✨│ai-help

[glacial pollen]

you're meant to use run bat

or rather, go webui bat

no need to play around environment

wdym by " not for me "

[trim wave]

no dw

it blocked

bcs windows

[glacial pollen]

so wut now

[trim wave]

i mean python was asking for permissions

but it opene

[glacial pollen]

Oh

welp

[trim wave]

the webui

[glacial pollen]

For future, allow for such

( as long the thing is trusted

[trim wave]

ok thank you

anyways

is there a specific time of audio

i have to put

for the original voice

[glacial pollen]

tbf... I still haven't fully figured it out but, seems like proper labeling / transcription + audio's quality itself is more important than the total length of the dataset ( samples ) butttt

I'd say, anywhere from 8 mins to 30 should be just fine

( I'd probs say, 15 to 20 mins being a golden middle )

but tbf, you could go with even 2-5 mins and it'll do okay ish
( And again, rvc quality enhancing could always be an option

[trim wave]

kk

and i use the rvc voice for these?

or bad idea

[glacial pollen]

As in, rvc's output for gpt-sovits samples?

orrrr, feeding rvc with gpt-sovit's outputs?

[trim wave]

bro

me english not main leanguage

haha

i jus wanna know

what i originally put

for the audio

[glacial pollen]

do you mean:

Using rvc's generated output as input for gpt-sovits?

or

Using gpt-sovit's generated output as input for rvc?

[hallow thistle]

Crazy.

[glacial pollen]

aka, rvc -> gpt-sovits or gpt-sovits -> rvc

[trim wave]

ohhh

wait

[glacial pollen]

Because if that's the case then I wouldn't recommend the first
instead, the latter, for sure

[trim wave]

yeah

i was talking

[hallow thistle]

Use RVC audio into GPT-SoVits or put GPT-SoVits audio into RVC?

[trim wave]

abt the first

dk if it was a good idea

rvc into gpu

gpt i mean

[glacial pollen]

well

it definitely won't go that well

being fully honest with you, gpt-sovits is way more sensitive to bad audio

than rvc is

It's just less forgiving

+, doing AI -> AI in terms of making models vs improving the output is overall not the best idea

[trim wave]

yeah

i know what u mean

[glacial pollen]

You see, rvc itself causes like a 15 to 40% loss of the voice's fidelity in the first place

and then gpt breaking it further

[trim wave]

but like i have to go for only voice clips on youtube of the charater voice i wanna clone?

[glacial pollen]

However, it works well when you do gpt - > rvc because rvc models are, baseline, higher in quality

that's why it will work

[glacial pollen]

Well, more or less yes

Then there's transcription

processing stuff and ye, training

[trim wave]

i see

what would you recommend so?

to do

[glacial pollen]

train gpt-sovits

train rvc model

Get output from gpt-sovits to have accurate speech and stuff ( according to your character's style etc )

and then feed rvc with gpt-sovits' outputs

aka. you improve sovits' output quality thanks to rvc's superior quality

so ye, gpt-sovits -> rvc ( not training tho, just inferencing / " cover " )

[trim wave]

and then i can use the voice for tts?

[glacial pollen]

no, you get your character's speech or whatever from gpt-sovits

and that output ( low quality ish )
you give to rvc

then rvc outputs you that, but better

rvc's only purpose in here would be improving the quality, nothing else

[trim wave]

okay i see

and where do i get from the orginal voice of the character

before gpt sovits

youtube?

[glacial pollen]

Well... this is actually up to you

You see, I work with Anime / japanese content so

[trim wave]

idk if theres a special website

for that

[glacial pollen]

I can as well use anime blu ray rips and isolate the vocals

[trim wave]

yeah maybe i will do the same

[glacial pollen]

or I can use visual novels, games etc ( and rip the files

[trim wave]

thats why

oh

i have davinci resolve studio after so i can isolate vocals easily

its just abt where do you find them

[glacial pollen]

well, you'd better use uvr / mvsep and the fv4 model for isolation

I can confidently say it's one of the best if not the best atm for " full voice " extraction

Either way, idk man where you can find it

[trim wave]

did u test the davinci one once?

[glacial pollen]

If you ask about Anime or something

then Nyaa is a good source

[trim wave]

y'eah

i know this website

[glacial pollen]

Davinci resolve is not a dedicated software for voice separation

[hallow thistle]

I used Demucs to separate background noise from an audio track.

