grad norm g below 1k = fine

[analog obsidian]

they can be a bit misleading too, for example, g/total going down while your model is frying

mel always goes down
fm always slowly rises
kl always converge at some value

d/total is a bit useful since it tells you if your discriminator is either too strong or too weak

[analog obsidian]

theyre ok

[bleak nymph]

why what happens if it goes above 1k

[analog obsidian]

chances of the model having some weird issue increases

can be anything, it's random

may also affect the quality too

what i do for my models is to save every 10 epochs and hear them

[bleak nymph]

interesting

[analog obsidian]

then you just choose one epoch that isnt robotic and sounds ok

[alpine pelican]

client or sever for okada which is better fir quality

[knotty moth]

setting input volume above 100 may possibly cause distortion due to clipping

[knotty moth]

https://cdn.discordapp.com/attachments/1159290139609137264/1388070244848898059/image.png?ex=68619ee7&is=68604d67&hm=f1dc11ec347261b91e54dbbcac6fbd0079c9aa7c168f09f95e42af56911d6fb9&

tl;dr it affects more of audio latency

[bleak nymph]

i let it train for a little while longer

it doesn't look like much has changed i think

[bleak nymph]

@analog obsidian it's at epoch 300 now

is it done yet? or still need more training

[analog obsidian]

hear it

[bleak nymph]

oke

[analog obsidian]

don't blindly trust the graphs, like i told you before, g/total likes to go down even when the model is overtraining

what is the current value of d/total?

[wraith cipher]

Very quick question chat, so if i download a model from weights, and mess with the settings a bit I can use it as a real time voice with okada? Im assuming this makes it sound better quality?

[bleak nymph]

i stopped the training

i'm testing it out

by ear

[analog obsidian]

if it's overtrained you'll notice the model sounds compressed, robotic, very unnatural

[bleak nymph]

compressed as in the effect or like compressed as in quality

[analog obsidian]

as the effect

[bleak nymph]

ok

well

i notice all of the above

😭

jfc

[analog obsidian]

now ur job is to go down every 10e

eventually one epoch will sound ok

see? thats why i told u to forget about the graphs xD

[bleak nymph]

lmaoo yeah 😭

[wraith cipher]

@analog obsidian is this accurate?

[analog obsidian]

weights models are regular rvc models

so yeah they work with realtime voice changer apps like w-okada and vonovox

and also local rvc

[wraith cipher]

which do you recommend?

[analog obsidian]

idk i don't use other people models

[wraith cipher]

I see, so they are all fairly similar though quality wise? I know very little but from what I can see online there seems to be a lot of settings you can tweak to change the quality of the voices

^ at least in w-okada

[analog obsidian]

the quality of the model depends in how good the quality was in the dataset, if the dataset had trash quality, you'll get a trash model
w-okada settings only change how the inference works, you can't change how good a model naturally sounds

lets say a model has trash quality, even if u set extra 5s and enable fp32 inference, the model will still sound bad

[wraith cipher]

okay thanks for the thoughts Im going to attempt to download okada and a model and see if I can get it working

[analog obsidian]

good luck

[brittle wing]

I'm getting "No features exist for this model yet. Did you run Feature Extraction?" no matter what dataset I'm using. I even tried a dataset I recently successfully trained with and I'm still getting the same message. Is it a bug?

[ruby sparrow]

hi guys morning afternoon or evening. i have a question. i dont have a nvidia only amd so im using google colab to train a ai voice. rvc v2 disconnected. what does the red thing mean when i run dependencies. cuz i always keep getting the red thing on different browser and even my brother pc who has amd also. i check the video and the guy in the vid never showed what happen after he run dependencies

[wraith cipher]

ive got the voice changer working only thing is, whats the best way to tweak the settings to get the desired effect?

its kinda inconsistent with the settings and Idk if I should be more focused on tweaking my mic sensitivity or the voice gain/tuning

[analog obsidian]

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

[cyan elk]

i wanted to wait to respond until after i trained a model to see if it worked but i'm like 10 hours in and just got to epoch 100/500 so uh it'll take a couple days 💀 but i did follow your instructions and it seems to have worked perfectly so far, so thank you very much!

[sacred lichen]

@golden walrus hello, do you perhaps have some model that support spin? I wanna test new embedded

[pine star]

is it possible to combine two voices while having it sound normal when training?

[next oracle]

what happened to the total number of AI Voice uses?

[knotty moth]

test several model checkpoints

the results may be somewhat unpredictable

[golden walrus]

my model suck

[sacred lichen]

Just wanna see how spin work with rvc, doesnt matter if it sucks

[proper oracle]

when i try to load in it keep sayin waiting for web and itm does forever

[tiny wolf]

google colab isnt working it says no version found for fais gpu

[golden walrus]

okay, i will send you the model.

[knotty moth]

it's more likely outdated

[tiny wolf]

um i kinda wanna use it cause my gpu is integrated

and i have no idea if i can run it

[knotty moth]

try either of these:
https://docs.aihub.gg/rvc/cloud/applio-colab/
https://docs.aihub.gg/rvc/cloud/applio-kaggle/

[tiny wolf]

tysm

[proper oracle]

why do i have so much delay

ive had like no delay before on higher settings

[tiny wolf]

how do i do real time voice changing in this

[knotty moth]

smh you didn't say for voice changer

it would instead be this https://docs.aihub.gg/rvc-voice-changer/cloud/w-okada-kaggle/

[tiny wolf]

mb

[tiny wolf]

oh ok

is that real time right?

[knotty moth]

the colab one may risk your account banned unless paying for the compute units

[tiny wolf]

ahh ok

[tiny wolf]

do i gotta repeat these steps everytime i use it?

[knotty moth]

the session state and data won't be saved after disconnection

(ask chatGPT if you don't understand it)

[tiny wolf]

ah ok

the time limit is more than enough

[edgy compass]

how do i get more volume?

[neat meadow]

How do I get the voice changer to work in asio?

