✨│ai-help

[boreal mango]

idk if theres a peter griffin skin

[low shard]

!realtime

[vernal bone]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[low shard]

you can use either wokada deiteris fork or vonovox, it's better you read both pros&cons

[boreal mango]

tysm

[boreal mango]

whats the difference between local and cloud

i replied to the clanker </3

[low shard]

local = runs on your hardware, like chrome right now
cloud = runs on a remote good pc server, it would be better than your hardware, but with limited free gpu hours

[formal wind]

Why tf you trynna import a json file

[boreal mango]

i think cloud would be better no?

[formal wind]

Oh lord

[boreal mango]

im usually listening to music and stuff on the side

oop

[low shard]

cloud depends also on your internet connection, is a bit harder to setup and you have limited time depending on the service
it would be better you try local first

[boreal mango]

okay!

tyty

[low shard]

you're welcome and let me know :D

what do you even mean a website with it? We don't allow it at all, it's written in the server rules and help guidelines and even in the docs websites

[formal wind]

Ai isn't meant for catfishing. It's meant to troll people as character.

[low shard]

AI is meant for good ethical fun, such as legal consent roleplay with friends, AI covers, etc,
not tricking and catfishing others as for example e girls

[viral mason]

https://tenor.com/view/star-wars-battle-droids-this-doesnt-look-good-not-good-binoculars-gif-22136439

[viral mason]

yet most people things that join here want to catfish

the ones that don't are actual people who should be shown kindness, not those things tho

they should get nuked

[untold igloo]

guys is there any way to convert .pth file .onnx but with like piper tts pipeline?

so like piper needs its specific input and output parameters with .onnx models, so is there any way to convert these .pth to .onnx for piper tts?

[formal wind]

Agreed

[chilly furnace]

svs cant be converted to tts

you need to retrain them in that engine

[untold igloo]

:(

i will shift to kokoro tts then, thanks for answering tho

[marsh scroll]

RTX 2060 super, win 10, trying to use a voice changer
https://www.youtube.com/watch?v=SxdnGxicJOg used this tutorial

[low shard]

that tutorial is outdated, uninstall everything
are you trying to do e girl trolling like in the video?

[serene pumice]

i want to do covers but idk how

[marsh scroll]

no just looking at any voices, thats the vid i found tho

[safe echo]

hi whats the best virtual cable please?

[celest pier]

I thought egirl trolling was against the rules

💔

[safe echo]

hi when i try run i get this error does anyone know why?

[brittle wing]

would zluda work with a cuda based TTS?
(I dont know anything about how far zluda development has gone or anything about coding at all)

[safe echo]

@odd shale thank you but this also does not work and it loads voice for 5+ minutes and doesnt work

VONOVOX just says warming up voice forever and never does

[red stratus]

Question, I don't need realtime voice conversion. I'm trying to process some recordings for the best quality.
I'm using the deiteris fork of w-okada with

3090
RMVPE
CHUNK: max (2730.7ms)
EXTRA: 5s
Advanced Settings:
Protocol: sio
Crossfade: 0.2s
FP32: on
disable jit: on

Is this optimal? It takes about ~4s for the voice conversion, but honestly it could take an hour if it means the best quality.

[simple ore]

yes

most tts can be patched to use Zluda

(not the og zluda, but Lee's fork)

[low shard]

wokada deiteris fork is meant for realtime voice conversion, not for pre-recorded audios

use applio instead https://docs.aihub.gg/rvc/local/applio/

[red stratus]

perfect, thank you!

[low shard]

It's hard to understand which program you're using

Please Elaborate:

• your PC GPU (Such as Nvidia RTX 3060, AMD RX 9070 XT..)
• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do? There isn't a program that does everything, there's a program for each thing:
  • AI Covers
  • Train RVC Models
  • E girl trolling
  • TTS
  • Roleplay in VC
  • Roleplay in Games
  • etc
• what tutorial link are you using
• a screenshot of the program

[low shard]

you're welcome, let me know :D

[safe echo]

gotcha

[low shard]

it's written we are against that in the server rules, help guidelines, docs website,
it is catfishing also by definition on wikipedia
I'm glad u understand

[safe echo]

yes that says intent for desception, which, im not doing.

but i understand the rule 🙂

[low shard]

There's no thing as Applio Pro, you probably confused with Google Colab, which is paid because you're using a remote good pc to run the software, the software itsself is open source

do you have a good pc to run it locally?

[low shard]

you can run it locally, you don't need cloud

check https://docs.aihub.gg/rvc/local/applio/

[sick ruin]

Anyone from Brazil to help me?

Could you send it to me?

[brittle wing]

thanks I been working on Dia TTS

[sick ruin]

Could someone teach me how to install by showing the screen?

[simple ore]

this is a typical method https://github.com/resemble-ai/chatterbox/issues/52#issuecomment-2925328247

but with 7000 and 9000-series amd it is no longer necessary as the are now rocm windows pythorch builds

[brittle wing]

but pytorch isn't supported on amd

right?

[simple ore]

no

https://github.com/ROCm/TheRock/

depends on your gpu

[brittle wing]

7900xtx

[simple ore]

you can try

[brittle wing]

I might be living in a CAVE this means I can run any TTS I want 😭

[simple ore]

python -m pip install --index-url https://d2awnip2yjpvqn.cloudfront.net/v2/gfx110X-dgpu/ torch torchaudio torchvision

it is still beta

there are nightly/weekly builds

[brittle wing]

whatever TTS I use it will probably check for cuda and ignore dis tho

[simple ore]

no

it works as cuda

[brittle wing]

it works as cuda or disguise itself as cuda? because it will have to perform Matrix multiplication similer to cuda

well

ignore that

I will try nonetheless

thanks alot for sharing this

[vestal gust]

hi guys what voice changer should i use for real time, i have a 1650 for my graphics card

amd ryzen 7 2700X for my cpu

[pine isle]

what does index mean here

#1409235754206494782 message

[deep blade]

yo how is it able to make

an ai voice of a youtuber

[viral mason]

u can use Applio on Kaggle I have a whole video on how to do it

it doesn't show how to clean a voice but it shows how to train one

[mystic nova]

how to fix delay

[viral mason]

have good settings

try these out

if u use vonovox I can't really help for now

[final gyro]

@low shard Hi I'm trying to download Deitieris W Okada because I want to roleplay with my friends on Roblox, the cpu I have is an amd ryzen 7 5700G.

My gpu is Radeon RX 6600XT

And the instructions say to download the 24.6.1 driver for my cpu but there are later versions than that, so the question is, do I still download the 24.6.1 driver? or can I download the latest one instead.

@odd shale

Nvm

[red stratus]

How can I ensure that applio is using my gpu for inference/tts? Seems pretty slow but maybe my expectations are too high. 40s output tts takes ~3s to create. Is that roughly what others are getting? (3090)

[sweet wren]

anyone know how to fix unsupported gloo device?

rtx 5090

[sweet wren]

nvm, just transferred to the newest update. was on a build from early february

[formal wind]

is it possible to install a custom pretrain on kaggle?

[minor isle]

how to fix peoples voice coming through okada

which they can hear back

on discord

[rain axle]

@minor isle Your microphone is too sensitive and your headphones are bleeding sound. So the mic picks it up and runs it through W-Okada. You need to decrease headphone volume or use headphones that have better sound-isolating capabilities.

[sharp rampart]

hi i wants to start the program for voicechanger start.http.bat but if i clicked on it, it just asked if i wana to start that and nothing happen after

[minor isle]

ahh thanks

what if decreasing sound also doesnt fix it?

[rain axle]

@minor isle Are you looping the incoming sound through the VAC somehow? W-Okada has to pick up the other voice from somewhere.

