✨│ai-help

[merry kernel]

holy shiet :v

even pitch change?

[low shard]

Not sure if for that too since I don't have an Intel arc

[merry kernel]

oke

[low shard]

But anyways you will get that triangle everytime you need to do that

[merry kernel]

let me tried

[low shard]

Whenever you get the triangle, follow the traingle

God bless the little triangle

[merry kernel]

i have some other question

so i need to

[low shard]

Valorant should work fine at lowest, its not super heavy

[low shard]

Sure

[merry kernel]

open the web all the time for it to work right?

[merry kernel]

i run valo at low setting

chill

im fps gaming
not graphic gaming

[lyric fossil]

yes

[low shard]

Yeah, it uses a Web User interface

[merry kernel]

and

[low shard]

https://rentry.co/forkvoicechangerguide#why-does-it-run-in-a-browser-and-not-its-own-window

[low shard]

Low-key goated

[merry kernel]

i have a fan next to me

it kinda bug my ai voice

eventhough i chane this to max

[simple ore]

the safe load error is because you've installed it from scratch with most current torch while the application is ancient and has unsafe load calls

[merry kernel]

okie

thank you all

for the help

[merry kernel]

but how can i fix this

[lyric fossil]

i have tried local wokada fork rvc to play within r.e.p.o (a less than 1 gb mp game) and my gpu can't handle that. i dont think valorant either

[merry kernel]

lmao :)))

damn

[lyric fossil]

a530m is laptop right?

[merry kernel]

yeah

i have no hope now lol

[lyric fossil]

well you most likely can only run it (locally) without playing game like talking on discord

you can try cloud computing

i use wokada kaggle and now i can run rvc with online multiplayer voice

[knotty moth]

RTX 4060 laptop is recommended spec, unless your budget is a concern

[low shard]

You got an Intel arc 530m GPU?

[lyric fossil]

no, i have a gpu that is equivalent to that intel.

[merry kernel]

bruh

i really need to change back to cpu to change chunk

that hurt

[lyric fossil]

yeah but it's not recommended to use cpu. gpu is more reliable

[merry kernel]

wait

i have 2 gpu on my laptop

i set my arc to play game

the other rest when im gaming

can i change the gpu to run that

[low shard]

He means the triangle

[glacial pollen]

you should be able to

[low shard]

What's the other one?

[merry kernel]

let me check

its on board gpu

bruh

i change the chunk to 750 and it not function anymore lmao

[low shard]

That's 100 times worse lmfaoo

That's integrated graphics

Extremely more weak

[merry kernel]

wait

why it burn

[low shard]

Wdym

[merry kernel]

both my cpu at the same time

i mean gpu

[knotty moth]

what

[merry kernel]

wait

let me send you

huh

it stop working

lmao

ah there we go

it using both

[low shard]

Huh did you set the GPU to Intel arc

If you want

[merry kernel]

yes

[low shard]

Instead of locally (runs on your hardware)

There's cloud (remote good PC(

But that got limitations for free

And also depends how fast is your internet speed

[merry kernel]

nah im too lazy for that

and fuk my internet btw

it suck

it some how take 2min to boots up the voice

and it missing some of my voice :v

it need to upgrade my laptop fr

[lyric fossil]

is rmvpe_onnx obsolete than the normal rmvpe?

[low shard]

I mean you can't expect to run much well ai with a weak PC and shitty internet tbf

The same with games

[low shard]

No, thats just the version for non Nvidia

[merry kernel]

okie

thanks for the help

[low shard]

I mean, for now it can kinda work for discord, just be sure to set the chunk higher than the perf and to close useless programs

Also be sure to set line 1 as input for discord

[merry kernel]

its kinda un useable to play game

it burn my gpu

[low shard]

Yeah your laptop isn't that powerful, it's just a laptop

[merry kernel]

yeah

[low shard]

But for discord can work

[elfin wharf]

Hi everyone,

I’m trying to run the Mangio-RVC WebUI on Windows (v23.7.0, downloaded the _INFER_TRAIN.7z bundle) by double-clicking go-web.bat, but it immediately spits out errors and never brings up the Gradio UI. Here’s what I’m seeing in the console:

call venv\Scripts\activate
infer-web.py --share --port 7867
usage: infer-web.py [-h] [--device DEVICE] …
infer-web.py: error: unrecognized arguments: --share --port 7867

CUDA initialization: The NVIDIA driver on your system is too old (found version 98304).
Please update your GPU driver or install a PyTorch version compatible with your driver.

sqlite3.OperationalError: unable to open database file

[merry kernel]

yeah for sure

[glacial pollen]

Alright so, you should get the newest applio or my fork if you wish @elfin wharf

Think mine's gonna be easier for you with ui
if you're completely clueless

[merry kernel]

it kinda buggy though

oh

is there any beatrice model

[elfin wharf]

alr imma give it try or smth

[merry kernel]

it work fine with my pc

[glacial pollen]

https://github.com/codename0og/codename-rvc-fork-3/releases/tag/v3.1.0-rev2
Here it is

[merry kernel]

i mean laptop

[glacial pollen]

Instructions are simple:
https://github.com/codename0og/codename-rvc-fork-3

install then run

tho, prior to that

what is your gpu man?

@elfin wharf

[elfin wharf]

thats a really good question, my pc is really fucking old so i might aswell take a quick look

[glacial pollen]

yea go take a look and once you're done, @ me

[low shard]

Beatrice models are way more lightweight, bc they suck in quality compared to rvc

No one trains then so good luck to even find any other Beatrice model

[merry kernel]

oh

shiet :v

[low shard]

Also Beatrice models are only for original wokada, they are removed from the deiteris fork bc they suck in quality

[elfin wharf]

Intel(R) Core(TM) i7 CPU 860

[low shard]

I would just suggest you to uninstall the original Wokada

[elfin wharf]

@glacial pollen

[low shard]

That's a CPU not GPU

[glacial pollen]

Alr, so what is the gpu

a

well, that's the cpu yea

gpu would be like so

unless you don't see anything then.. well

uhhh, you're out of luck

[low shard]

@elfin wharf use task manager, check performance tab

[glacial pollen]

^

[low shard]

Mangio rvc is abandoned since 2023

Never follow video tutorials for rvc

Or for ai generally

AI progresses at Sonic speed

[elfin wharf]

GTX 560 TI

[low shard]

Damn, you sure it's that?

That GPU is extremely outdated

[elfin wharf]

i am texin on my pc which is hard asf for me rn

[low shard]

That GPU is too old for even to get recognized in the program

[glacial pollen]

well, not only it's outdated but 1 gig of vram...