[knotty moth]

https://rentry.co/lessdelayasio

[knotty moth]

you mean above the volume limit?

[neat meadow]

I already have an audio interface driver, should I download flexasio?

[knotty moth]

the guide says so

[neat meadow]

Don't the two drivers collide?

[knotty moth]

wdym? what another driver?

[neat meadow]

Audio interface driver

[knotty moth]

nah another one again

[neat meadow]

flexasio is a virtual audio interface driver

[hallow thistle]

I use ASIO4ALL, but I haven't installed the program, other than being ReaRoute ASIO.

[neat meadow]

Can you choose the output for the input?

[hallow thistle]

It's just this on W-Okada. As what I remember, ASIO4ALL lets you to change and select the main input/output devices there within the program.

[neat meadow]

Can you put output in the input? Not the microphone

[hallow thistle]

Your question seems confusing, but the answer is "no".

Virtual Audio Cable provides both its input and output at the same time. A pair of speakers/headphones don't function as an input device, they are output devices.

[neat meadow]

I'm planning to set input:microphone output:output in the daw program and put output in voice changer input. Can I choose output in flexasio input?

Unable to set line1 in the output because the daw program returns to asio

[hallow thistle]

[neat meadow]

i'll try thx

[neat meadow]

and why you use 44100hz??

[knotty moth]

I think it should be 48k for server mode, otherwise it may throw some error

[neat meadow]

you are server mode

[hallow thistle]

Is that a problem? I did that mainly for example, not as a practical scenario.

Although I may set it to server mode in case if I have to use W-Okada with ASIO. However, I mostly use client mode in general.

[ocean holly]

What is the main factor that determines the quality of the voice changer? in this case w okada.
Equipment I have 3 1080TI's laying around and some hardware. My Main desktop is running 32gigs of drr4, intel i9 11th gen and an rtx 3090(24gig vram),
My Idea: connect the graphics cards into a rig(already done) add some kind of software layer to it so I can route all that power into w-okada or something simllar, then conntect the whole thing with fiber optics to my computer to stream the audio,
Why? I want my main computer to run without the AI bloat and such, peak performence, and if I can just use a 2nd machine for all the ai stuff and stream it to my pc, that would be ideal, I was also thinking 3 1080 ti's might also be better then this setup for AI voice changers in general, not mentioning the fact the desktop at the moment needs to run both the voice changer and all other apps I need at the same time.

[simple ore]

1080ti is not ideal for the voice changer, but you can run it

have a second computer with 1080 running the server, connect to use from the browser on your main pc

[ocean holly]

Thats actually a pretty good idea, do you think the 3 1080 ti's can be the 3090?

[simple ore]

i dont think you can use 3x cards at the same time

[ocean holly]

Is the gpu even the deciding factor is all of this?

This is currently my setup using my main computer.

It works very comfortably.

[simple ore]

gpu is important if you want to run a voice changer and a game on the same pc at the same time

[ocean holly]

What if I'd like to build a pc for the AI voice?

[simple ore]

a 3050 or higher would be ideal

[ocean holly]

Hm

[simple ore]

1080 are old, but you can still use one to run the voice changer

[ocean holly]

Oh yeah but the performence will take a huge hit.

[simple ore]

since that's all it is gonna do, there will be no performance issue

game on 3090, vc on 1080 in 2nd pc

[ocean holly]

The only reason I will switch the voice changer to a seperate computer is if the voice changer runs better on the 2nd pc

or the same.

[simple ore]

vc client (browser) on 3090 pc

vc server (the actual voice changing) on 1080 pc

[ocean holly]

Oh yeah I understand that part

[simple ore]

you can just have a second card in your main pc

but uness you got an actual pcie x16 slot with at least x8 bus I would not recommend it.. 1080 on x4 gonna suck

[knotty moth]

so using a single 3090 is better

also RTX 2060 or 3050 is still better than the GTX cards

[ocean holly]

Why would it being real time matter? its pulling resources from the gpu, coulden't it combine the power?
(Note. I could be talking out of my ass, please call me out if that is the case.)

[simple ore]

what?

[ocean holly]

Would a 3050 on a separate machine that is only used for the ai voice be better, or on par with the 3090?

[knotty moth]

mb I meant if in case using multi gpu workload for the voice changer alone, perhaps

which means like utilizing dual T4 in kaggle

[simple ore]

it wont be dual gpu, you pick a gpu to run on in the voice changer

so a heavy game would run on 3090 while vc can run on second card (or second pc)

[ocean holly]

I see! the reason I decided to go with duel gpu's is because I already got a 3 gpu rig setup and running, so I was thinking it would be mega easy to just run ai on it if its better.

[knotty moth]

^this, and I don't think SLI/NVlink would benefit from it

1080 Ti will be just 1080 Ti

[simple ore]

sli is long dead, nvlink requires higher cards

[ocean holly]

Yeah I guess so! if that is the case, then I would have to look into making a computer for it spesficly

If you guys have some time, and are even mildly intrested in it, I would love some help regarding that

[simple ore]

make something out of spare parts

[ocean holly]

I want something that can be as good as my 3090 setup.

I don't have spare parts on that level.

[ocean holly]

These are my current settings, I need a system that can support this at the minimum

[simple ore]

1080 is 8 years old, so something at that level

[knotty moth]

sell those GTX cards and you could get a decent RTX card in exchange

[simple ore]

intel 11th gen is 4 years old

[ocean holly]

Definitely an option.

[ocean holly]

it is old but isnt the cpu a non factor in a machine that only runs AI?

[knotty moth]

I'd still not consider intel 10th gen or the AMD counterpart too old

[simple ore]

well... what's your goal for the main pc?

[ocean holly]

Well I don't care about the main computer at the moment?

I'l talking about the other machine.

[simple ore]

then find something old that can be used to run your 1080

i mean.. gee, my old spare rig with 4790k can run it

[ocean holly]

The main computer does its thing, boot games, do work, etc.
the AI pc needs to be able to run, the, well, AI.
to my understanding, the AI works like this: The better the gpu, the better it works. (faster response time, less glitching, lower delay, etc.)