[minor isle]

no not really, voice only comes from the headset speakers

im not sure how its even picking it up, mic is pretty far away from the headset

[prisma rover]

guys i have a problem the voice is lagy it worked but it is lagy you know how to fix ??

[prisma rover]

ITS LIKE delayED

[brittle wing]

you think Comfyui could be used with this?

Comfyui setup offers two options manual and one via cuda

[dawn tendon]

Why does it not let me?

Does someone know?

[simple ore]

you can install it in comfy venv

[pine isle]

somebody help pretty plz #1409235754206494782 message

[viral mason]

Good settings

If u use Vonovox can't help with settings rn

[prisma rover]

ty

[viral mason]

Np!

[low shard]

huh nowhere talks about drivers, be sure to use the AI HUB Docs guide: https://docs.aihub.gg

[low shard]

ill check

[low shard]

check the command prompt

[low shard]

it's hard to understand which program and version you're using, please elaborate :

• your PC GPU (Such as Nvidia RTX 3060, AMD RX 9070 XT..)
• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do? There isn't a program that does everything, there's a program for each thing:
  • AI Covers
  • Train RVC Models
  • E girl trolling
  • TTS
  • Roleplay in VC
  • Roleplay in Games
  • etc
• what tutorial link are you using
• a screenshot of the program

[ruby latch]

If I want to clone my own voice what script should I read Is there some generic one or

[shell flicker]

where do i find the download to the main ai voice changer again?

for amd

[simple ore]

do you need it for TTS?

or do you want to make AI cover with your voice?

[viral mason]

The bee movie script

[shell flicker]

i just want the voice changer do i use dml or std?

[viral mason]

I don't want an std

[manic parrot]

hi guys, how many epochs should I use for a 52-minute audio to get a perfect result?

[potent pier]

yo anyone want to help me? I have a 4070ti super and a i7 14700kf,
i want to consistently female troll online. meaning i dont want to be caught

[viral mason]

There is no set epoch count for any model you train, just train it and listen to how it sounds until it's good

[manic parrot]

I've heard that if I use too many epochs, it can get distorted. So I wanted to know how many epochs people usually set on models that have an average of 1 hour of data

[potent pier]

yo dnot hate the player bro 😭

free bread

[viral mason]

Shhh egrill troller

[potent pier]

yo bro just help me

[viral mason]

I mean a model could be good at 200 epochs or around even 800 but over 1000 is bad

[manic parrot]

Okay thanks

[viral mason]

Btw try a batch size around maybe 12 or higher

[manic parrot]

What sample rate should I choose, 40k or 48k or 32k

[low shard]

E girl trolling is catfishing, which is illegal. You have been warned both verbally and via sapphire and continuing to ask will lead to further actions possibly like a ban

[low shard]

Catfishers aren't allowed to get any help

[viral mason]

it depends on if the audio matches one of the three, u can force the audio into becoming one of them in like audacity

[viral mason]

u can change sample rate like this ^^

best settings to export audio as well after fully cleaning btw on audacity

[potent pier]

okay nick088

[manic parrot]

oh okay thanks for video

[simple ore]

you do not need to do it for audio used for training

the preprocess takes care of resampling

[viral mason]

oh does applio and mainline change the sample rate after selection?

[simple ore]

when they load source into preprocess they resample to the selected sample rate automatically

[viral mason]

ohh

I wish I knew this kinda stuff lol

[manic parrot]

My dataset is .mp3 but it says wav or flac are lossless

If I convert the mp3 to wav will this solve the problem

[viral mason]

you should try to avoid changing the format of the audio file but it'd be better to use either wav or flac

u can change it though audacity or just some website mp3 converter

[manic parrot]

but when i make mp3 it just 10mb and when i make it wav, it will be 100mb, Won't this slow down the training?

[viral mason]

it shouldn't no

usually it's just the length of the dataset that may slow training

[manic parrot]

oh okay

[golden walrus]

sire, can i ask if KL drop to negative is a normal thing?

i asked gpt and it said it's no good

[manic parrot]

Is it faster to train the model with the A100 in Google Colab or with the rental 5090 GPU works on windows 10

[viral mason]

Kaggle would be better because it gives 30 hours instead of like 2

[manic parrot]

Even if I have compute units in Google Colab, can I only keep it open for 2 hours

[viral mason]

Idk what that means lol

[manic parrot]

So, without making a purchase on Google, you can only use the Tesla T4 GPU and for a limited time, but if you buy the Compute Unit, you can also use GPUs up to the A100

I just wanted to know which one is faster for training A100 or 5090

[viral mason]

yikes I wouldn't buy compute units for training models..

it's easier to just use kaggle completely free with the gpu t4 x2

[simple ore]

it should never do that

[golden walrus]

shiet

[analog obsidian]

check your pings 👀

[golden walrus]

so it's normal based on the explanation

[simple ore]

K/l going to negatives is not normal

theoretically it should be impossible, but rvc uses some scuffed formula

[golden walrus]

i guess so, i have no idea

[next pecan]

im using the fork voice changer and whenever im using it normally without playing any game it works okay but whenever im playing a game the voice pitches goes so high and the voice doesn't work smoothly
it used to work great in the prev versions but now it lags and i can't use it anymore while playing game any help?

[magic hare]

I want to try to improve my RVC quality, I've been using these settings on this software for a while, but it's always a bit delayed, or it's just not really good sounding. How should I improve what I use? I just want the voices I'm currently use to be more clear sounding, and also be a bit faster too. What should I do?

[viral mason]

what is this???

[magic hare]

https://github.com/w-okada/voice-changer

[viral mason]

that isn't deiteris fork or vonovox

is it some outdated or new voice changer I never heard of

[magic hare]

I don't remember.

[magic hare]

What would you recommend I use, then?

[viral mason]

depends honestly, do u have nvidia or amd

I can't help much with non nvidia users

[magic hare]

I use Nvidia

[manic parrot]

best settings for me

[viral mason]

tune?

that also looks old..

[magic hare]

Yeah that looks very different than mine.

[acoustic burrow]

hey, total noob here ... I put my Dione Launcher on English, why is my Applio in German still?

[viral mason]

tbh u could try out both

-realtime

[vernal bone]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[acoustic burrow]

nvm found it

[magic hare]

Where can I download them?

[viral mason]

each has a guide, just scroll down to the download section

[magic hare]

Oh, ty

[acoustic burrow]

Uhm, the Mainline RVC Folder if I use Applio via Dione, is the one called \apps\Applio ?

I want to use the Tensorboard Batch and in the Guide it says I should just place the bat in the "Mainline RVC" Folder, I'm not sure which folder path is meant exactly

[languid venture]

hi its normal its so laggy i have a amd gpu?

[acoustic burrow]

Do I break anything if I place and execute the Bat in the wrong folder? Tbh I'm not sure if it is \apps\Applio or \apps\Applio\applio

alright both seem to be wrong

I checked the docs, it doesn't specify what "Mainline RVC Folder" means

\DioneLauncher
\DioneLauncher\ _Install
\DioneLauncher\ _Install\apps
\DioneLauncher\ _Install\apps\Applio
\DioneLauncher\ _Install\apps\Applio\applio

None of these folders seem to be the "Mainline RVC" wtf do I do wrong

[simple ore]

Mainline RVC <> Applio

[acoustic burrow]

It gets me an error message
"Installing virtualenv...
Das System kann den angegebenen Pfad nicht finden.
Creating a virtual environment named tensorboard_venv...
Das System kann den angegebenen Pfad nicht finden.
Activating the virtual environment...
Das System kann den angegebenen Pfad nicht finden.
Installing TensorBoard into the virtual environment...
Der Befehl "pip" ist entweder falsch geschrieben oder
konnte nicht gefunden werden.
Downgrading packages for troubleshooting...
Der Befehl "pip" ist entweder falsch geschrieben oder
konnte nicht gefunden werden.
Der Befehl "pip" ist entweder falsch geschrieben oder
konnte nicht gefunden werden.
Der Befehl "pip" ist entweder falsch geschrieben oder
konnte nicht gefunden werden.
Der Befehl "pip" ist entweder falsch geschrieben oder
konnte nicht gefunden werden.
Launching TensorBoard...
Der Befehl "tensorboard" ist entweder falsch geschrieben oder
konnte nicht gefunden werden.
Keeping the command prompt open...
Drücken Sie eine beliebige Taste . . ."