[low shard]

It's ancient

[elfin wharf]

as i said my pc is very old and probably witnessed the rise and fall of the roman empire

[glacial pollen]

and lack of pytorch support

yea, you're out of luck buddy

[low shard]

Yup, too weak to do anything

[glacial pollen]

you'll need to suffice with cloud / colab etc

[elfin wharf]

good to know i wasted my time for weeks lol

[low shard]

Your only way is cloud (remote good PC) which is more easy to break and has limitations in free tier

[low shard]

What are you looking to do? You can try cloud

[glacial pollen]

Either cloud or just commission anyone
yup

Guess I should legit think of building kaggle of my fork

oof

eventually

[low shard]

You don't know over half of RVC users have a bad PC?

[glacial pollen]

@low shard is encryption needed there too? btw

[elfin wharf]

nevermind ima except it being not possible for me lol

[glacial pollen]

I mean yea, true

[low shard]

You can do it on cloud

Just not locally on your pc

But yes u can do it on cloud

Cloud is like sites

[elfin wharf]

dont have the money for that

[low shard]

There's free ones

[elfin wharf]

still thx

[glacial pollen]

there's free alternatives

with minor or major limits, depending on what u gonna use

it's not that bad tho

[low shard]

Like colab with max 4 hours GPU daily (that can be random, not granted)

[glacial pollen]

Before I got a decent gpu that's what I'd do years ago

[low shard]

So yeah it's possible for free

[glacial pollen]

So def worth giving a shot

[low shard]

Just you don't have unlimited granted time lol

[low shard]

Of course

[elfin wharf]

i dont know what i have to use thats the thing. i want to train a model with audio input or smth idk

[low shard]

Then RVC

Train (make) RVC Models on cloud:

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

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio (UI)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • 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

mainline is broken rn

But the other things work

[glacial pollen]

Yeah, you can use the clouds as suggested.
Then you can perhaps use the models on your cpu ( gonna be hella hella slow but it is do-able. Afaik pytorch has no strict requirements for cpu inference, in most cases )

[low shard]

Hopefully that will work, other RVCs aren't having issues but for example Wokada deiteris fork gets caught even with encryption

[glacial pollen]

yeah but like, since when encryption's needed for kaggle?

I thought it was just colab, mainly

[low shard]

Since always, kaggle is owned by google

[glacial pollen]

Is kaggle now having a fuss with rvc?

oh

welp

Makes sense then

[elfin wharf]

i think i have tried osmething liek that before, i got errors that i dont even understand since i have zero python knowledge and dont know anyone who is a pro atz this

[glacial pollen]

Thought the restrictions were a lil more loose on kaggle ngl

[low shard]

Kaggle directly kicks you out of your profile and bans you if you get caught

[low shard]

Absolutely not, the contrary

[elfin wharf]

but imma try what u guys suggested

thx

[glacial pollen]

I see... well thanks for info

[low shard]

That's because you used video tutorials, those are old

[low shard]

Let us know

[glacial pollen]

unfortunately ^

[low shard]

You're welcome and good luck

I'm so glad I got a desktop

[elfin wharf]

i also used video tutorials and tried others without lol i think its cause i dont understand the errors

[low shard]

I assure you cloud is a pain

[elfin wharf]

yes i willl

[low shard]

Video tutorials are outdated

Don't ever follow them for rvc

That's why you got errors lol

[elfin wharf]

got it thanks

[low shard]

Yw

[soft eagle]

Hello, first I am sorry if this is the wrong channel, I had downloaded the w okada a while ago (I suppose around july 2024) and it is the v.1.5.3.18.a version. Should I update it (as in are there some necessities in the new/newer version for some voices) or is it fine if I leave it be? If it's best I update it, how should I do it or where should I look? Every video or tutorial always seems to be saying different things so I am a bit confused

Is it even possible to update or will it require an uninstall and a new install

[knotty moth]

the fork ver is recommended https://rentry.co/forkvoicechangerguide

[soft eagle]

I see thanks, I'll look into it

Just to make sure: the b2332 right? There is quite a few

[pastel lichen]

GUYS, i m new here. I need to downald applio to use the models right?

[low shard]

yes

[low shard]

for pre-recorded audios only, yes

but

what's your pc gpu?

[pastel lichen]

its just a laptop whit window 11

[low shard]

AI is an intensive task, you can't just run it on a random laptop lol

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

check your pc gpu

it's not like chatgpt, chatgpt runs on cloud (remote good pc), installing RVC (Open Source AI) locally, will mean that it runs locally on your hardware

and you need good hardware to run AI, just like you need good hardware to run games or video editing for example

[pastel lichen]

Intel(R) UHD Graphics 620

[craggy bough]

☠️

[low shard]

dang, is there no other gpu?

that's integrated graphics, weak asf

wouldn't even be recognized from the program

if there's no other, it would just run on your cpu, which isn't good since it will take A LOT of time

[knotty moth]

unusable, if there's no dedicated Nvidia/amd gpu, you could only run in cpu mode

[low shard]

it's better you use cloud (remote good pc), ofcourse this has limitations even if the program is free and open source, because you're using someone's else pc gpu service

@pastel lichen you're looking only to inference (use models) for pre-recorded audios, right?

[pastel lichen]

yes

[low shard]

Train (make) RVC Models on cloud:

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

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio (UI)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • 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

mainline is broken right now

but the rest should work

don't expect unlimited free time ofcourse

it runs on a remote server with good pcs

[pastel lichen]

OOO OK OK THX

[low shard]

you're welcome, and let me know for any issues

[elfin wharf]

sorry to bother, does someone know what i did wrong or what isnt working ? this is usually the point i would just give up

[low shard]

what cloud link are you using?

[elfin wharf]

google collab mainline

the hina one if that helps

[elfin wharf]

omg how embaressing

lets pretend that didnt happen

thx

[low shard]

lol, lmk

[soft eagle]

So normally we use the start file, no? I don't see it in the fork extracted files... is the process different? And do I just delete the entire old folder?

[knotty moth]

run the exe file

[low shard]

just run the .exe, and ofcourse delete the old original wokada

[soft eagle]

Thanks

[knotty moth]

note that it will open in a browser and you have to allow mic permission

[low shard]

and if you're wondering why it opens in a browser, no it's not a website and it still runs on your pc, it just uses a web user interface: https://rentry.co/forkvoicechangerguide#why-does-it-run-in-a-browser-and-not-its-own-window

[elfin wharf]

lol aint gettin past that stage. am i not supposed to copy and paste the token int the key thingy ?