I don't want to invest in a 2nd computer for ai for it to just run okay, I want something that can run good. really good. at least to the level my current computer runs things.

[simple ore]

there's more to AI than just a realtime voice changer

[ocean holly]

Of course, but I don't need those.

I use the voice changer for work related reasons.

[simple ore]

if you just want to use the voice changer, find a 10-year old piece of shit pc, plug your 1080, enjoy

[ocean holly]

Okay. I see a pattern here.

Maybe I'm just confused!

Let me ask this.

[ocean holly]

If I do as your saying, and plug the 1080 "piece of shit" and run it for solely AI. would the ai, run better, or just as well as the computer I'm currently running.

meaning I can keep the chunk and extra.

the same or change them to even better?

[simple ore]

3090 is 12-cylinder turbo charged engine, you have installed it into a regular car.. okay, it's fne. your 1080 is 6-cylinder engine, find a piece of shit honda civic to install it in and it will be fine

[simple ore]

when you try to run a heavy 3d game and CUDA application on the same GPU at the same time, they compete for resources

unlike CPu where you can assign an application to only some of the cpu cores, there's no such thing for GPUs

so the game is asking the GPU to generate as many frames as possible and the CUDA application is asking it to crunch some nmbers as fast as possibl, that results in weird things

when you have a separate card, even in the same PC, to run CUDA task, everything work fine without delays or lag

[ocean holly]

So that would mean If I did plug the 1080 I could do this for example, and it would work?

[simple ore]

not 2.7ms chunk lol

[ocean holly]

Ptff

[simple ore]

GTX 10xx-series 320 ms chunk + 2.0s extra

RTX xx90 (e.g. 3090) 72 ms chunk + 2.7s extra

[ocean holly]

Okay-

SO!

Now I understand xd

[ocean holly]

The 1080, or in this case the gtx seris, is too slow for me. I want to get something that has at least the same as the 3090.

The 3090 is largly overpriced, are there any equivalent cards you can recommend

would a 40 Series be better?

[simple ore]

i can only tell you to check 1080 and play with the chunk to see how low can you go

since you're going to use it as the dedicated card, it can probably go lower

[hallow thistle]

GeForce RTX 40 GPUs can be better than RTX 30, but their prices can be something.

[simple ore]

used 3070

can probably sell your 1080s for $100-150 each

[hallow thistle]

Second-handed ones, real.

[ocean holly]

Clearance (FOR PARTS) Gigabyte RTX 4090 Windforce (No Core & VRAM) With Box
found this for 135 dollars
https://www.ebay.com/itm/266771335969

[simple ore]

for parts

do you read?

[ocean holly]

English is not my first lang.

[simple ore]

it is as dead as a brick

[ocean holly]

I'm sorry but I don't understand what for parts mean.
Now I do though!

thanks.

[simple ore]

it is the usual chinese leftover

"The core chip and vrams are removed. "

[ocean holly]

Yeah seems like it, thank you

You've been a huge help, and continue being really nice!

Making this whole thing be wayyy easier.

[hallow thistle]

The "No core and VRAM" means there are neither actual chip or VRAM chip present on the "bare board", making the GPU unusable without them. This kind of bare board GPU is usually as donor part or a decoration in your room.

[ocean holly]

Thanks!

Atm, I'm looking in ebay and amazon for used rtx 3070s

as you guys suggested

Finding them for around 300 dollars so far?

[hallow thistle]

So this is what Gigabyte NVIDIA GeForce RTX 4090 board without the chip and VRAM looks like. https://i.ebayimg.com/images/g/vVEAAOSwodJl~gRA/s-l1600.jpg

[ocean holly]

What should I opt for in terms for motherboard cpu or etc

[knotty moth]

you should read the description carefully

[ocean holly]

ah yes that topic is already done

[knotty moth]

in order to avoid falling into trap

or getting scammed

[ocean holly]

I understood my mistake!

Thank you for your worry

[meager pollen]

Can somebody give me there voice changer settings please? Just one that’s efficient

[ocean holly]

Whats your gpu?

[meager pollen]

NVIDIA Geforcce Rtx 2050 (4gb)

Not much to work with

[hallow thistle]

And which W-Okada version are you using? There are two versions of W-Okada.

[meager pollen]

Voice changer client demo

Onnxgpu-cuda

[hallow thistle]

Wow, that's an old W-Okada, outdated. Download and use the better W-Okada from this link instead. https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#download-nvidia-on-windows

[ocean holly]

Btw from what I'm seeing I can probabbly obtain an rtx 5080, do you think that would be better for the AI pc?

I'm kinda stuck on picking all the other parts for it

does it need a cpu? what about the mother board? etc

[hallow thistle]

RTX 50 GPUs are too new. Not every AI program has developed to work on this specific RTX GPU, although a few AI programs like Stable Diffusion can.

[ocean holly]

Then w okada won't work?

I'm building this solely for voice changing

[ashen cairn]

I need to prepare a voice sample to upload to rvc gui. This is a recording of the voice that I want to use in this program. I've been trying for an hour and it doesn't detect it after creating a zip file and importing it. I'll send it in dm because it's not possible here.

[knotty moth]

RTX 50-series doesnt work for applications using pytorch 2.6/older

[ocean holly]

And I assume w okada runs that then?

[hallow thistle]

That doesn't mean W-Okada won't work at all. There's a specific version of fork W-Okada developed for RTX 50 GPU, the program itself is experimental and might have a bug.

[ocean holly]

Oh I see

[knotty moth]

I'd recommend vonovox in the guide

[ocean holly]

I also know this is a hot topic, but is w okada even the best ai realtime voice changer?

[knotty moth]

may not be beginner friendly but it works for your 5080

[ocean holly]

I'll check it

[odd shale]

Hello peko!

[ocean holly]

This whole thing left the beginner friendly a long time ago

I wanna build a computer for an AI voice changer-

Also, separate question, would it be worth it paying someone to make me a voice model based on a database?