Can't find the path it says and that the command "pip" is either spelled wrong or couldn't be found.

I get this same error message in every folder, including \apps\Applio

I removed the underscore in my installation folder's name, thought it may cause the error. It doesn't. Getting still the same error.

[simple ore]

do you have python installed? 3.11

[acoustic burrow]

Idk if 3.11 but my processes stated that Python was running. I'll try to install that version.

After installing Python 3.11.0 I still get the same error when executing the .bat

[simple ore]

can't you just download the compiled version of applio?

https://huggingface.co/IAHispano/Applio/resolve/main/Compiled/Windows/ApplioV3.3.0.zip

[acoustic burrow]

I tried, but when I tried to unpack it, it gave me an Archive Error, even though I use the newest WinRar Version.

I got an error where it couldn't further unpack the archive.

That's why I tried it via Dione, as the Website stated.

I mean, I can try again. I just load with about 10-11 mb/s so it takes a while

[simple ore]

probably did not download fully

[acoustic burrow]

well, I can't unpack it if the download doesn't successfully end, since I can't unpack ".opdownload" file extensions ...

[simple ore]

it obviously did not

[acoustic burrow]

Well, unpacks now again ... let's see if the download's clean now

okay, so I can uninstall DioneLauncher if I have the compiled version of Applio?

[simple ore]

you dont need Dione to run it

[acoustic burrow]

Alright, works, TensorBoard too. Thank you for your time

[deep nova]

any way to fix browser version cutting out

ive tried messing with chunk and stuff but it doesnt work well

it starts of good then goes quiet halfway through the sentence

[viral mason]

You have kinda stinky settings with chunk and extra

Also the problem could be from the model itself

Unless you have that same issues with all models

[viral mason]

Try these settings

As a start

[deep nova]

same with all my models

ill give this a try'

[viral mason]

Hmm, could try turning off sup1 and echo to see if that helps

[deep nova]

how am i supossed to get my chunks like this

wait mb

im stupid

[viral mason]

You're good lol

[deep nova]

nvm worked ty

it was just the chunks being bad

i never thought of making extra high and stuff

[viral mason]

Nice! Glad I could help ^^

[brittle wing]

worked!

[knotty moth]

which pretrain are you using?

if you're not sure, better leave the learning rate 1e-4

the important thing is to test several model checkpoints

[simple ore]

look at avg 50 charts instead

[deep nova]

sorry thanks for asking but i resolved the issue ty for your time

[brittle wing]

what is the best voice changer for AMD gpus?

[brittle wing]

did u figure it out?

[shell flicker]

yes

[carmine delta]

can anyone help me with this

after i speak it has such a big delay and when it speaks it sounds like ass

and lagged

[brittle wing]

well what is its name? am trying to figure out 😆

[shell flicker]

oh its the dml version

[brittle wing]

you mean wokada?

[shell flicker]

yes i could send the link if you want

[brittle wing]

please

my eyes hurt

I been looking around for long

[shell flicker]

wait wrong one

wait i think i installed the wrong its saying error make sure to look at supported gpus but it work weird

also whats you gpu brand?

[brittle wing]

amd 7900xtx

[shell flicker]

oh nice

ok so try the std version or dml

https://huggingface.co/wok000/vcclient000/tree/main

[brittle wing]

what's the difference ?

[shell flicker]

i remember last time i used it i thought i had used std but dml is what i use for something like this

sooo cuda is for Nvidia and dml or std is for amd

[shell flicker]

ill look into it a little later though

[brittle wing]

no need, it works for you now that's what's important

also.ur name color changed

[viral mason]

good settings

[shell flicker]

it said i leveled up

https://huggingface.co/wok000/vcclient000/tree/main

this is the link

[brittle wing]

[shell flicker]

also heres the link to the main github

https://github.com/w-okada/voice-changer/blob/master/README_en.md

[brittle wing]

yup almost finished downloading

also...

[viral mason]

-realtime

[vernal bone]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[viral mason]

u should only be downloading the first two

not from some random link

[brittle wing]

do NOT search in Google imgs when typing wokada

[shell flicker]

https://github.com/w-okada/voice-changer/blob/master/tutorials/tutorial_rvc_en_latest.md

[brittle wing]

big mistake

[shell flicker]

ok.......

[viral mason]

[brittle wing]

it says it's only for Nvidia no?

no mention of rocm, zluda, direcrtml

not even a crumb for amd

[shell flicker]

he's right

[brittle wing]

[shell flicker]

https://github.com/w-okada/voice-changer/tree/master

just go to this

[viral mason]

I believe it has the amd version in the guide but I could be wrong, only vonovox is nvidia only

[shell flicker]

it does

[viral mason]

my bad lol

[near palm]

is this normal?

[shell flicker]

https://www.youtube.com/watch?v=_JXbvSTGPoo&t=331s

[simple ore]

not the avg_50 charts again

all of them

dont search

just expand them and collapse others

and change the smoothing to 0.5-0.6

click on each chart

collapse regular loss, expand loss_avg_50

same for grads

start checking models from 12k+ steps and higher

not really what you think it is

it is really weird metric

lowest value is collected for individual steps which are not really at the save point or at the end of epoch or at 50th step

so they may dip really low while the average is not so much

it is loss_avg_50/g/total

[knotty moth]

no, the og version is terrible for amd gpus @brittle wing

[shell flicker]

oh ok

whats the new one?

[knotty moth]

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

[near palm]

it's kind of just loading infinitely, it went up to 1500 before i gave up
it's never done this before

[cold patrol]

what are the best models atm

[shell flicker]

why is it on a web client and not on a app

[knotty moth]

read this section

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

[shell flicker]

ohhhhhhhhhh ok im just dumb

[merry eagle]

-web

-rvc

[vernal bone]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[simple ore]

what Fork?

Applio is RVC fork

Kaggle is a online ML service provider

where you can run Applio

what's your GPU?

well, there's original mainline RVC which has not been updated for ages

there's Applio

https://huggingface.co/IAHispano/Applio/resolve/main/Compiled/Windows/ApplioV3.3.0.zip

Applio can be set up to mimic RVC mainline (mostly) - FP16, prepeocess using automatic slicing with noise filter and post normalization

Kaggle is online cloud provider

it is not local

[manic parrot]

You can run applio on your local pc / google colab / kaggle. all the same things but working in different places

You have to choose one of the three. If you have money, google colab is best because A100 Gpu is fast or you can buy 5090 cloud gpu. Otherwise you can use your own GPU or Kaggle (2x Tesla T4 Gpu) So if you want speed you can buy google colab but 3080 or T4 can also handle this job.