[low shard]

no not in the key thingy in colab, that was just for mainline

you need to put it in ngrok_token

or, use a different method like gradio

[elfin wharf]

imma use gradio

[low shard]

yeah it doesn't need a token

[elfin wharf]

sorry that i keep spamming with images, i really have no clue whatsoever. do i have to select anything ? i ran every cell prior to this, and i got this error with discord or smth

i apologise for my stupidity once again...should have clicked the link

[low shard]

the links I gave you have the guides for the tools, be sure to read them

[elfin wharf]

yes i saw it, thx, i am currently trying to export the model files cause i keep getting errors it cant export to drive, but besides that i think it worked so thank you very much

[elfin wharf]

nope no chance at uploading

i am restarting it from new. shits pissing me off at this point

[glacial pollen]

yea, that's how it is

frustrating at times

[elfin wharf]

yup cant mount drive

[glacial pollen]

and it's why kaggle is typically more friendly, I suppose

[elfin wharf]

i am trying for 2 weeks now. i have never been this angry

[glacial pollen]

it's fine. Imagine I had to go through that procedure around 80 times, back in the day

or 90, for one specific character

lost progress like 5 times in total

oof

[elfin wharf]

yea that sums it up pretty good

[glacial pollen]

feels bad then ye

[elfin wharf]

plus i have no python understanding

[glacial pollen]

you can always copy logs to here

we'll be glad to help you understand the error

[crude flame]

just download the files you need, no need to export to drive

[elfin wharf]

cant even do that i am already restarting

anyone know what the f its problem is ? my drive is empty

i am close to just destroy my monitor lol

[crude flame]

You don’t have to mount it to drive

[elfin wharf]

i just realised its ass. i need to use it for commercial use. and i am going to be honest. fuck this stupid shit. i am done.

from what i am getting it doesnt permit it

completely wasted my time once again. thx anyway to those replying to me

[fair sparrow]

why is my voice stuttering? like it's cutting off each second. I'm using AMD.

[agile kelp]

the cmd is downloading stuff , are they necessary or they are smth else

[low shard]

the models quality depend on the other users that trained them, also it ain't easy to train a model, it's not a 1 click job lol

[elfin wharf]

i never got to test it out so i dont know about the quality

it doesnt matter anyway. no MIT license makes it useless to me. but still thx for trying to help

[low shard]

elaborate:

• what's your pc gpu
• what tutorial link are you using
• a screenshot of the program
• what program are you using it with?

[low shard]

elaborate:

• your pc gpu
• what you want to do
• what tutorial link are you using

[low shard]

so how can you say it's ass lmao

AI isn't a 1 job click or just chatgpt

[agile kelp]

its working now , is their a proper settings for any model ?

[low shard]

RVC & Applio are both MIT licensed

[elfin wharf]

i aint talkin bout the model. i was talkin about spending multiple weeks to find out its not possible. and guess what, i aint an informatics teacher or python crack like u, i dont understand programming

[elfin wharf]

i read something else. anyway i wont try anymore. i still appreciate you helping out

[agile kelp]

https://rentry.co/forkvoicechangerguide and i used this

[low shard]

elaborate the things I said, there is no perfect settings for every gpu

[agile kelp]

RTX 3070TI laptop
i want to understand the settings
link followed : https://rentry.co/forkvoicechangerguide

[low shard]

that's good, it was just downloading internal files

[low shard]

set chunk to 128
extra: 2.7

input: microphone
output: line 1
monitor (optional to hear urself): headphones

you can optionally use force fp32 mode in advanced settings for better quality and stability but more delay

you installed vac lite at the 3rd step right?

also be sure to close useless programs in background

[agile kelp]

not yet i'll do it

[low shard]

and what program are you gonna use it in?

[agile kelp]

Discord

[low shard]

alr, set discord input to line 1

[low shard]

A VAC (Virtual Audio Cable) makes a fake audio device, used to re-route the audio of different programs

In Wokada context, it's used to get the output of wokada as the input in other programs

TL;DR: you have to

you can't find any line 1 rn bc you don't have vac lite

[agile kelp]

ahh virtual audio cable ye ye i installed it

i'll follow your steps now thx

[low shard]

you could follow more complex steps for less delay btw https://rentry.co/forkvoicechangerguide#reduce-more-delay

[jolly veldt]

i installed clownfish and i saw that they added the option to use with the vc client, but why they did add it? does it change anything?

[low shard]

we don't use clownfish here

we use wokada deiteris fork, program designed to run RVC (speech to speech, retrieval-based-voice-conversion) models in realtime inference (use models)

[jolly veldt]

ik

[low shard]

you might have heard of wokada as MMVCserversio or vcclient on video tutorials

be sure to never use video tutorials are they are old

we commonly call it wokada

[jolly veldt]

i know about the vc client

[low shard]

recently there's a new version

[jolly veldt]

can my pc even start the program tho?

[low shard]

the deiteris fork, a modified version which is the best

[low shard]

what's your pc gpu

[jolly veldt]

gpu: gtx 1050 ti

16gb ddr3 ram

[low shard]

it can, just not intensive games

[jolly veldt]

wdym

can i run the vc and fivem at the same time?

[low shard]

it works, but will have massive spike lags on games like marvel rivals

ehh not sure how intensive is fivem

worth a try maybe

[jolly veldt]

i will lyk

[low shard]

ofcourse you will have to set game graphics to the minimum

[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

don't use other things off youtube

[jolly veldt]

i always run it low

[jolly veldt]

i was not using other things from yt tho

my first instinct was to use clownfish, cause i cant run ai voices

[low shard]

if your pc struggles to even run the minimum graphics alone, then yeah it won't work for wokada

[jolly veldt]

any workarounds?

can i create my website and host it there?

[low shard]

• buy a better pc
• use cloud (remote good pc), which break more easily (bc they get updated from the cloud provider) and have limited free gpu time

[jolly veldt]

• i dont have money
• i need it for long periods

[low shard]

you seriously need it to use it more than 30 hours weekly?

[jolly veldt]

can't i just, get a vps and host my ai there?

[jolly veldt]

err...

[low shard]

kaggle offers 30 hours weekly of good gpu

for free

[jolly veldt]

lets just say that...

i might play a little too much

like

5 - 6 hours a day

if its weekends

[low shard]

😭

[jolly veldt]

what do you expect

i still go to high school

[low shard]

I mean you could also use any other paid cloud service provider, but you will need to figure out how to run it there on your own

or you could pay for more gpu time on google colab, and use the wokada deiteris fork colab

[low shard]

same

[jolly veldt]

parla italiano zi

ho visto mo

[low shard]

lol, sorry but the server is english only for not having any moderation issues among staffers

[jolly veldt]

you speak italian

?