[hallow thistle]

The Vonovox, while not as widely known as W-Okada, its UI looks like this. https://raw.githubusercontent.com/dr87/Vonovox/main/docs/images/ui1.png

[ocean holly]

There already is one based on the database I wanna use, but I'd like to find someone who can remake it to be even better

[knotty moth]

with such high end gpu, I'd recommend ryzen 7700X or 9800X3D and an affordable AM5 mobo as possible

[ocean holly]

including more laughs and more to make the whole voice changer feel as "human" as I can.

[ocean holly]

Lovley thanks thats great.

[knotty moth]

try not to laugh challenge

or just non-raspy "hahaha"

[simple ore]

i mean.. geez.. if you're going to get 9800x3d, that's the top tier gaming CPU

using it for a voice changer spare parts PC is a huge overkill

@knotty moth why would you even suggest that?

it would be "upgrade your main pc, move its parts down to spare"

@ocean holly dont get carried away

[knotty moth]

I assumed for general use upgrade plan (main pc), not always specific voice changer

but ye mb the "moving parts down" problem may seem convoluting

[simple ore]

he clearly said he does not want to touch his main pc

which is 4-year old one with 3090

[knotty moth]

then he could keep the 2nd pc's cpu & mobo

[ocean holly]

Hm, is it though? I'm running the voice changer for an avarage of 6 hours a day, everyday, for more then a year.

I'd want this thing to be as good as it can be

[weary lagoon]

Hey hello,
I am having issues training a voice model using the RVC WebUI

please help

[odd shale]

First of all, what's your GPU?

And which RVC version are you using?

[simple ore]

there'a reasonable expense and there's crazy

if you can install a second card into your main pc, for the price of just 9800x3d you can do that

[meager pollen]

THANK YOU SO MUCH JUST SET UP AND IT WORKS SO WELL THERES LITTLE TO NO DELAY AT ALL

[weary lagoon]

NVIDIA RTX 3060 6GB Laptop GPU,
i cloned the RVC repo today. so i guess it's the latest version?

[odd shale]

I'm not sure, but you can also try Applio since it gets updated more often.

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[odd shale]

Right here you got some guides.

[weary lagoon]

Applio? i'll take a look

i tried training via rvc webui on Windows 11 and it failed. now i am on WSL2 Ubuntu 24.04 and i managed to make the model.pth; but I can't make the .index....

Thank you Leo

[simple ore]

you dont need wsl

[weary lagoon]

I just wanted to try in case it worked on ubuntu..

WOW APPLIO LOOKS VERY GOOD

And it sets to my language automatically, no chinese mistranslation gibberish

[pine star]

Is there a link to get started on training voice models?

[odd shale]

Of course.

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[odd shale]

You can start by reading these docs.

[pine star]

thank you!

[odd shale]

You're welcome buddy.

[pine star]

i think this is realy cool how people are making models and it's exciting to know how everything is done behind the curtains

[weary lagoon]

Yeah

[keen crescent]

• rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[elfin wharf]

I wish you all a good hello. I have a question which once has brought me to the limits of my nerves. Does anyone know either a working google colab image to video ai with 60fps and 1080p feature that does not take 192938389283883hours until its generated ? If not then does anyone know any ai that i can use ? One free video per day is enough. All help is greatly appreciated.

[meager pollen]

Anyone know how to remove voices on real time voice changer I think so put too many

[simple ore]

I too like to eat a 12-course gourmet dinner for free once a day.

[ashen cairn]

Please, maybe someone can create a model for RVC GUI based on a sample of my voice, the file is in .wav. I ended up with some squeaky voice and I spent half a day on this. It can be done quickly. I want to create good content and I can't fulfill my dreams.

[elfin wharf]

It seems like you love to provoke other people, no ? I am aware of the costs to run all of this you narrow minded idiot. Normally i would go further into this and explain my reason to why i temporarely look for this but i dont have the patience and power to do this. Its kinda ironic that only assholes are met on "weights" server, your personality is just as rotten as their founders.

[frank crater]

-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

[low shard]

AI needs good hardware, no one is gonna pay thousands of dollars to give you free unlimited power lol

[low shard]

Rvc GUI is extremely outdated, don't use video tutorials they are all old

Also it can't be done quickly, it takes a lot of time

What's your PC GPU?

[cedar surge]

Hello. I've been having issues with Deiteris' W Okada Fork; when I'm playing + recording with OBS, sometimes the VC has this issue where the Voice sounds as if it were "inhaling" while talking or sounds super raspy. I'm guessing this is due to strain on GPU. However, the perf number always stays green (which, to my understanding, is the indicative of the GPU doing more than it can handle). I've tried the troubleshooting of this document (https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/) but nothing seems to completely "solve" the problem. Does anyone have any tips for this?

[analog obsidian]

the voice sounding raspy is a model issue, probably using an asmr model/soft model instead of a regular voice

no way to fix that

just using a model trained with regular natural speech is the fix

[bleak nymph]

i tried a bunch of epochs and all were about the same except the more down i went the worse it sounded

[cedar surge]

Oh nono, it's not asmr and the samples have no whispering nor raspy. Also, it only happens when I game+record+vc, it doesn't happen if the GPU is not strained.

[analog obsidian]

did you trained the model yourself?

[cedar surge]

Yes, I did. It is possible I could've made a mistake somewhere along the way though.

[analog obsidian]

hmm try higher chunk size and extra 2.7s

and perhaps change f0 to fcpe

[cedar surge]

I did. No matter the chunk, it happens from time to time.

[analog obsidian]

enable sup2

random inhaling sounds happen due to noise being present in the input audio

since breaths are technically noise

rvc sometimes mistakes noise for breathing

[cedar surge]

Let me try this

[cedar surge]

I have another Okada VC that doesn't do that (the quality is way shittier though, that's why I wanna make this one work)

[analog obsidian]

i havent compared vonovox vs wokada myself so idk

i think vonovox realtime inference has a completely new realtime code

[crude flame]

its pretty much wokada but fast

[analog obsidian]

wokada uses mainline realtime code iirc

[crude flame]

instead of using fp32 it uses tf32

[cedar surge]

sorry, where is that option?