[simple ore]

3080 is perfectly serviceable for model training

yes, you can run applio on your computer

you misunderstood.. if you ask for how much epochs you should train the answer is 'nobody can tell you, you need to look at tensorboard charts, then test the saved epochs'

[manic parrot]

here is the local installation guide https://docs.aihub.gg/rvc/local/applio/

[simple ore]

lets say you have 1 hour file and you're using a default pretrain. set max epoch to 300, then run the training and run tensorboard in another window to check the progress

[native kernel]

Traceback (most recent call last):
File "core\inference\cuvc.py", line 407, in core.inference.cuvc.RVC.init
File "core\inference\cuvc.py", line 404, in core.inference.cuvc.RVC.init.set_synthesizer
File "core\inference\cuvc.py", line 263, in core.inference.cuvc.RVC.init.set_default_model
File "core\inference\cuvc.py", line 392, in core.inference.cuvc.RVC.init.get_synthesizer
File "F:\Vonovox-1.6.6\runtime\Lib\site-packages\torch\nn\modules\module.py", line 2624, in load_state_dict
raise RuntimeError(
RuntimeError: Error(s) in loading state_dict for SynthesizerTrnMs768NSFsid:
size mismatch for enc_p.emb_phone.weight: copying a param with shape torch.Size([192, 256]) from checkpoint, the shape in current model is torch.Size([192, 768]).

Error: RVC model not properly initialized. Check above error for more details.
Failed to initialize audio processing
Critical error in start_vc: Failed to initialize audio processing
Traceback (most recent call last):
File "gui\gui.py", line 1493, in gui.gui.GUI.start_vc
File "gui\gui.py", line 1530, in gui.gui.GUI.initialize_voice_conversion
RuntimeError: Failed to initialize audio processing

Sometimes I get this error when trying to use some voice models on Vonovox

is it because Vonovox was made for RVCv2 and this model could possibly be RVCv1?

Metadata says "type": "v1",
"tags": [
"Anime Character",
"Fictional Character",
"RVC v2",
"Japanese"
],

Type is v1 but it's tagged as v2 lol

[simple ore]

likely

size mismatch for enc_p.emb_phone.weight: copying a param with shape torch.Size([192, 256]) from checkpoint, the shape in current model is torch.Size([192, 768]).

[native kernel]

yeah im guessing that's the problem im new to this ai voice model stuff

[simple ore]

768 is v2, 256 is v1

[native kernel]

yep then that's the problem

[simple ore]

your file is v1 model

[native kernel]

guess I have to use the W-okada program?

[simple ore]

yeah

[native kernel]

is v1 a lot worse than v2?

[native kernel]

thank you for the help!

[simple ore]

i would not say much worse

less precise phonemes

[native kernel]

Oh okay I see..

I'll stick with Vonovox for now and if I really want to try a v1 model i'll use W-Okada

thank you for the helpp

[manic parrot]

türk müydün la

evet

zaten ayırdıysan parçalara tüm sesi

bi daha ayırmadan dırekt yuklıcen dataset kısmına

[manic parrot]

tek tek derken

seçemıyon mu hepsini ctrl a ile

[native kernel]

Should I add silence/gaps in between audio clips for training?

So far it's just a blast of voicelines one after another

[near palm]

my okada won't work, i'm stuck permanently loading for hours
is there a way i can fix this?

[manic parrot]

burdan skip de

zaten kesiliyse

kesmez dıyo

bide senın dosya seçmene gerek yokki

zaten yüklemiceksın

sesleri

folderın

yolunu

yazcan oraya

yok

tüm sesleri assets/datasets klasörune atıp

dataseth path e assets/datasets yazcan

tensorboard tutoriali var

sitede yine

https://docs.aihub.gg/rvc/resources/training/#tensorboard

nerde durman gerektiği yazıo

Normalization ne bılmıyom da

none yap

sitede olmayan ayarlar önemsız yazıyodu

8

batch size mı bu

turkce kullanıyon anlayamadım o yuzden

tamam 8 o zaman

evet 1000 yapcan

her 10 epoch ta kaydetsin

eğer kestiysen

o silent şeyini

0 yap

konuşma modelı mı yapcan

tamam

o zaman

spin seç

diğerini de rmvpe veya fcpe

fcpe seçersen daha hızlı train olur

rmvpe ile çok fark yok aralarında

yok

Spin

contentvec in

daha gelişmiş haliymiş

content vec eskiymiş

ondan dedım

burda bişey seçmene gerek yok

kendın pretrained model eklemıceksen yanı

o detaya gırıyo

zaten

önce index

sonra eğitim

@rotund vigil 1 epochu kaç sanıyede uretıyo senın gpu da

assets/datasets klasorunu ss atsana

lazım değil aslında

ben dc açık olmadan yapıyorum

doğru

bide dataset path kısmını at ss

slash ı yanlış gibi

ben assets/datasets

yazınca

olmuştu

bundan olmayadabılır ama

bide slashı duzeltıp dene

valla herşey doğru gorunuyo

wav zarar görmüş olabilir onu yaparken

daha deneyımlı bırısıne sor ben ayarları bılıyorum sadece

[deep furnace]

hi can someone help me get a voice changer i really don't understand how to download that thing

I saw one here but i really don't understand anythinghttps://discord.com/channels/1159260121998827560/1393389200862089240

[simple ore]

you likely did not cut those pieces to acceptable size anyway

[simple ore]

this is an english channel, speak english or go to DMs

point preprocess to the folder with your files, run preprocess, see the output window

[manic parrot]

Is there a problem here? Should we write the path of the audio file directly or the folder

[polar tendon]

RVC reprocesses the audio files when pre-processing, not sure when skipping tho.

[simple ore]

remove the file name and /

you need to provide a path to the files

[manic parrot]

so this is true right?

[simple ore]

yes

user built-in slicer

[deep furnace]

To respect the guidelines (sorry I didn’t)
I have an nvidia geforce rtx 3050 and I’m on windows 11
Everything is just very overwhelming and I don’t understand anything from the already existing guide

[simple ore]

Automatic if you have a lot of silences or Simple if you've trimmed silences in Audacity

[viral mason]

do you know which one you want to download? either being Deiteris or Vonovox?

they're super simple to set up

[deep furnace]

I didn’t even understand there was 2 bro

What’s the difference between the 2 ?

[viral mason]

from what I understand Vonovox is just better in general as it has a lot more options but the only flaw for now if it has very few model slots, and Deiteris is still good and has a lot of model slots like over 100 if you really need that many. Vonovox has better quality overall but less slots for now and Deiteris is slightly lower quality but not really noticable

I still use Deiteris tho because I need all them model slots

[deep furnace]

What’s a model shot ?

[viral mason]

that's where u put the voice model u wanna sound like

[deep furnace]

Oh ok so if I have 10 of them I can switch between 10 different voices ?

[viral mason]

yup!

[deep furnace]

Yeah ok I see then

https://docs.aihub.gg/realtime-voice-changer/local/vonovox i go there then right ?

[viral mason]

u can use either guide from here

-realtime

[vernal bone]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[viral mason]

the second one is vonovox

[deep furnace]

Ok yes so that’s the one I’m on

I need the VAC and anything else ?

[viral mason]

just that and the voice changer

[simple ore]

you have a problem

you're using a standard pretrain, but your dataset was extracted with SPIN embedder

you not gonna get a good result

you either need to use contentvec (delete extracted folder from logs/yourmodel and re-run feature extraction) or you need to use spin pretrain

use custom pretrained under advanced options at the bottom

download it from #1235952130855010365

[deep furnace]

Hmmm

I have the .zip vac

But then it doesn’t seem to work

[simple ore]

what kind of question this is

[deep furnace]

@viral mason what does it change if I say I’m an advanced user or not ?

[viral mason]

idk what this means

[deep furnace]

"Additional options" what kind of additional options

[manic parrot]

Is there a big quality difference between Spin and Contentvec?

[viral mason]

probably don't click that bc idk what it does

[deep furnace]

ok i didn't

it's installing the vac

[viral mason]

nice

[simple ore]

download spin pretrain from https://discord.com/channels/1159260121998827560/1408339243948052501

[viral mason]

I recommend restarting your pc now just in case

[simple ore]

personally I have not noticed much difference for english

for other languages it may matter

anyway, download the pretrain files, put them into \rvc\models\pretraineds\custom

both

you need both G and D weights

[deep furnace]

it's downloading 3.5gb on my pc is that normal ?

after clicking on setup.bat

[viral mason]

yes any ai voice changer is very large because it has a lot of stuff in it

[deep furnace]

i don't think it's working

i clicked on setup.bat, downloaded a bunch of things

then i clicked on start.bat, i have to press a key but it doesn't do anything after

[viral mason]

the start didn't do anything?

odd

[deep furnace]

i can stream i maybe ?