[low shard]

yea, but we set those rules because it adds our staff team just more trouble if everyone speaks whatever language they known, I know italian but it wouldn't be fair to just let you speak italian with me on the server but not other russian members speak russian (bc we can't moderate it) for example

[jolly veldt]

oh okay

[low shard]

yeah it's just for more fairness in moderation, sorry about that but we think everyone should be treated equally, I hope you understand

[indigo dagger]

ok i have a new question, all my genorations with the rvc ai applio sound like crap when i try to get a voice to do a cover of a song

[low shard]

@jolly veldt btw, your pc gpu would be enough to run wokada deiteris fork only in calls basically

about games, since you need it for alot of time, it's best you either:

• pay google colab
• use your own cloud provider (but you'd need to be careful since you'd need the commands to download and start the program, maybe you could try importing the wokada deiteris fork colab notebook)

[low shard]

try other models and be sure you cleaned well the vocals of the song

not all rvc models are high quality since they are made by the community

[indigo dagger]

i did thay all sound terrible

Twilight Sparkle (My Little Pony) (RVC V2, RMVPE) (250 Epochs)

Rainbow Dash (My Little Pony) (RVC V2, RMVPE) (400 Epochs)

My Little Pony: Friendship Is Magic (Shining Armor) (E480)(Harvest)

thay all sound horrable

[low shard]

harvest is an outdated pitch extraction algorithm

[low shard]

I can't do anything about it, it all depends on how good the other users train those models

maybe try training your own

also reminder that you need to clean those songs vocals good, and that rvc has limitations on things like laughing

[indigo dagger]

i do i use an ai to separate the lyrics with the background music

[crude flame]

-# harvest isnt actually that bad

[low shard]

what do you use?

you use what we suggest in https://docs.aihub.gg/rvc/resources/dataset-isolation/#vocal-isolation--cleaning right?

[low shard]

we don't even accept it anymore

[crude flame]

yea

but

[low shard]

but rmvpe is goated 🔥

[crude flame]

true

however

your mother harvest can add nice bass to the model

[indigo dagger]

https://plugins.audacityteam.org/ai-plugins/ai-plugins

[fleet raven]

can someone help me download Gabox's voc_fv4?

[low shard]

it's better you use the docs I sent you

[indigo dagger]

ok, what else shuold i try if that dose not work

[low shard]

the docs have multiple options

with multiple models

please read it out

[fervent rover]

Hey y’all it is be an while since last time, but I just wanna to know, is RVC Mainline Colab Finally Worked and Fixed, I Just Wanna To Know

[indigo dagger]

Ok thanks for nothing, I did everything it still sounds horrible

[low shard]

nope, also for that you can just check #📰│dev-updates

[low shard]

i also just checked some of the models you said, yeah they aren't super good quality, can't do much about it other than trying to find others or making your own

[indigo dagger]

Ok, I'll have to learn how to make it

[fervent rover]

Oh Okay, I just wanna to know if it works or not but I guess it didn’t work I guess But Thanks

[low shard]

https://docs.aihub.gg

[brisk steeple]

you guys know why the MMSCServerSIO doesn't open in the browser ? im sorry im new in all this

[brittle wing]

Again how do I resume on Applio no UI colab after switching accounts? @simple ore sorry for pinging but I've been asking that question several times already

[simple ore]

load the backup

run install steps

then start training

[bronze pecan]

where in gods name do I go to download the latest version!?

[stable ferry]

Hello, can someone please send me John Rzeznik version of the group Goo Goo Dolls?

[surreal latch]

Can someone explain this error to me?

will it affect training?

[bronze pecan]

I jsut want to instal the program QQ

where did the files go!?

there are piles of Wieghts and training guids but FA on how to install the client 😦

[surreal latch]

Are you referring to ApplioBackup?

[bronze pecan]

I want to use a voice model

there is software that does that

I need to put it on a computer

there is no where I can find to get it

[surreal latch]

oh, sorry

[bronze pecan]

all the howtos are on trainging

there used to be a link to pre-built binaries

they are gone

IDK Hot TF you guys are doing anything as is 😦

MMVCServerSIO prebuilt binaires

[simple ore]

there's a typo in the script

needs to be ==

[surreal latch]

Is this where I should modify?

[simple ore]

yeah

[surreal latch]

Thank you so much

[simple ore]

fixed it in the repo, so should be good now

[bronze pecan]

any chance you know where I can get the client?

is this still the latest? https://huggingface.co/wok000/vcclient000/tree/218057d3d625b410d0afb1c569948d64987c8fbf

or like?

[simple ore]

all you need to do is to jus scroll up to get the link

https://rentry.co/forkvoicechangerguide

[bronze pecan]

well I couldn't find it so I asked and asked and asked

[eternal aurora]

damn it im still getting this issue

@low shard

[low shard]

elaborate the issue

[eternal aurora]

So basically I been making my own voice right

And I want to use Apollio for it

So I’m on kaggle doing it

[low shard]

Did you use the applio kaggle link from github

[eternal aurora]

And when I do the steps it comes with this issue when I press the link of Apollo

[eternal aurora]

Yep

[low shard]

You mean applio?

[eternal aurora]

Yh it keeps on auto correcting lmao

[low shard]

Show a screenshot of the kaggle output and tell all the steps you did

[eternal aurora]

Issue is I’m abt to sleep and I just saw this would it be fine if I show u tmrw? I’ll ping u

[elder coral]

How do you use pugins in Applio

[simple ore]

there are only two - 1) elevenlabs, which is useless without APi key 2) uvr, which is somewhat broken and not up to date

[elder coral]

How to use elevenlabs

[simple ore]

generally you just download the .zip from release page and then drop it into plugins tab on ui

and re-start

[elder coral]

I get errors after installing the plugin maybe I need to use the api to continue

[simple ore]

No

did you download Elevenlabs-TTS.zip ?

[elder coral]

yes and i got errors on the text-to-speech model

[simple ore]

without the api key you get

[elder coral]

How do i make the model train so fast

and not overtrained

My voice will sound like bill gates robot

[outer isle]

Not enough data present in the training set. Perhaps you forgot to slice the audio files in preprocess? What This?

[valid spruce]

Hi guys

Gn

What is the difference between the hifigan vocolder and the standard Applio vocolder?