[pulsar yarrow]

Is there any way to mute the playback of the voice from lighthost? I cant mute it because it also mutes the voice, meaning I cant actually output it anywhere

[cedar surge]

found it

[analog obsidian]

u cant change it in vonox afaik

[cedar surge]

nono, no using vonovox

[analog obsidian]

oh ok

[cedar surge]

Using this one

[analog obsidian]

ah yeah the fork of w-okada, same code but optimized

i have compared them multiple times and they produce the same result

if ur results are very different from each other then ur doing something wrong with one

or could be some weird windows interaction

both apps are extremely buggy and unstable

i wont be surprised if somehow one works but the other not, despite being the same thing

[manic mulch]

Which Real time voice changer do y’all recommend

I’m trying voice.AI

[odd shale]

Don't.

That program is paywalled garbage.

Use either vonovox or deiteris' fork instead.

-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

[honest gate]

hey im tying to use rvc but for some reason stops working as soon as i put on my headset

can any1 help?

[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

[willow grail]

the voice changer dont work fore me how to fix it?

[sudden tree]

hey can someone help me find what pretrain i need

Error(s) in loading state_dict for Synthesizer: i am getting this error

for the yuumy juice wrld model

[knotty moth]

I dont quite understand but you can't train using the existing model

[sudden tree]

im not

im using someone elses model

in applio

[simple ore]

what did you use?

details

screenshots

errors

[sudden tree]

Yuumy juice wrld model

I unstalled

[simple ore]

this says nothing to me

[sudden tree]

it said the n_speakers was 9 and there was no config

I am on applio 3.2.8 bugfix is that why

from february

[knotty moth]

then wdym of this?

[sudden tree]

did he lowkey upload virus

[knotty moth]

pretrain is only needed for training a model (finetuning)

[sudden tree]

i was thinking it was because i didnt have a klm pretrain but I realized you do not need pretrain for model

yep

[simple ore]

can you point me at what you had downloaded and tried to use as pretrain?

[sudden tree]

nothing I only download model

it wont let me use yuumy model tho

[simple ore]

URL

stop dancing around

[sudden tree]

I think its because it cant find a config.json

[knotty moth]

so do u want to use the klm pretrain or the juice wrld model?

you cant use both of them

[sudden tree]

https://huggingface.co/YuumyS/DRFLSinging/resolve/main/DRFL.zip?download=true

[sudden tree]

juice wrld

[simple ore]

okay, so a voice model

you can't use it as a pretrain

[knotty moth]

then go straight to the inference

[sudden tree]

i did it throws that error though

[sudden tree]

yep, my error is it just doesnt let me use the model and throws that state dict error

I am using applio 3.2.8 which should be fine because he released the model when that was the SOTA

[simple ore]

hold on

[knotty moth]

did u change the speaker id? otherwise leave it 0

[sudden tree]

what do you mean by this? I changed nothing just plopped it in and it threw an error for having n_speakers = 9 or something

[simple ore]

works fine as is

[sudden tree]

which version of applio you on?

[simple ore]

3.2.9

ish

[sudden tree]

could my version be causing this?

[knotty moth]

nvm ye just try the latest 3.2.9

[sudden tree]

i just dont wanna install new applio bc it says windows protected your poc

pc

should be fine but its unsigned still

[simple ore]

download compiled version, unzip into C:\Applio

[sudden tree]

oh so I can just update preexisting folder?

thats a lot easier than fresh install ngl

[simple ore]

I dont think it is 3.2.8 issue

[analog obsidian]

wait, it could be a v1 model?

but iirc applio infer works with v1

[simple ore]

dtype=torch.float16)}), ('config', [1025, 32, 192, 192, 768, 2, 6, 3, 0, '1', [3, 7, 11], [[1, 3, 5], [1, 3, 5], [1, 3, 5]], [10, 10, 2, 2], 512, [16, 16, 4, 4], 109, 256, 40000]), ('epoch', 345), ('step', 35190), ('sr', 40000), ('f0', True), ('version', 'v2'), ('creation_date', '2025-03-15T04:26:57.464005'), ('model_hash', '6e68437756a365f420100c31d8971f70a967755ff9670f7a460cab7b74bd4514'), ('dataset_length', '00:21:41'), ('model_name', 'DRFL(Scheming Model v1)'), ('author', 'None'), ('embedder_model', 'contentvec'), ('speakers_id', 1), ('vocoder', 'MRF HiFi-GAN')])

MRF HiFi-GAN

yes, you need 3.2.9

[sudden tree]

oh shit

I am nervous ngl I just tried to install 3.2.9 and i got a very sketch cmd prompt

[true basin]

theres 3.2.9??

[simple ore]

2 months ago

[true basin]

i only see 2s

and 1.5s

[simple ore]

???

[true basin]

like 2.078

2.0.78

[simple ore]

Applio

not realtime stuff

[sudden tree]

thanks for your help @simple ore I really appreciated it man

saved me from confusion

[true basin]

why does

the pitch slide

not move

😔

[granite crest]

Yo guys is there a guide to setting up version 2.14 with an nvidia 3080?

[dim spire]

im only need help turn sims 4 into ai models and make photoshoots for my ai modelling show if someane can help me send me a pm is the user like antm by tyra banks

[viral mason]

just make it yourself, it's easy and there's good free options

[simple ore]

2.14 what?

[pine star]

is it possible to combine two voices and having a new voice while sounding normal?