[viral mason]

do u have image perms?

[deep furnace]

no

[viral mason]

hmm

[deep furnace]

i have image perms yes

but i can't stream apparently

[viral mason]

so u can send a video of what it's doing maybe?

[viral mason]

u might need to level up

[simple ore]

logs folder has your model name

inside there's extracted folder

delete that and re-extract

you ran out of space on your hard drive anyway

applio is 7-8GB, each model save is 1.2GB

you can use a different drive

other than C:

[brittle wing]

I downloaded the one you sent by deiteris and works perfectly thanks

this is probably the most helpful discord has ever been to me

[short hazel]

Hi

What's the best setting for my voicie changer so I can do voice message

I want to quality can be good as others because when it comes to my laugh it just went sounds weird

[viral mason]

which voice changer do u use?

there are multiple

[short hazel]

Uh

I think is the

Honestly idk it name

[viral mason]

I'd say send a screenshot but idk if u can

[short hazel]

Yeah but

This channel deosn't allow ss

[short hazel]

Id send trhough dms

[viral mason]

u could do that yea

[simple ore]

if you already dropped them into the folder I said you should, then you just need to select D/G files

or if there's only one set it will be picked automatically like on the screenshot

[analog obsidian]

batch size 64 was used to train a pretrain of almost 60 hours

if ur using the spin version of lcore you also need to extract ur dataset features with spin

show the extract feature step, be sure you selected spin

[merry eagle]

-rvc

[vernal bone]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[merry eagle]

<@&1159293204038955078> i am having this issue can someone please give me install.bat file cuz i don't have that one idk why

-locally

[simple ore]

dont run it as admin

[merry eagle]

thanks buddy at first it didn't work but now its working

[merry eagle]

can you tell me where i paste the model?

[honest river]

Hi guys!

I need help in running RVC, which software would be best?

[viral mason]

what are you trying to do? make a model of a voice or use a voice changer to change your voice into a different character?

[honest river]

use a voice changer to change your voice into a different character.

[viral mason]

mk so do you use nvidia or amd?

[honest river]

nvidia

But my friend also wants to do it, he doesn't have nvidia or amd

He has intel HD graphics only

[viral mason]

cool u got two options for a voice changer, Vonovox and Wokoda Deiteris. Idk what to do for your friend as I don't have intel but I'm pretty sure there's an intel version of Deiteris

-realtime

[vernal bone]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[viral mason]

just click the guid on either and read up on them

[honest river]

Cool!! Thank you so much!

[viral mason]

I explained the difference between the two here in case u wannt know

[honest river]

What do you use? or do you make models?

[honest river]

Yes I do.

I am reading that rn

[viral mason]

I use Deiteris atm but if u don't need over 8 models at a time u should use Vonovox since it's better quality overall

[honest river]

just 2 models

[viral mason]

Vonovox should be fine then, make sure to download vac lite (it's in the guide) and the voice changer and u should be good

[honest river]

I can directly download that and run the rvc in realtime?

[viral mason]

yup btw rvc doesn't stand for realtime voice changer lol

[honest river]

the vonovox is the voice changer right? Or do I need another tool for that? or does Vonovox uses rvc to change the voice?

Sorry kinda new and confused

[viral mason]

you're good

[honest river]

Got it 😂

[viral mason]

vonovox is the voice changer :D

[honest river]

Amazing! Thanks

[viral mason]

Np!

[honest river]

And for my friend he has to use Deiteris as he uses basic intel graphic card?

[knotty moth]

please elaborate:

• your GPU & PC spec and the operating system used
• what purpose do you want to do for?
• what kind of voice models do you want to use?
• what tutorial link are you using/referring to?
• show the screenshot of the program you're using, esp its version

[viral mason]

I think Desuwa can help more with that

[knotty moth]

it could only run in cpu mode, whose performance will be much worse than average gpus like RTX 2060

[tidal swift]

i wanna use realtime RVC but my gpu is Intel UHD graphics and RTX 2050 is that enough or not? i want to use it smoothly while using discord/games, but i don't know if this is enough or not?

can somebody help?

[honest river]

1. Nvidia GPU, windows 10 pro, 8gb ram
2. Just to try how it out, me and my friend want to change our voices
3. As of now, none
4. Haven't checked it out

[honest river]

But my friend has a really low powered laptop. Like windows 10, with basic intel graphics card and 4 gb ram

[knotty moth]

please elaborate:

• your GPU & PC spec and the operating system used (your RTX 2050 is usable but may be only for discord and casual uses)
• what purpose do you want to do for?
• what kind of voice models do you want to use?
• what tutorial link are you using/referring to?
• show the screenshot of the program you're using, esp its version

[knotty moth]

I dont think even 4 gb ram could keep up with such modern workloads, including the browser you'd need to run the cloud services

[oak edge]

applio

[oak edge]

gang can I ask some doubts in new applio interface

[honest river]

What about in my case?

[honest river]

I searched that, didn't get how to run in colab

[oak edge]

it's pretty easy gang, you want to see?

[honest river]

Sure

[knotty moth]

no, applio is NOT a voice changer program you'd need

[oak edge]

i mean they can train stuff and do right?

[tidal swift]

1. i used Nvidia RTX 2050, on windows 11 with 8gbs of RAM
2. i wanted to have fun with my friends using silly voice while also using it for roleplaying with friends too
3. i'm still not sure, but probably more leaned into animes, rpg, english voices?
4. https://www.youtube.com/watch?v=0uyGj7Fel58, i've used it for the colab one but it didn't work cause too many errors on my part but idk why so i just figured to try the local one. but not sure if my device can handle it or not

[honest river]

then what can i use in my laptop?
and if i get my friend to update from 4gb to 8gb ram?

[knotty moth]

I think you should stop invading @honest river's topic and go back to your usual hangout in https://discord.com/channels/1159260121998827560/1159265300915687534

[oak edge]

damn aighttt

[knotty moth]

a decade old laptop should be upgradable right?

8 GB is bare minimum but may be restrictive, 16 GB would be ideal

[viral mason]

training in Kaggle is 100X better because it gives you more time

for free

[oak edge]

what's kaggle?

[oak edge]

temme about it now I'm curious

[viral mason]

the better alternative to google collab my friend

basically I have a whole video that shows u how to use it

it's slightly outdated but should work

kaggle.com/code/deiant/applio

https://dashboard.ngrok.com

both links u need to train on Kaggle applio

quality is doo doo because discord

[oak edge]

WAIT WILL IT WORK FOR APPLIO

[oak edge]

ohh

how different is it compared to regular applio

[viral mason]

only difference is just that batch size is multiplied by 2

so if u set the batch to 3 you'd get 6

only a kaggle thing

it's weird but managable

[oak edge]

i mean

can you tell me how to change

vocoder as well

is it possible

[viral mason]

wdym?

[oak edge]

it usually trains with HiFi Gan right

[viral mason]

ye

[oak edge]

but there are other versions

and hifi gan is kinda the old one

at least according to applio's page

i want to try and use the refine gan if possible

[viral mason]

if u wanted to change it I'd ask Noobies or Lyery maybe

idk how

[knotty moth]

to use only a single gpu, the "GPU number" can be set to "0" instead of "0-1"

[oak edge]

alright alsooo there is new thing

like spin

next to contentvec

[viral mason]

would that be done in the training tab?