[brittle wing]

Yes but what about the initial cells.

Should I preprocess dataset etc

[elder coral]

this is not sliced there are voicelines one by one

there are 8 audios

[outer isle]

I put audios files in the audio slicer, is that ok?

[elder coral]

Yes

[brittle wing]

So... installation, load a backup & train?

[simple ore]

No, they are loaded from backup

[brittle wing]

Exactly

Can you tell me the order

First install, the backup and train at last right

[weary socket]

Hi friends, anyone knows how to make the model less sensitive to background noise?

[brittle wing]

Use Melroformer denoise 2

What about this cell

It's essential I see

[weary socket]

how does it work? do I add it during inference?

I am using realtime gui btw

[brittle wing]

You process the denoised, deechoed vocals for a dataset

[weary socket]

oh, my dataset is clean

[brittle wing]

Be more specific are you making your own dataset of a model

And?

[weary socket]

just during inference, the air-conditioner noise is being picked up quite often

[brittle wing]

It's probably from the dataset

Depends on dataset quality

[weary socket]

the dataset is studio quality, I confirmed it. the noise was translated to random human vocal

[brittle wing]

"the noise was translated to random human vocal"

?

Model artifact

What's the model.Is it on weights

[weary socket]

i mean random voice with breathing and stuff

[brittle wing]

So I can check the audio samples

[brittle wing]

A voice out of place

You mean?

Well that's because you didn't cut it off

From your audio samples

What's the model

[weary socket]

RVC v2

[brittle wing]

The model name.

[weary socket]

I just trained it

[brittle wing]

So I can find it on weights

....

Then there's no way I can tell you

[weary socket]

I can send you the pth

[brittle wing]

You can send me an audio sample

Pth, index

[weary socket]

there is too much audio sample

[brittle wing]

Send one.

[weary socket]

i didnt train index...

is that why?

I mean

I tried to train

but it result failed during inference

[brittle wing]

You should have

[weary socket]

I used this

[knotty moth]

the audio to infer should also be denoised

[brittle wing]

Jesus Christ alive

[brittle wing]

Oh hi any tips on how I should process my audio for inference

[brittle wing]

That's a script?

[weary socket]

I used input noice reduction option in RVC gui

[weary socket]

I thought it will work.

[knotty moth]

-gui

[patent trellis]

RVC Easy GUI

This program is pretty abandoned and old, don't use it.

[weary socket]

just kmeans clustering

[brittle wing]

Christ alive-

[analog obsidian]

train a model with noise

[weary socket]

what should i use then

[weary socket]

really?

[brittle wing]

Applio

[analog obsidian]

yeah, actually helps

[knotty moth]

show the screenshot of what you're using

[analog obsidian]

the result will be noisy but the model is going to be able to handle noise

[brittle wing]

Wait it has index!

[weary socket]

is that for realtime conversion?

[weary socket]

it does not work

[knotty moth]

the fork version of wokada is more recommended https://rentry.co/forkvoicechangerguide

[weary socket]

great to know

thx

[knotty moth]

also regarding to handle the mic noise, try using 3rd party noise suppression if the built-in one is not enough

[weary socket]

any suggestions for the 3rd party software?

[brittle wing]

I'm just sayin-

That your model has an index to it

[weary socket]

fine. any valid script to train the appropriate index?

[knotty moth]

any of mainline RVC or Applio

[weary socket]

I did not find that script in RVC webui repo

will check applio

[knotty moth]

easily found in the training tab

[knotty moth]

I mean you need to do preprocessing and feature extraction for the model dataset you want to train on

[weary socket]

I used script to do that part

ans also training

under the infer/module/training there are all necessary script I guess, except for training index

Have not tried the webui yet

[knotty moth]

unless you're using an outdated shit, ctrl+F the "index"

should be there in the webui ones

[weary socket]

I just modified the extract_index file from applio to fit the RVC webui, works good

also I feel the applio's code are clean and easy to read

RVC webui repo really gave me a hard time reading

[knotty moth]

which RVC webui that doesn't have index training? pls screenshot it

!give-media-perms 20m @weary socket

[weary socket]

maybe i just didn't spot it

[knotty moth]

that's what ctrl+F could save yourself from the hassle

[weary socket]

at least not under infer

too much files contain index in the root

[knotty moth]

are you running it in a VM container that doesn't allow opening the webui in the browser?

[weary socket]

yes

I can, but for security reason I cannot

[knotty moth]

even if so, go-web.bat in the mainline RVC should show the link address from the console to open

though I have no idea on working it in docker shit or whatever at all

[weary socket]

you right, I found it

its in the webui script

embeded

shit code

defined in the infer_web.py

lmao

[knotty moth]

you can blame rvc boss as the developer but they no longer maintain on it

[brisk steeple]

Hii quick question, what I need to put here so I can start using this ?

In the input, output and monitor ?

i follow the guide and everything

and i have the vac thingy for the Line 1

[weary socket]

the devices you want to use for input output and monitor, find them in the dropdown menu

for me it's microphone, voicemeeter aux input and None for monitor

[hallow thistle]

Use "Virtual Audio Cable" lite instead of VB-Cable or Voicemeeter. VB-Cable gives random issues to Windows users when they wanna use it with W-Okada.

[hallow thistle]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

• Don't simply mention your issue, like "my rvc is not working".
• Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
• The more context, the better.

__**BE POLITE.**__ <:matsuripray:1159685390156967936>

• Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
• It's okay if you're frustrated, but don't take it into this server.
• Don't DM without prior consent.

__**BE PRODUCTIVE.**__ 🤝

• Don't ask for every little instruction. Put your own effort & test things by yourself.
• Don't ask to ask.
• Check if your answer is a Google search away/on our guides website.

If you've changed your mind about your message, feel free to delete/edit it.

[hallow thistle]

Which W-Okada version are you using? I use fork W-Okada.

[weary socket]

Voicemeeter works for me so far, what is the random issues specifically?

[hallow thistle]

Most of people here complained to me VB-Cable and Voicemeeter not working with W-Okada and did not output any audio signal.

[weary socket]

🫠 I guess they did not config it correctly. It also took me some time to learn this app

[hallow thistle]

[weary socket]

I thought voicemeeter is not working initially, but it happens that I did not enable b2 for the aux output

But good to know virtual audio cable. Will switch to that if I ran into other issues

[knotty moth]

they mean the virtual cable, including the built-in one in voicemeeter (not the whole software itself)

[lone abyss]

hiii

any working colab rvc interface? 😔

[hallow thistle]

RVC can work locally. What is your PC GPU?