[analog obsidian]

merge results are random

not consistent

[pine star]

oh so they won't sound like the combination of the two people?

even if the two people sound somewhat similar

[knotty moth]

if you mean the voice changer, you should get here
https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

[analog obsidian]

they're the combination of both models timbre
but the results of the merge cannot be predicted

most of the time the merge ends up sounding unnatural/bad

[crude flame]

If it's two people then the model will just switch between the two depending on the pitch

[analog obsidian]

but its all luck, sometimes it does sound good

[crude flame]

You need more than two

[pine star]

oh oki

[analog obsidian]

yeah keep trying until you get a good result haha

[pine star]

time to crank the slot machine

[analog obsidian]

🔥

[knotty moth]

Vocoflex by dreamtronics is an option if you really want to blend the timbre

[crude flame]

^ don't forget this when merging voices

[restive blaze]

can anyone help answer my question how are 2 providers gpu prices extremely diffrent
like vast ai has a 4090 for 0.30-0.40 but runpod has it as 0.60 does that mean vast is less reliable and cant fulfill the same things runpods can

[simple ore]

different providers, different service.. shared means there may be a wait time

[restive blaze]

but its the dame gpu hows the price that diffrent

is their no catch?

btw its vast ai

the cheaper one is that a bad provider?

[simple ore]

just the level of sharing

[restive blaze]

sorr ybut what does level of sharing mean

[simple ore]

https://cloud.vast.ai/?gpu_option=RTX 4090

[restive blaze]

their 0.336

[simple ore]

runprod: Community Cloud $0.34/hr

[restive blaze]

thats insanely low isnt it?

[simple ore]

Secure Cloud - 0.69

[restive blaze]

no i have it on show secuure cloud only it showes the first one as 0.336

[restive blaze]

yeah ik but vast ai secure cloud has 0.336

for a 4090 24gbvram

is that not insanely low?

[simple ore]

no, about right

[restive blaze]

are you familliar with mistral?

mistral 8b in specific

[simple ore]

there are cheaper clouds too

[restive blaze]

yeah but running mistral 8b is it possible to reach 10 req a sec on a 4090 24gb vram?

[simple ore]

but I guess there may be some shenanigans involved in the calculation of hours

[restive blaze]

as my pc 4070 ti super only reaches
Metric Result Notes
Requested QPS 10 Number of requests sent per second
Actual Completed QPS 1.5 – 2 Requests fully processed per second
Average Latency 3 – 5 seconds Time taken to complete each request
Effective Throughput ~90 – 120 requests/min Completed requests per minute

[restive blaze]

ill check

[restive blaze]

if it preformers like this on a 4070 ti super 16gb vram how much could it preform on a 4090

[simple ore]

mistral 7b?

[restive blaze]

no 8b

[simple ore]

okay, it looks small enough to run with full gpu load on 4070tis

[restive blaze]

but mine wasnt able to

mine was able to achbive 2 req a second

[restive blaze]

im aiming for 10

im aiming for 10 req a second

[simple ore]

not sure how can I check that

[restive blaze]

.

this data could help?

[simple ore]

i mean personally

[restive blaze]

?

[simple ore]

yes, you can train a model, then use it in realtime

[restive blaze]

if it can only handle 2req in a 4070 ti super 16gb vram

ho wmuch do u think it can in a 4090 with 24gb vram

like 8?

[simple ore]

😮

[restive blaze]

?

i dont understand

[simple ore]

Anyway, I can not answer you that

I run stuff locally on 4070tis

[restive blaze]

oh ok ty

[wary shore]

Mind checking my DM? I could offer you free 10 hours' RTX 4090 GPU just want to collect some user feedbacks

[simple ore]

if you train a model using, for example, french speaking audio, the inference using this model, even without an index, may sound with a small accent

having both speaking and singing audio by the same person is better than not

you can include many

3 minutes is barely enough

[normal lotus]

how do i download

[honest gate]

i fixed it, but i use a r7 9800x3d, rtx 5080+rtx pro 6000 blackwell, tiny 11 core 24h2, 128gb ram, and i was trying to do real time voice replacement bc im trans, got it working by running it on my 5080 instead of pro 6000

after reading realized my main gpu wasnt supported so switched to secondary

[knotty moth]

just for the voice changer+gaming?

lmao your money spent on pro 6000 could instead be for three 5090s, also the former is known to have kinda worse optimization on the geforce driver than the 5090.

and so I think ye it may only support the consumer RTX 50-series

im not sure what you're actially using, but vonovox is recommended for your gpu

https://docs.aihub.gg/rvc-voice-changer/local/vonovox/

[cyan elk]

nearly 50 hours later but it finally finished and it works lets gooo 🎉

[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

[deep charm]

if my mic intrrerupts what can i do

like when i speak sometimes it stops

[simple ore]

in discord?

[deep charm]

no like when i try to listen to my voice

in the app

[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

[deep charm]

amd rx580x 8 gb windows 10 girl voice ive used this tut https://www.youtube.com/watch?v=SxdnGxicJOg&t=305s i cant send ss of the program here

[low shard]

Did you even check the comments?

[deep charm]

yes

[low shard]

I said that video uses an over year old program 😭

It has outdated info

There's no updated video tutorial

[low shard]

Uninstall everything, it was just a waste of time

[deep charm]

then whats the new program

[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

You must read the guide

[deep charm]

alr

[low shard]

Uninstall everything ofc

[deep charm]

i got amd tho

[low shard]

Yeah, it works

[deep charm]

i see only the nvidia download

[low shard]

You just gotta read it and get the AMD version

[low shard]

It's literally below that

Also, don't skip parts

Skipping a single part Can fuck up your whole audio system

Be sure you delete the old program, and vb audio cable

Get vac lite

[deep charm]

below nvidia download is opening on windows

[low shard]

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

[deep charm]

i openmed other link

[low shard]

What

[deep charm]

idk

my brain slow

[low shard]

Everything is in the guide, read it carefully

Because messing up can accidentally make you not be able to hear anything from your PC anymore

[deep charm]

do i restart

after installing vcable

[low shard]

Did you click yes?

[deep charm]

yes

[low shard]

Did you set your default audio devices?