[glacial pollen]

It is just a bonus

whereas uvr / mvsep are made for that, and so are the models, for specific usages, with specific traits

[trim wave]

yeah true

[glacial pollen]

so yeah, as they say, pick your poison 🤔

[trim wave]

which one do you use

[glacial pollen]

gabox's voc fv4

[trim wave]

oh

[glacial pollen]

You can find it on hugging face

[trim wave]

is it uvr or mvsep?

[glacial pollen]

both

I just prefer uvr atm

mvsep has some issues on my end so

either way, it'll perform the same way here n there

if you prefer mvsep use that, if you like uvr, you can use that as well, no difference ( maybe in speed? but yeah, Haven't tested it on mvsep so can't say - btw, by mvsep I mean the local one, not the website

[trim wave]

ok thank you for everything brother

i'll try to start

[glacial pollen]

Np

Best of luck, and in case of problems, there's ' Audio separation ' discord, they can help you out

uvr and mvsep devs and such, in there

[trim wave]

oh yeah wait before forgetting to ask

whats the rvc software u use

[glacial pollen]

I use my fork

[trim wave]

fork?

[glacial pollen]

https://github.com/codename0og/codename-rvc-fork-3/tree/main

[trim wave]

oh

[glacial pollen]

Just my own take on applio, let's put it that way

[trim wave]

i thought u were making a joke with that

[glacial pollen]

lmao, no

[trim wave]

can u send me the uvr link please

@glacial pollen

and i'll let u

(i'll try)

[glacial pollen]

you gotta join the server, audio separation one

Links are there

[trim wave]

oh

[glacial pollen]

for specific uvr beta

and patch for it

[trim wave]

wdym there

[glacial pollen]

there

[trim wave]

i know

but wheres the link

to join serv

[glacial pollen]

👀

[trim wave]

bruh

was it that easy

[glacial pollen]

it always is that easy

just people never use google

¯_(ツ)_/¯

lol

[trim wave]

mb

[glacial pollen]

Either way, that's the place

Pretty sure people gonna help you in there

[blazing solar]

Can someone help? Why i cant download the audio

[glacial pollen]

Is this android

[civic ivy]

обезьяна

[formal wind]

Welp. I keep getting an error

"Process Process-1:
Traceback (most recent call last):
File "C:\Users\User\Downloads\Codename-RVC-Fork-V3.0.4\Codename-RVC-Fork-V3.0.4\env\lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Users\User\Downloads\Codename-RVC-Fork-V3.0.4\Codename-RVC-Fork-V3.0.4\env\lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "C:\Users\User\Downloads\Codename-RVC-Fork-V3.0.4\Codename-RVC-Fork-V3.0.4\rvc\train\train.py", line 590, in run
reference,
UnboundLocalError: local variable 'reference' referenced before assignment"

It happens whenever I start trying to train

I can do every other step including generating the index

its just training

[opal kelp]

I have three datasets, two of 18 seconds and one of 24 seconds, how many epochs are recommended for each?

[gloomy cairn]

like this outdated video https://www.youtube.com/watch?v=bP8AMf20MAY&ab_channel=kalozalt can I make real time voice change?

[low shard]

Oh you were the Mac guy

[gloomy cairn]

ya

[low shard]

RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models

Wokada = uses RVC for realtime inference

Use #🔍│help-w-okada

[gloomy cairn]

Yeah I'm trying to use the speech to speech, my bad with the wordings @low shard

[formal wind]

You awake to help me out?

[low shard]

Yeah you meant realtime voice changer for calls or just ai covers?

[glacial pollen]

that's weird

show me your log folder

[formal wind]

You want the log folder or the model folder thats in the logs folder

[gloomy cairn]

The former yeah but not for calls. itll be for content similar to the start of this video https://youtu.be/NTfVUDYSmpE?si=pnOhAE1iC9nES--n

aka titled How to Dub Anime with the help of AI

[glacial pollen]

Hmm.. does the reference folder contain files?

[glacial pollen]

Hm. That's quirky ngl
That error shouldn't be triggered under any circumstances

What were the steps you did prior to training?

preprocessing, feature extraction, index and training?

Well, tell me if you got the zip or cloned the repo

[formal wind]

Wdym

[gloomy cairn]

I guess I dont really need realtime voice changer but just that functionality of speech to speech that doesnt sound so AI. wups.

[glacial pollen]

how did you get the fork, via repo dl or zip dl ( releases )

[formal wind]

Zip I think

[glacial pollen]

lemme inspect the training script rquick

[formal wind]

Where tf is that

The train.py?