[lone abyss]

I have one of the default gpu's

intel arc graphics

[hallow thistle]

Intel Arc is a dedicated GPU. I'm not sure if any working RVC program can work with this specific GPU.

[lone abyss]

yeah ik that's why I always depended on the colabs

but what happened

cuz since rvc began in early 2023 there were always issues but they just made a new one

[hallow thistle]

RVC program like Applio will still work anyway, but it will use CPU instead.

[lone abyss]

do you have a link?

cuz everything I tried didn't work

[hallow thistle]

There's Kaggle, but you'll have to register with your phone number on this one. The thing is main notebook for Applio on Kaggle is outdated, so you'll have to look in this thread #1364884749482590209 message on how to import the notebook link to your Kaggle.

[elder coral]

My GPU is not good enough maybe Imacs are great for making voice models

[hallow thistle]

[elder coral]

If it took so long my model might be

[elder coral]

do i need a macbook for bette rpurposes of trianing model?

[willow crest]

guys can someone help me please

i cannot find the download file to the Realtime Voice changer

can someone send it please

[woeful obsidian]

New here
I downloaded a model that dosent have an index file
How do i use the webui to create one

[hallow thistle]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

• Don't simply mention your issue, like "my rvc is not working".
• Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
• The more context, the better.

__**BE POLITE.**__ <:matsuripray:1159685390156967936>

• Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
• It's okay if you're frustrated, but don't take it into this server.
• Don't DM without prior consent.

__**BE PRODUCTIVE.**__ 🤝

• Don't ask for every little instruction. Put your own effort & test things by yourself.
• Don't ask to ask.
• Check if your answer is a Google search away/on our guides website.

If you've changed your mind about your message, feel free to delete/edit it.

[hallow thistle]

Don't try to ask anyone to ask you like that. What is your PC GPU?

[hallow thistle]

If one RVC voice model doesn't come with an index file, it's pretty much it. Voice model trainer/uploader might have not make an index file during training process. Although the voice model can work without index file anyway on RVC.

[willow crest]

i have a r5 5600x and a NVIDIA RTX 3060 as Grapics card

I Downloaded now the git-lfs-windows-v3.6.1 file but if i try to open it just give me that error. "Coould not find Git; can not register Git LFS"

[hallow thistle]

Also, you've already got help in #1192011222023950368 make sure to stay on that one.

[willow crest]

yes then lets start writing there

[hallow thistle]

Git isn't needed to install W-Okada, unless you wanna make your own W-Okada or other Python related program.

[woeful obsidian]

but is it possible to create it in rvc webui as i see a train feature index along side train model and one click training

because when loading the g or d model in webui i get error in index option and cant convert the song

[hallow thistle]

You think it's possible to make an index file with no given dataset audio? I don't think any RVC program would work like that.

[woeful obsidian]

totaly new to this ai stuff
so if i cant make a index from the pth how can i use the pth as the webui gives an error
do i use the built in index in logs like guanguanv1?

[hallow thistle]

If the RVC fork program you're using requires you to use index file, you can borrow an index file from any voice model that has it, but the result voice model won't sound as it should be.

[woeful obsidian]

as in any index or a index of the same artist just form a diff version

[low shard]

absolutely not, Mac doesn't even have training support lol

Nvidia GPUs with 8gb atleast VRAM are the best for RVC training

Nvidia GPUs are the best for ML

[low shard]

What's your PC GPU

@lone abyss #🧬│ai-chat message yep you're cooked

[lone abyss]

well

I guess I'll be using the bad weights interface forever

[low shard]

Train (make) RVC Models on cloud:

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

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio (UI)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • 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

[lone abyss]

[low shard]

Huh? What's wrong with weights? If there's any issues you should report them to discord.gg/weights

Weights just uses RVC in a more simplified way

[lone abyss]

no issues but the interface is LIMITED. You can't control the index amount and the little things. sometimes a lot of things end up being buggy, and it's either doing it right when I press it or there's a queue of 1000 people

[low shard]

Because it's meant to be user friendly, you could suggest them to add more advanced settings maybe

About the queue, yeah there were some accidents before because too many people used it, I think the queue is more stable now that they added limitations bc of it

[low shard]

@lone abyss btw the only broken one between those is mainline

Let me know

[eternal aurora]

@low shard what screen shot did u want?

this is the install for pyngrok

this is where im guessing the issues come from

[eternal aurora]

If this issue can’t be fixed should I switch to mainline?

[low shard]

you went on kaggle, made a new notebook, and imported from this github link https://github.com/IAHispano/Applio/blob/main/assets/Applio_Kaggle.ipynb right?

[craggy magnet]

hello evryone can somone help me it doesnt want to launch

so iam thinking to reinstall

but doesnt want to tbh

so if there is some spcial cure to this help me thanks helpers

[low shard]

elaborate:

• what's your pc gpu
• what program are you talking about
• what you want to do
• what tutorial link are you using
• what's the error

we don't know which one of the thousands programs or tutorials you're using

[craggy magnet]

3060 ti vcoicchanger in browser it doesnt want to launch so i need help with that and no error popup

[low shard]

lemme guess, you used a youtube tutorial for a realtime voice changer?

[simple ore]

the usual "update gradio to 5.23.1"

[craggy magnet]

nope

i didnt

well i worked for few months

[low shard]

share the tutorial link you used

or a screenshot of the program folder

[craggy magnet]

but today morning stoped working

okay wait

[low shard]

!give-media-perms 1h @craggy magnet

[craggy magnet]

whats the command on here to

ask the bot for the tutorial

i used that

[low shard]

for realtime voice changer?

[craggy magnet]

yes

that one

[low shard]

did you use https://rentry.co/forkvoicechangerguide ?

[eternal aurora]

How do I do that?

[low shard]

@craggy magnet it's better you share a screenshot of the program folder so it's sure it's wokada deiteris fork

[simple ore]

use an up to date kaggle notebook

import one from Applio's github

[eternal aurora]

Okay

[craggy magnet]

yes that one

step by step

https://rentry.co/forkvoicechangerguide

[low shard]

and what's happening exactly that doesn't work?

[craggy magnet]

well i launch it

[eternal aurora]

apparently not

[low shard]

you got the nvidia (NOT 5000 serie bc u got a 3000 serie) right?