[deep charm]

i went there and there was my default

no need to change

idk why

[low shard]

No, you need to change, show a screenshot

[deep charm]

where

i cant send herer

oh i can

[low shard]

That's the specific part that If you mess up fucks up your whole audio system 😭

[low shard]

Check also the playback

[deep charm]

its good

[low shard]

Alr seems good then

Maybe you had it already

[deep charm]

idk

[low shard]

Anyways

Download the AMD version

Be aware that open source ai is complex to run, it isn't like ChatGPT

[deep charm]

alr

i extracted it

now?

i run the file

my antivirus

wait

[low shard]

What anti virus so u have

Put the folder as an exception

The program is open source

You can even check the code yourself

[simple ore]

some antivirus that is worse than a virus

mcafee or norton crap

[deep charm]

norton

[simple ore]

what a joke

[deep charm]

it installed itself

[deep charm]

it closed

nun appeared

and i opened the web link

and it closed

now i got this

what now

[low shard]

You shouldn't close it

It uses a Web User Interface

[deep charm]

it closed itself

wait

lemme reextract it

[low shard]

Extract everything, not just the .exe

[deep charm]

ok

wow

[low shard]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#why-does-it-run-in-a-browser-and-not-its-own-window

[deep charm]

the winrar file got removed

itself

and when i try to download it again

[low shard]

Disable your anti virus

[low shard]

It doesn't do things by itself

[deep charm]

i did

[low shard]

It's your anti virus

[low shard]

Are you on a school PC?

[deep charm]

mine

home

[low shard]

Be sure to fully disable it and add the file and folder as an exception

Norton is noisy asf

[deep charm]

ill just unninstall it

cant even unninstall it

what is happening w my shi

now i dont got permissions

on my c driver

[simple ore]

https://support.norton.com/sp/en/us/home/current/solutions/v15972972

[low shard]

real

[deep charm]

i opened

the file

it said sum

and then it closed

and i got this again

why this so hard

@low shard

what do i do

[low shard]

this is an open source ai program, it's not a product made by a company like ChatGPT

its made by the community, ai isn't user friendly

[low shard]

you sure your anti virus is off?

[deep charm]

ive unninstalled it finnaly

[low shard]

can u show a screen recording of u downloading it, extracting it and opening it?

[simple ore]

with norton you can never be sure

there may still be services and startup tasks leftovers

[deep charm]

lemme try

[low shard]

lmao

[deep charm]

ive looked in task manager

no norton

[simple ore]

and they'll keep nagging you 6 month later

is windows defender enabled?

reboot and make sure it is

[deep charm]

yes

of course its downloading now

what now

@low shard

[low shard]

gpU: ur rx580
chunk: 400
input: microphone
output: line 1
monitor: headphones optionally to hear urself

then add a model https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#adding-models and click start

then set the input in the other program as line 1

[deep charm]

do i change the f0

?

i need help to set up my settings

can u help

?

@low shard

i sound horrible

[low shard]

on wokada, 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)

• Reduce the delay via: https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#reduce-more-delay

be sure to get a good model

not all models are good

also set extra to 1.5

[deep charm]

now i sound off

[low shard]

elaborate

[deep charm]

my friends tell me i sound bad

[low shard]

find a better model

use the suggested models in the guide

or search in over the 20k rvc models

[deep charm]

what guide

[low shard]

the one where you got the program

it has EVERYTHING in it basicallu

[deep charm]

ok

the words

i speak

are weird

thats the thing

its not even saying them correctly

@low shard

how fix

[simple ore]

that's not how it works

it loads the whole dataset, all files, shreds them to little pieces and learns from that

yes

[low shard]

show a screenshot of ur wokada

[low shard]

set extra to 2.0

then

CHECK THIS TRIANGLE ALWAYS

ITS THE MOST IMPORTANT THING ON AMD

[deep charm]

done

[simple ore]

does not matter

[low shard]

if it still sounds bad, try another model and check that ur microphone is good

[deep charm]

can u suggest a model pls?

[simple ore]

what's your gpu?

should not be a problem to handle more

[low shard]

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

[simple ore]

is it running on CPU for some reason?

is it a laptop?

may wanna invest into some proper cooling stand, yes not surprising

AI is a computing intensive task

but I'm make sure it actually runs on GPU

check task manager/performance

does it look like that when training?

that's hot

may wanna get a cooling stand or something or it gonna melt

[low shard]

gotta be rich

[knotty moth]

I have told you to stop using the old mangio

[low shard]

Mangio is outdated asf

it's the same as using windows xp in 2025

mangio is abandoned since 2023

all y tuts are old

[analog obsidian]

applio is not going to fix your hot gpu temps, imo dont train with that, the chances of u melting ur gpu are pretty high

use cloud solutions

ok but be careful, i know someone that actually potentially melted their gpu because of rvc training on a laptop

[crude flame]

i used to train on a 3060 laptop gpu

it never melted

battlefield 5 made it crash though

i did like 20-30 min

[analog obsidian]

maybe you got a better cooling system

or his cooling may be defective/not good enough

imo i'd rather not risking my laptop for a rvc model

[crude flame]

idk, it was really hot

i almost burned myself lol

[brittle wing]

Yo, do you guys know any apps to make rvc models, I'm a beginner?

[viral mason]

You can use Applio on Kaggle, it's all online and free, I can help u btw just dm me

[coarse pier]

whats the difference between okada and vonovox?

[analog obsidian]

vonovox runs better, in simple words
its a completely new thing

the quality is the same between all realtime clients

what changes between them is optimization and quality of life settings

[lime flint]

Hey everyone,

I’m really interested in learning how to build automations for clients — things like automating tasks between apps, setting up workflows, etc. The problem is… I’m a total beginner. I have no idea where to start or what tools I should even be looking at.

I’ve heard of tools like Zapier and n8n, but I don’t really understand the differences, use cases, or how to use them in real-world client projects.

Are there any YouTube channels, tutorials, or other resources you’d recommend for someone who’s just starting out and wants to learn this from scratch?

Any advice, direction, or beginner-friendly content would be super appreciated!