[viral mason]

yea applio has spin embedder

[oak edge]

what's that

[knotty moth]

it is in the preprocess section

[oak edge]

is it better than contentvec

I'm training tamil songs for context

also i'm noticing

there are like small "tch" sounds in every converstion

like it's observable when listened through headset

like some spikes every few seconds

in song

[viral mason]

?

[viral mason]

could u send a video or smth to show what u mean?

[oak edge]

OHHH YES what's normalization

[oak edge]

yea sure I can send a voice sample

[knotty moth]

not sure but otherwise it may be before preprocess (besides the model name & sample rate setting)

[oak edge]

also is kaggle here has faster training than colab? (I'm using 2 accounts in colab currently and training this 50 minutes file for like 8 days)

i mean I've been trainin everyday though

[viral mason]

normalization is really good as it helps models not having issues with volume such as a model being way too loud or quiet

[tidal swift]

i wanted to put the screenshots of the error but i can't put image here?

[viral mason]

it shouldn't take more than a day 😭

[oak edge]

i'm a perfectionist

[knotty moth]

!give-media-perms 10m @tidal swift

[oak edge]

i'm at 440ish epochs now i think

[viral mason]

Kaggle trains quick so at most an hour long dataset would take like maybe a day or 2

[oak edge]

and yea it doesn't sound good enough

[oak edge]

THAT'S LOVELY mwah

🔥

[viral mason]

[oak edge]

no homo btw gang

[ruby latch]

I will do that

[viral mason]

I am homo

[tidal swift]

here's the error output from the colab one

[oak edge]

ahh no offense then

[viral mason]

lol ur good

[oak edge]

ganggg

[viral mason]

anyways it shouldn't take long for u to train the model on there

[oak edge]

i will love to try it tomorrow morning as the first thing, will the model sound better in terms of quality

also do you know about singer pretrain?

[viral mason]

singer pretrain?

[oak edge]

I'm having really hard time with my voice and pronounciation not getting applied to songs

[oak edge]

yeaa

this is singer pretrain

[viral mason]

interesting

[oak edge]

🥲 but the thing is I sing in tamil

and tamil is dravidian south asian language

and I'm not finding any pretrains suitable for it

(yup default one really sucks hard with pronounciation)

[tidal swift]

this one is the fullscreen one if this helps

[oak edge]

i've tried titan 40k

[viral mason]

I mean it is still Applio and does depend on quality of the dataset

[oak edge]

and yea not that much of an improvement

[analog obsidian]

the pretrain language doesnt matter, is the embedder that needs to be finetuned

[viral mason]

titan is undertrained I think, causes harmonic glitches and stuff too

[oak edge]

i recorded myself for 60 minutes my throat all dried up

[knotty moth]

afaik voice changer colab no longer works

[oak edge]

ohh temme

[oak edge]

is the new spin one good?

that's in applio

[tidal swift]

is there any alternatives for colab voice changer?

[oak edge]

my stupid gpt said it's better than default

yea it even said to try out snowie i think

[knotty moth]

run locally with your rtx 2050, though only viable for discord & casual gaming

[oak edge]

since I speak in tamil

[analog obsidian]

its better for english but for other languages idk

[tidal swift]

is there a link for a guide or document to do that?

[oak edge]

ganggggg

[knotty moth]

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

[manic parrot]

why pth file is always 52,7 MB, is there a problem with it

[knotty moth]

why the sun rises in the east?
it is mostly so, not less nor more

[manic parrot]

I don't know how the size of a 20 epoch and a 300 epoch file can be the same size

[viral mason]

it's normal

[viral mason]

I believe that's for russian voices

[manic parrot]

which software, deiteris w-okada fork or Vonovox is good for realtime changing

[viral mason]

do you need a bunch of diff models at one time?

if not download vonovox (it's nvidia only tho in case u don't have that)

[manic parrot]

Can't I switch between models in vonovox?

[viral mason]

u can I just mean it has wayyyyyyyy less slots

like 8

and Deiteris has over 100

[manic parrot]

okay its not problem for me

so vonovox is better than deiteris right

[viral mason]

yea

[manic parrot]

I have topic in #1192011222023950368

[viral mason]

everytime I see that copy paste I think of Nick

[trim onyx]

hey is anyone here familiar with wan ai or comfyui?

[latent kettle]

I'm familiar with comfyui. And I also know what Wan is but I never used it because of hardware limitations

[trim onyx]

can i use a rx 9070xt

if so hep pols

[latent kettle]

https://github.com/patientx/ComfyUI-Zluda

[manic parrot]

Is this point overtraining

[latent kettle]

See loss_avg_50

[latent kettle]

You can wait if it never came back again to the lowest point it's overfitting

[manic parrot]

So should I wait for it to drop further

300

2

240 epoch

yanı

[trim onyx]

im having issues

im not sure whatto do next

[simple ore]

latest

[viral mason]

@warm holly here's the guide if u wanna make an ai voice

[short torrent]

where would i find the newest version of deiteris w okada

[viral mason]

-realtime

[vernal bone]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[viral mason]

first guide

[short torrent]

thank youuuu

[jagged coral]

which vc version is best to use on 5070ti? ty

[vernal bone]

❓ Help Guidelines

If you need urgent help, please checkout our AI Hub Docs or ask for help here following the [Guidelines](#1402790586028789830 message)

[low shard]

Please Elaborate:

• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do? There isn't a program that does everything, there's a program for each thing:
  • AI Covers
  • Train RVC Models
  • E girl trolling
  • TTS
  • Roleplay in VC
  • Roleplay in Games
  • etc
• what tutorial link are you using
• a screenshot of the program

[reef bane]

hello i tried to train a female voice model. first i used 1 hour dataset and finished training at 100epoch i use that model at real time rvc i get male kid voice. then i tried training with 2 Hours dataset but i got same results again. how can i get feminine voice tone? thank you

[gritty vault]

can someone send me th E-women Voice changer, I cant Find it for some reason

[low shard]

Are you trying to do e girl trolling lol?

[gritty vault]

Close enough.. I dont feel like explaining

[low shard]

E girl trolling is catfishing, which is illegal. You have been warned both verbally and via sapphire and continuing to ask will lead to further actions possibly like a ban

[gritty vault]

SO i should have explained dawg.

[low shard]

E girl trolling is catfishing, which is illegal. You have been warned both verbally and via sapphire and continuing to ask will lead to further actions possibly like a ban

[gritty vault]

im not trolling anything. Im making a song and need a girl voice in the backround

god

admins be trippin

Jr admin for a reason

[viral mason]

how dare you call it that

dissapointed

[gritty vault]

The model i need is quite litterally called "E-women"

[viral mason]

you called the voice changer "th E-women Voice changer"

https://cdn.discordapp.com/attachments/1369012167621214240/1409928788078231583/Flocci_stare.gif

[low shard]

From what you said it was meant for realtime voice changing, which doesn't do that, but is for realtime used in programs such as discord or games, and you did say it was related to e girl trolling

We don't want anything related to e girl trolling / catfishing here for moral and legal reasons, since there have been too many cases of people trying to extort or do other bad things with it, as seen in outdated YouTube tutorials which "normalize it"
I hope you understand what I'm talking about

[low shard]

Btw be aware that the version and program depends on your PC GPU and Operating System, if you're on Nvidia Windows, try Vonovox, else use the wokada deiteris fork version for your specific OS & GPU :)

The current latest is 3.3.0 as of now (just checked), be sure to be on windows 10/11 and to have an Nvidia GPU for that precompileds
If you have an AMD GPU, follow the process for modifying that precompiled in either https://docs.aihub.gg/

[reef bane]

can anybody help?

i tried different pretrained model and got same result again

[simple ore]

change the pitch

[low shard]

^ be sure to also use the tensorboard while training, to monitor how it's going (instead of always training the same amount of epochs, as there isn't a right amount)

[rare grotto]

nah forget i asked anything, i'll just go back to figuring this stuff out on my own

[glass lily]

how do i use codename fork w mmvc?

never heard of it but a model i wanna use requires it and i cant find a tut anywhere lol

[simple ore]

what model?