[craggy magnet]

in my file manager and it for millisecond show the window and than close

i look detail in task manager

[low shard]

then do it, the docs link has to be updated from @crude flame

[craggy magnet]

but i open and instant close

[eternal aurora]

where do i import it from

[simple ore]

open cmd.exe

and launch it from it

[eternal aurora]

okay

[simple ore]

that way you'l have an error message if there's any

[craggy magnet]

looks like worked thanks guys alot for help

[eternal aurora]

download raw file?

[simple ore]

no

[low shard]

open kaggle

make a notebook

file > import notebook

select the github import

thenn just select applio's github

[eternal aurora]

okay gimme a sec

[low shard]

did you just open cmd or also opened the program? bc cmd is just the command line prompt for windows, not the voice changer too

to open the program in cmd to check if there's any issues you should put the program name in it

[craggy magnet]

i did nothing was

and than reopen

and worked

idk my pc is acustic

[low shard]

maybe you just needed to reopen the program lol

want me to check your settings?

[craggy magnet]

i did few times before

and now randomly started working

so

i even restart pc few times

[hallow thistle]

Make sure you don't "run CMD as administrator", otherwise W-Okada would gonna run as administrator as CMD when run from it.

[low shard]

@craggy magnet so did it open your browser now?

[craggy magnet]

yes

[low shard]

be sure also your windows version and nvidia drivers are up to date

[craggy magnet]

they are

[low shard]

want to share me a screenshot so i can suggest you the best settings?

[craggy magnet]

yeah

[hallow thistle]

You can send any screenshot to here now just to make sure what you're saying is true.

[craggy magnet]

evrything opened fine

[eternal aurora]

ty

[eternal aurora]

yh just done that

[low shard]

your settings are good

the only things i can suggest is using force fp32 mode in advanced settings for more stability and better quality at the cost of some delay

or try https://rentry.co/forkvoicechangerguide#reduce-more-delay for less delay (but you won't be able to use noise suppression)

also you could set the monitor option to your headphones optionally to hear yourself

[craggy magnet]

yeah i know that one well guys thanks alot for youre time and help

[low shard]

yw

[craggy magnet]

is it big diffrence between the delay with it and without?

[low shard]

with what

[trim nest]

how do i set up mmvc with a 3060 ti

[hallow thistle]

Did you follow a tutorial video on YouTube?

[trim nest]

they are all for old mmvc

[hallow thistle]

Yeah. Are you looking for this better one? https://rentry.co/ForkVoiceChangerGuide

[trim nest]

i looked at it but i dont have that version, i use the original one not modified

i can choose chunks rather than use a slider

[hallow thistle]

Download and use this better fork W-Okada instead.

[trim nest]

is it better?

the latest from w okada is from 2025 that one is from 2024

[hallow thistle]

Fork W-Okada is better in term of performance.

[trim nest]

can u help me with this version?

[hallow thistle]

I can help you with good settings for fork W-Okada and the virtual line cable program.

[knotty moth]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

• Don't simply mention your issue, like "my rvc is not working".
• Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
• The more context, the better.

__**BE POLITE.**__ <:matsuripray:1159685390156967936>

• Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
• It's okay if you're frustrated, but don't take it into this server.
• Don't DM without prior consent.

__**BE PRODUCTIVE.**__ 🤝

• Don't ask for every little instruction. Put your own effort & test things by yourself.
• Don't ask to ask.
• Check if your answer is a Google search away/on our guides website.

If you've changed your mind about your message, feel free to delete/edit it.

[knotty moth]

asking to ask deserves to be rejected

[bleak nymph]

@analog obsidian uh so it looks flat to me

i think it means the dataset is no good? or it's too early to tell

screenshot taken at 90 epochs

[fading lodge]

i can

i got settings if u need

[bleak nymph]

i still don't really know what any of the other graphs mean i only know this one means down = good and up = bad 😭

[bleak nymph]

at 100 now and still looks the same really

[analog obsidian]

doesnt look flat to me

[bleak nymph]

i mean not totally flat

i meant rather it's not improving

[analog obsidian]

dataset size?

how many mins?

[bleak nymph]

hold on lemme check

i just put several files of my singing together and cut out the silence and some bad ones

should have my whole range (except lows since i am a bass and i sing songs by sopranos and tenors)

[bleak nymph]

how do i check again

[analog obsidian]

id recommend you to train this using the klm 4.9 pretrain

[bleak nymph]

oh

why's that

actually it might have some lows

[analog obsidian]

model_info.json inside your model log foldee

[analog obsidian]

og pretrain has low voice range

very low

[bleak nymph]

oh damn, what's the range

[analog obsidian]

speech range

[bleak nymph]

like what's the vocal range for it

[analog obsidian]

idk, i think its less than c5

[bleak nymph]

mine is an entire octave (F#2 to F#3)

[bleak nymph]

what 😭

5th octave???

[analog obsidian]

its a monotone pretrain trained using audio of people hating their job

what do u expect

they all sound ded inside

[bleak nymph]

😭 no i meant like 5th octave is like really high or i'm misunderstanding

[analog obsidian]

hmmmm

[bleak nymph]

that's like EXTREMELY high falsetto

unless i'm misunderstanding

opera style singing

[analog obsidian]

wait no
its the voice range of jpop lol

female voice range

but if you want to inference male only audio or -12 pitch, i dont think that is going to affect you much

its only bad for high pitched audio

[bleak nymph]

ah

[bleak nymph]

i mean i wanna inference high-male audio

tenors and sopranos

[analog obsidian]

uhhhh

well your dataset has those notes right?

[bleak nymph]

well

no

[analog obsidian]

🦈

[bleak nymph]

😭

[analog obsidian]

u can try one of your epochs now and see if it can handle that

it will sound undertrained but its just too see if your model can handle what you want to inference

[bleak nymph]

ok

it won't process it 😭

i think it can't inference and train at the same time?

nevermind

it comleted

it just took like 2 mins

oh my god

THIS IS KINDA

GOOD??

IT JUST NEEDS MORE TRAINING

LMFAO

IT'S LIKE BOTH THE ORIGINAL SINGER'S SINGING STYLE AND MY VOICE IT SOUNDS CURSED

doesn't sound THAT much like me

[analog obsidian]

use the index

[analog obsidian]

like 0.5 index

ofc needs more training

keep training and hear every 10 epochs and see if there any improvement in similarity with the dataset

edit: low batch sizes like 4 may also cause the voice to sound a bit different

[elder coral]

3050 of geofrce

[bleak nymph]

i used the settings u told me

i saved them so i can use it lmao

[analog obsidian]

batch size 8 is more accurate but impacts the generalization

[bleak nymph]

wym by generalization

[analog obsidian]

model being able to do stuff
bad generalization in rvc sounds like a metallic sound

but also the model will glitch sometimes

while trying to inference stuff too different from the set

batch 4 reduces those moments but sometimes the voice may sound slightly different

[bleak nymph]

oh

[analog obsidian]

but you can kinda fix this if you use the index

[bleak nymph]

i generated index

is that what u mean

[analog obsidian]

yeah

use a value of 0.5 or 0.7

[bleak nymph]

😭 i think it's a bit too late to change settings

can u change settings whilst training?