Thanks in advance 🙏

[languid lodge]

is possible to train rvc model with colab?

[simple ore]

yes

[simple ore]

https://www.theregister.com/2025/06/29/ai_agents_fail_a_lot/

[languid lodge]

There's any colab still working? One I've found so far don't work 'cus colab changed the way it works.

[simple ore]

-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

[calm jackal]

I am wondering that my audio files that I training with, do they have to be in wav or mp3 format?

[low shard]

Wav, so they are lossless

[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

[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

[calm jackal]

Ok what colab do you recommend I should use for training?

[low shard]

You should first check that your PC is good enough, colab is only for people with bad pc

What's your PC GPU and operative system?

Don't use cloud as your first option

[viral mason]

I've been told flac and wav back and forth, which is really best for training?

[low shard]

Cloud has limited free time

[calm jackal]

I don't have pc unfortunately, I use my phone

[viral mason]

you're cooked

[calm jackal]

Why?

[low shard]

Both are good, both are lossless, simply FLAC is a bit smaller

[calm jackal]

Why?

[low shard]

Oh damn, your only option is cloud then, but remember it's harder on phone, the user interface isn't made for phones

Train (make) RVC Models on cloud:

1. Prepare the Dataset
2. Setup RVC:
   Choose a cloud way to use RVC,

• Google Colabs (max 4 hours of daily T4 16gb gpu not granted for free, not much hours for training, but easy to use, there's a paid tier):
  • Applio (UI)
  • Mainline (U, broken right now)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus, either T4x2 16gb each or P100 16gb, only free):
  • Applio (UI)
  • Mainline (UI, broken right now)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly, Free Studios run 24/7 but require restart every 4 hours. There's a paid tier):
  • Applio (UI)
  • Mainline (UI, No guide as of right now)

3. Be sure to know about the tensorboard

Google Colab = Easier but risk of getting disconnected
Kaggle = Harder but way more gpu time
If you are looking for the easiest way and for free, is using https://weights.com/ which ofc uses RVC

RVC Inference (use models) on pre-recorded audio on Cloud

You can use either:

• Weights.com: Easiest Possible Ever Automatic
• Ilaria RVC Zero: Fastest free on cloud
• Applio UI Colab: RVC Fork with some extra features like TTS
• RVC AI Cover Maker UI: Automatically Separates the Vocals and Instrumentals, converts the voice and mixes them back

You can use them, you will connect to a remote good PC with limited free time

Just the interface won't be comfortable

[calm jackal]

Ok

[low shard]

Phones aren't powerful at all, AI is intensive asfff, it's like running GTA 6 on a phone

[calm jackal]

Damn

[low shard]

ChatGPT runs on cloud, that's why it works on phones

[calm jackal]

Ah, do you have a pc?

[low shard]

It doesn't run directly on your phone

[low shard]

Yeah why?

[low shard]

You can still make it on your phone ofc

Just not as comfortable as a pc

And not able to do it locally

[calm jackal]

If I may ask you, can you please make it on your pc? My voice model I mean

[low shard]

Sorry but I'm not looking nor have enough time for that, it takes a lot of hours to train a voice model

[calm jackal]

And speaking of phones, I went into gradio when I used applio but when went to the dataset section, i couldn't click on the file

[low shard]

You can either make a free model request in #1159289738314919936 , or make it yourself on cloud

Have you checked first the model isn't already made by someone else?

[calm jackal]

And what cloud are you talking about exactly?

[calm jackal]

No

It hasn't been made by anyone else

[low shard]

The ones I mentioned above #✨│ai-help message

[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

[viral mason]

why does this happen in the UVR hugginface space?

it'll add a pop at the very start for no reason

[calm jackal]

.... @low shard

[low shard]

Alright, if that's so you can make a free request, or read the guides for doing it on cloud

[low shard]

@viscid moss are u aware of this?

[calm jackal]

Ok

Are these the guides?

[viral mason]

it happens mostly on datasets for me although when extracting vocals from songs or just from anything with already existing background music or anything like that it doesn't add that pop

[low shard]

That's our ai hub documentation yes

[calm jackal]

Ok

[simple ore]

load the file in audacity

zoom in to the start

[viral mason]

would using fl studio work as well, I really don't use audacity for editing my datasets

[simple ore]

if it does no start from close to 0 you'll hear the pop

whichever method lets you see the waveform

[viral mason]

ah

[simple ore]

pop happens if you cut he audio wrong

[viral mason]

how do you cut audio wrong?

[simple ore]

like trimming the silence at too high value

[calm jackal]

If I don't have any storage on my phone to do the model on cloud, what can I do?

[simple ore]

this is low enough not to pop

[low shard]

If you don't have any free storage, delete something on your phone

[viral mason]

?

[simple ore]

this is gonna pop

[languid lodge]

any of these is compatible with colab's python 3.11?

[low shard]

Elaborate:

• your PC GPU
• your operative system
• what you want to do
• what tutorial link are you using

@languid lodge cloud shouldn't be your first option, you should check if your pc is good enough first

[calm jackal]

I posted it already, the thread on #1159289738314919936 @low shard

[low shard]

alright, goodluck

[calm jackal]

Yeah let's hope so

[jade tusk]

Hi! I found a character's AI, but it's in English. Is there a way to make it speak Japanese?

[odd shale]

Elaborate about your issue with detail.

Are you referring to an RVC model or some sort of character ai chatbot?

[jade tusk]

RVC model

[odd shale]

You can try with testing a japanese speaking/singing audio with the model.

[split jacinth]

what is now the best way to use use rvc models.. like i talk and the app uses the model
i am still using rvc gui so i need a new one 😭

idc about the time of rendering, i want the one that currently has the better quality, locally

[viral mason]

that is confusing

[viscid moss]

That's weird. It's working for me without that pop sound

[split jacinth]

like I record a mp3 and then the app uses a model which i give and it makes the original mp3 sound like the model

[viral mason]

what

[split jacinth]

nvm

[simple ore]

that's how it works

although