[formal wind]

Can yall give me the best settings for these settings in UVR5 no ui colab.
The reason i'm asking is because sometimes it sounds robotic and noisy at curtain parts with a lot going on. Its not an issue with the hugging face uvr. Just colab

[short torrent]

oh, I am using nvidia & windows where would I find the installation for Vonovox?

[oak edge]

Yooooo

i have some doubts regarding the kaggle

applio*

can I upload data set from google drive link?

[prisma rover]

yooo my voice changer only use my cpu but i want to use my gpu and i didn't find my gpu in the options i only find my cpu any one knows how to fix it ??

[gentle cipher]

Hi I have a question regarding yk the voice changer we all use. So like idk why when I turn on my voice changer it works good but when I play a YouTube video or when I talk to someone, it also changes the voice talking in the youtube video and with my friend. I only want my voice to be changed not them. Can anyone help me?

[low shard]

Reading the 2nd link guide of that message

[low shard]

"my voice changer" is too vague, there's too many programs that can be called like that

Please Elaborate:

• your PC GPU (Such as Nvidia RTX 3060, AMD RX 9070 XT..)
• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do? There isn't a program that does everything, there's a program for each thing:
  • AI Covers
  • Train RVC Models
  • E girl trolling
  • TTS
  • Roleplay in VC
  • Roleplay in Games
  • etc
• what tutorial link are you using
• a screenshot of the program

[low shard]

I will reply in https://discord.com/channels/1159260121998827560/1410221779900698744

[manic parrot]

vonovox setup.bat is not working

[low shard]

Don't run it as admin

[manic parrot]

I didn't start it as admin anyway.

[simple ore]

the install bat needs a change

@shy spruce curl calls need --ssl-no-revoke

it can be antivirus that is messing with it, but the end result is curl fails to download the files and thus the errors above

[manic parrot]

are you sure Vonovox is safe

It doesn't have many stars on GitHub. This creates a trust issue

For some reason, Vonovox seems to me to be something that is only promoted to people by this server. Because the number of people using this software is very small and it is unknown to anyone

[patent locust]

Can GTX1060 6GB be used for AI video generation?

[simple ore]

can you race Nascar in Ford Focus? Sure you can.. you're just gonna finish 5 hours later than others

[simple ore]

in reality no

it has no fp16 or quantized data type support

AI Video Gen requires 16GB VRAM + >32GB RAM to have a good time

[manic parrot]

vonovox does not work, nor does deiteris saying don't use w-okada either, what should we use then

I haven't seen a single suggestion that works without any issues

[simple ore]

is it on colab/kaggle?

[manic parrot]

Its colab

[simple ore]

did you actually run install part?

[manic parrot]

yeah

[simple ore]

show it

[manic parrot]

its ticked

[simple ore]

the log

or better the URL to this colab

[manic parrot]

https://colab.research.google.com/github/deiteris/voice-changer/blob/master-custom/Colab_RealtimeVoiceChanger.ipynb

[manic parrot]

What should I understand from here

[simple ore]

it failed to download the prebuilt voice changer

so it failed to unzip it

[manic parrot]

So what's my problem here

My colab is working properly

[simple ore]

the colab is failing to download 1.2GB file

[pallid monolith]

Error: "NoneType' object has no attribute 'host_api' loop stop Error: 'NoneType' object has no attribute 'host_api

vcclient is terminating...

vcclient is terminated. [18000]

native client is terminated.

How to fix

[manic parrot]

why there is a colab version of this then

[oak edge]

help I'm getting this in kaggle

applio

@knotty moth

[knotty moth]

~~@low shard I'd leave you or anyone else to deal with this guy ~~

[oak edge]

[stiff basin]

when i click on the http_start.bat i see a flash of the cmd but it doesnt open anything, can someone help?

[low shard]

start_http.bat is apart of the original wokada, dont use it

Please Elaborate:

• your PC GPU (Such as Nvidia RTX 3060, AMD RX 9070 XT..)
• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do? There isn't a program that does everything, there's a program for each thing:
  • AI Covers
  • Train RVC Models
  • E girl trolling
  • TTS
  • Roleplay in VC
  • Roleplay in Games
  • etc
• what tutorial link are you using
• a screenshot of the program

[low shard]

what's ur pc gpu and OS first? cloud shouldn't be ur first option

[oak edge]

it's 4060

[stiff basin]

my gpu is 4070 ti super windows 11

https://www.youtube.com/watch?v=SxdnGxicJOg

thats the tutorial i used

[oak edge]

but yea I don't want to use my gpu as it have caused issues before when running RVC and UVR

[stiff basin]

trolling
and i dont got screenshot cuz it doesnt open

[low shard]

that tutorial is extremely outdated, delete everything

are u trying to do e girl trolling like in the video?

[stiff basin]

yea

how should i download the voice changer the correct way?

[low shard]

E girl trolling is catfishing, which is illegal. You have been warned both verbally and via sapphire and continuing to ask will lead to further actions possibly like a ban

The video you used is extremely outdated, u can just uninstall everything and forget to catfish people

[low shard]

u sure? the rtx 4060 should work fine with them

[oak edge]

my data set my voice is 2.29 GB

and around 3 hours of voice data

so I don't think the vram and ram is enough

it will fail real bad at extracting if it survives preprocessing

(cuz the max I could do was 20 minutes locally and that almost BSOD my pc like twice)

[low shard]

btw u don't need a too much long dataset, quality is also important

u could try locally with applio https://docs.aihub.gg/rvc/local/applio/

If you dont want to try, are u sure ur using the same proecess shown in https://docs.aihub.gg/rvc/cloud/applio-kaggle/ ?

[oak edge]

i've followed the docs hub way and as well as a video sent by @viral mason

[sly storm]

why does my voice changer start_https.bat close right upon opening after i created a shorcut

[oak edge]

the direct documents in aihub is throwing this error

[low shard]

start_http.bat is apart of the original wokada, dont use it

Please Elaborate:

• your PC GPU (Such as Nvidia RTX 3060, AMD RX 9070 XT..)
• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do? There isn't a program that does everything, there's a program for each thing:
  • AI Covers
  • Train RVC Models
  • E girl trolling
  • TTS
  • Roleplay in VC
  • Roleplay in Games
  • etc
• what tutorial link are you using
• a screenshot of the program

[low shard]

from the error, it looks like you didn't setup the correct ngrok authtoken, don't miss on https://docs.aihub.gg/rvc/cloud/applio-kaggle/#ngrok-setup

[oak edge]

I'm at the initial step, not even going to the running notebook part

i created the kaggle account verified with face

and then I clicked on import notebook

then pasted the link given in support page

into import notebook

and that's where it throws this error

[simple ore]

without creating a new notebook?

just 'create', then 'import notebook'?

and what link did you paste?