[analog obsidian]

things like batch size? no

[bleak nymph]

this is what i have it set to

[analog obsidian]

maybe train more and see if it has any improvements

[bleak nymph]

alr

[analog obsidian]

hear epoch 150 then epoch 200

[bleak nymph]

i do think it has good potential

[analog obsidian]

if 200 still sounds natural you can try training more

[bleak nymph]

thing is it's so fucking slow to inference whilst training

[analog obsidian]

yes lol

[bleak nymph]

um i think it broke it

my training stopped randomly

at 144

[analog obsidian]

uh maybe u got out of vram?

idk

[bleak nymph]

it just says this

[golden eagle]

voice model https://www.mediafire.com/file/pl5wq0iwak18y50/0426+(6)(1).MP3/file

[bleak nymph]

none of it is moving btw

idk i'l just restart

[golden eagle]

upload

[simple ore]

what about before those lines?

[golden eagle]

ok

[analog obsidian]

you didnt get any out of vram message or something?

[simple ore]

you probably just left off 'overtraining detected no improvement, stopping here' from the screenshot

l2make good log screenshots

[analog obsidian]

pls remove overtraining detector

🦈

[simple ore]

means the training is on going

speed dropping to s/it is likely vram ovefilled to shared memory because you ran inference

[golden eagle]

ALL RIGHT ALL RIGHT😡

[bleak nymph]

no i turned off overtraining detector

i restarted and it worked anyway

[golden eagle]

COME DOWN

UPLOADED

S/IT

[bleak nymph]

i'm on epoch 179 and lowest is epoch 169, nice

[golden eagle]

YES

[bleak nymph]

sir are you ok

do you require medical assistance

[analog obsidian]

dont blindly trust the graphs, actually hear the epochs

[bleak nymph]

idk i'm looking at the cmd prompt

tell me which is lowest

also

[analog obsidian]

thats just a random value of the generator

cant be used to tell which epoch is the best

[bleak nymph]

the training COMPLETELY breaks if i inference while training

[analog obsidian]

other of the many random nonsense features applio has

[bleak nymph]

oh

damn

rip

[analog obsidian]

so best metrics are ur ears

[bleak nymph]

is applio even the most accurate

voice cloner

[analog obsidian]

both applio and original rvc can do accurate reconstruction of the dataset

[bleak nymph]

i tried inferencing some other files and they sounding really unhinged

i think it's wayyyy too undertrained

[analog obsidian]

well its batch size 8

and the original pretrain is a bit slow in giving proper reconstruction

eventually it'll do it

if you want fast reconstruction then just use klm 4.9

ive heard it can do the same thing that the original one does in like 20 epochs

[bleak nymph]

daaamn

[analog obsidian]

that doesn't mean you should train your model for 20 epochs lol

[bleak nymph]

so if i use batch size 4 and use "klm 4.9"?

[analog obsidian]

it just does things faster

[bleak nymph]

yeah yeah ik what u mean 😭

[analog obsidian]

https://discord.com/channels/1159260121998827560/1339155300720054316

klm 4.9 ^

it learns your dataset faster but you'll need to train almost the same amount of epochs as if you were training using the original pretrain

at the end, they do the same thing

both should result in your model sounding like the dataset

this is all trial and error btw, so you have to actually try stuff and see which is better for your specific dataset

[bleak nymph]

ah

[low shard]

i mean it can train locally, but the issue is that it would be limited by vram.

you still want to do it locally (on ur pc hardware) or want to use cloud (remote good pc) which works for free?

[golden walrus]

guys. I want to ask if it is possible to use an embedder that only support 16khz data

For well, 48khz

[elfin nebula]

Is there a tutorial for that?

[simple ore]

read the github page

you do need hip sdk installed and stuff

[twilit forge]

im training a model with a dataset of 6mins i was wondering if there is anything specific i should be doing (batch size, epochs, etc) i did run it using batch size of 6 i feel like it still sounds like the model is breaking here is an example of it

i was thinking i should lower the batch size to 4 and it will fix the breaking but was wondering if there is anything else i should be doing while training

@analog obsidian @simple ore

[simple ore]

batch size is something you can experiment with, also leaning rate and lr decay in config.json

[nocturne patrol]

does anyone know what this is? i keep pressing reload and reload without inatalize and when it comes back to the default screen that keeps popping up when i choose my mic for the inputoh

[twilit forge]

has this started overtraining yet

[viscid moss]

nah

[viscid moss]

says you don't have perms + don't use that

[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

[nocturne patrol]

ohh

[viscid moss]

@nocturne patrol use this one, first link

[nocturne patrol]

dang it says the connection for this sight is not secure

[viscid moss]

ignore it, it's safe

[midnight fern]

Im using the okada kagglr but no one can hear it in game

Im ising virtual cable and everything

[nocturne patrol]

it doesnt let me bypass it

[viscid moss]

can u hear urself?

[midnight fern]

yes

[viscid moss]

it's just the cable then

[midnight fern]

and ive used the virtual cable before

[viscid moss]

which cable u have?

[midnight fern]

idk i had it for along time

how do i delete a virtual cable on mac

ill redownload one

[viscid moss]

idk about mac, sorry

[midnight fern]

ok

[viscid moss]

https://rentry.co/ForkVoiceChangerGuide#virtual-audio-cable

check this options for mac

[viscid moss]

send me a screenshot

[nocturne patrol]

alr

[pulsar heron]

https://rentry.co/ForkVoiceChangerGuide

for me it works

[nocturne patrol]

it doesnt for me idk why

[twilit forge]

is it overtrained now?

[pulsar heron]

i also need some help

but unfortunately i can t add files here to show you my problem :))

i ve just added a new post in ai support forum channel, hoping that someone sees it and helps me

my generated voice is always messy even if the audio i recorder for training is crystal clear

i want to create cover for songs

[nocturne patrol]

is there another link to the website

[viscid moss]

lol

[nocturne patrol]

???

how do i fix it

[viscid moss]

best way i found

[viscid moss]

seems like, at 7k

[nocturne patrol]

im already confused

but thank you