[oak edge]

wait a second, I created a notebook

[oak edge]

https://github.com/IAHispano/Applio/blob/main/assets/Applio_Kaggle.ipynb

this link

[simple ore]

i'm making an update, hold on

[oak edge]

i'm at this page rn

[simple ore]

okay, you import that kaggle url

[oak edge]

awwwwwwwww shit

[simple ore]

then you do need to provide ngrok token

[oak edge]

my bad

[oak edge]

yup yup found the issue

let me try again

[simple ore]

from new notebook, that's correct

you go file, import notebook, then provide the url

but we are about to make a new build

remove --branch 3.3.0 from install cell

[oak edge]

ohh

[low shard]

i'm guessing you're talking about RVC or Wokada (hopefully not from youtube video tuts, they are all old), you can just lower the trained model accent by lowering the index

If you want, you can send a screenshot of ur entire program so i can help you up with settings :D

[stiff basin]

i'm sorry and i understand the warning and i'll make sure not to break any rules, i dont wanna ask for anything but i just want to clarify that i wasnt planning on catfishing anyone. the trolling i mentioned was intended for joking around with some good friends i've known for years and not strangers. on top of that i also make music and wanted to try ai voice tools for vocals, all plugins or softwares where either bad or paid. il'l avoid anything that could be seen as catfishing, i just wanted to be upfront about my actual intentions

[low shard]

!give-media-perms 1h @round dew

here, just gave you temporary perms

you're using an extremely outdated version of original wokada, over a year old damn
dont ever use video tutorials for realtime voice changers, your settings are bad too

delete everything,

i'm guessing ur on windows 11 and you seem to have an rtx 4080
i need to know which purpose you need to give you updated programs, since there are different ones

I see you're using an e girl model, are you trying to do e girl trolling?

E girl trolling is catfishing, which is illegal. You have been warned both verbally and via sapphire and continuing to ask will lead to further actions possibly like a ban

You can just delete everything and forget about e girl trolling people

We don't help with e girl trolling, changing the topic with "seeing how good it is", won't make us help you

that doesn't mean they are meant for e girl trolling /catfishing, it's written right in the server rules and help guidelines, the models are meant only for good ethical fun/purposes such as trans people or roleplay, not for tricking/trolling/catfishing people
just because a youtube video does it, it doesn't mean it's good, most youtubers which promote the server for this got banned

[oak edge]

Hey man thanks it works now

any idea about the new embedder spin? (is it better than contentvec?) (My data set is in tamil - south asian language)

[low shard]

you've admitted to wanting to do e girl trolling when i asked you to
then tried to switch the topic to see "how good it is" when i told you that's not allowed

[low shard]

spin could be helpful especially for realtime model, the current pretrain by lyery, legacy core, is trained using vctk + m4singer dataset like the original pretrain, i'm not sure if it got tested for that language , but u could give it a try

[oak edge]

alright I'll try around with spin for a bit- fingers crossed. thanks for the help man

[low shard]

we don't want to be affiliated nor help anyone into trolling their friends with e girl models, it doesn't matter if you say after it's for "seeing how good it is" then change to "surprise ur friends"

[low shard]

You're welcome :D, i'm also updating my applio lightning.ai notebook currently btw https://docs.aihub.gg/rvc/cloud/applio-lightning-ai/ , it doesn't have as much free time as kaggle (80 hours monthly) but it allows web uis in free tier unlike google colab / kaggle

[oak edge]

WAIT colab doesn't give web ui in free tier?

i'm a bit confused I usually run applio and train in my free colab

(it lasts for 4 and a half hours every 3 days)

[low shard]

technically both google colab & kaggle don't allow web uis in the free tier, in fact both notebooks are encrypted, but if it ever gets caught they'd disconnect ur runtime or take action

usually it's safe since the code has layers of encryption tho, but im saying

[oak edge]

wait I've been doing it for months

[oak edge]

yeaa that's what keeps me alive ig

@low shard and yup colab fumbled with 2 hours and 40 minutes data set i think

[low shard]

im saying "if" hypothetically, it may happen in the future, like it may not, who knows, it doesn't get caught because it uses it's own encryption

[oak edge]

ohh alrightyy

[low shard]

kaggle offers T4x2
lightning.ai offers even better gpus for free but with shorter time

[oak edge]

how many hours could I expect

and cool down period?

[low shard]

depends on the GPU

damn u can get even a free A100 now

test if the T4x2 can do ur dataset with kaggle, else try lightning.ai

[low shard]

monthly

[oak edge]

a100 seems really pleasing can go to 30 batch size?

[oak edge]

ahh gotta be careful with my training set then

[oak edge]

so it's confirmed that my data set is too long for colab right?

[low shard]

i mean, im guessing that not all of ur datasets are 2+ hours 😭

[oak edge]

ehh these are like collective voice

i mean that's the only data set i have as of now Y_Y

[low shard]

yup

[oak edge]

WOW

[oak edge]

yup see

[low shard]

u don't need to unless u train like a pretrain tho 😭

[oak edge]

issue is my language mostly

[low shard]

ur dataset works fine with kaggle's t4x2 but didn't work with colab's T4?

[oak edge]

there are no pretrain for tamil

[oak edge]

yup

also I uploaded zip file for kaggle

it's around 1.79GB

and the audios aren't merged

as for colab

it's around 2.4 GB merged all into a single file using audacity

and 32kHz sampling rate

@low shard you think it maybe due to this audacity thingy?

[low shard]

exporting the audio from audacity shouldn't give issues

[oak edge]

then might be colab couldn't process that much amount

after all it only has 2 cores and kaggle has 4

[oak edge]

alsooo are the custom pre trains really necessary with this amount of data set? (i thought of using Titan 40k but then again..ahhh imma test all default with spin)

AHH main thing

forgot

what's the batch size i can go for with 2xT4 @low shard

[low shard]

btw be aware that T4x2 means it's using 2 T4 gpus, so the batch size has to be split for each gpu
like if u wanna train batch size 8, you need to put batch size 4 in the program

[oak edge]

i mean what's the max amount I can push

i can't see vram limit

in kaggle like colab

not even pushing, like what's the optimal amount to put @low shard

[opaque lance]

can someone help me with voicechanger?

[proud dawn]

Hello i need help

[low shard]

it uses 2 T4s, each has like 16, I'm guessing the mac could be 32, (which, as it's 2 gpus it has to be splitten to 16), but u shouldn't train on that

it would be better u check https://docs.aihub.gg/rvc/resources/training/#batch-size

[oak edge]

thanksie

[proud dawn]

i dont really sound like a girl with my voice changer

[low shard]

Hello :)
Please Elaborate:

• your PC GPU (Such as Nvidia RTX 3060, AMD RX 9070 XT..)
• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do? There isn't a program that does everything, there's a program for each thing:
  • AI Covers
  • Train RVC Models
  • E girl trolling
  • TTS
  • Roleplay in VC
  • Roleplay in Games
  • etc
• what tutorial link are you using
• a screenshot of the program

[oak edge]

shit i'll go for 8, so 8+8 16

[low shard]

try batch size 8 first, it's recommended for 30+ mins dataset in the docs

[low shard]

Are you trying to do e girl trolling?

[oak edge]

yu mine's like just 5 times more than that

[proud dawn]

No I want to roleplay in fivem

[oak edge]

holy

i clicked on stop trainin

i think shit's breakin down

nvm it stopped

holy poor kaggle Y_Y

[low shard]

yeah i love CPUs going over 100%

someone save that machine 💔

[oak edge]

poor guy is gonna be beaten when he finishes those 500 epochs

[low shard]

Hello! voice changer is too vague,
Please Elaborate:

• your PC GPU (Such as Nvidia RTX 3060, AMD RX 9070 XT..)
• your operating system (Like Windows 10, Windows 11, Linux, MacOS 26..)
• what you want to do? There isn't a program that does everything, there's a program for each thing:
  • AI Covers
  • Train RVC Models
  • E girl trolling
  • TTS
  • Roleplay in VC
  • Roleplay in Games
  • etc
• what tutorial link are you using
• a screenshot of the program

[oak edge]

it took 6 minutes for 1 epoch i'm cooked

[proud dawn]

im trying to roleplay on fivem as a girl, I'm on windows 11, rtx 4060 TI

And I can't send a screenshot

[low shard]

!give-media-perms 1h @proud dawn

[low shard]

u can now

[proud dawn]

This is what im using