[glacial pollen]

Nah, I'll handle that

inspecting the zip rn

[formal wind]

oh lol

[glacial pollen]

Yea, no abnormalities there

In fact, it never happened to me or any other users

huh 🤔

[glacial pollen]

Because it technically can't

[formal wind]

Bro why do I get cursed with new errors

[glacial pollen]

Did you click or tweak anything specific?

[formal wind]

I didn't really tweak much

[glacial pollen]

was there any other traceback ( earlier ) in console?

[formal wind]

Wdym traceback

[formal wind]

Ummm

I closed the console a while ago

[glacial pollen]

now, an important one

which sample rate did you check

[formal wind]

32

my dataset was 32k

[glacial pollen]

stock vocoder?

hifigan?

these

[formal wind]

Yeah

[glacial pollen]

well, weird shit weird shit, completely without any logical explanation

Move the fork folder to C drive

c/fork/run bat and all the rest

[formal wind]

Bet

[glacial pollen]

and retry, lemme know if that fixes the issue

and if not, this time fully inspect the console to see if there's no earlier tracebacks

[formal wind]

Sure can do

Holy fuck I see why its probably erroring now

[glacial pollen]

yea?

[formal wind]

An error occurred extracting file C:\ApplioForkTraining\logs\AdamBudgetCuts\sliced_audios_16k\0_0_11.wav on cuda:0: CUDA error: invalid argument
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

it says that for every processed dataset file

thats like 50 errors

[glacial pollen]

Oh yea, knew something was off

reference files aren't being moved to device

[formal wind]

Pre-processing it is fine.

extracting features is the issue

[glacial pollen]

because there's no gpu involved

both training and extraction uses gpu

Hmmm..

[glacial pollen]

@simple ore Any idea why that'd happen? ( zluda ) or is there something I don't know but you do

Either way @formal wind, I'm not sure what's the issue
From what I remember, zluda should work just fine and support the training ( otherwise what was the point of it in the first place as directml existed in rvc )

so I suppose, Noobies would have to take over this case

[formal wind]

Fuck man

alright

[glacial pollen]

but all I know for sure is that it ain't the fork

changes I've made aren't affecting that part of functionality so it must be your pc specific thing or indeed something zluda-related I am unaware of

[analog obsidian]

for amd it is best to train in the cloud, even if you configure zluda well, the performance is still terrible compared to nvidia

to be honest its not worth the suffering of doing the zluda rute for such horrible speeds

[formal wind]

So what am I supposed to do here exactly

[glacial pollen]

the whole deal was about using newest applio

Unless you know colabs / kaggles that use the up to date stuff

[analog obsidian]

kaggle uses the latest applio

[glacial pollen]

( matter of logging

[formal wind]

What...

[analog obsidian]

u just remove the line that tells it to download the compiled version

xD

[formal wind]

What applio is the latest

[glacial pollen]

oh, welp 🤔

[formal wind]

there are alot of applios on there

[analog obsidian]

just remove that

https://www.kaggle.com/code/deiant/applio

[glacial pollen]

welp ¯_(ツ)_/¯

[formal wind]

Oh I use that one alrighty

[analog obsidian]

it'll download main branch instead

[glacial pollen]

I didn't use kaggle so, wouldn't know

[formal wind]

Well all the struggle for nothing lol

[analog obsidian]

which has the very good avg logs

[glacial pollen]

Either way, think it'll be time to start making my own kaggle

[analog obsidian]

🦈 🫂

[glacial pollen]

for cases like this

[knotty moth]

noobies has done zluda stuff, in the end only the vram matters

[formal wind]

Thank you for your help both of you!

I look up to you Lyery 🙏

[analog obsidian]

no problem, just dont forget to remove that line

[harsh ravine]

any better ai voice models than weights?

[gloomy cairn]

I ran chmod +x run-applio.sh
./run-applio.sh to run on macOS using these instructions https://docs.applio.org/applio/getting-started/installationbut its either waiting for an input or its running indefinitely

what did i do wrong?

[simple ore]

empty train loader

[glacial pollen]

no I mean, why the extraction fails

[simple ore]

no vram?

[glacial pollen]

wait fr?

lmao

zluda can't be that bad... can it?

[gloomy cairn]

Traceback (most recent call last):
File "/Users/name/Desktop/AI/Applio-3.2.8-bugfix/app.py", line 22, in <module>
from tabs.inference.inference import inference_tab
File "/Users/name/Desktop/AI/Applio-3.2.8-bugfix/tabs/inference/inference.py", line 17, in <module>
from tabs.settings.sections.restart import stop_infer
ImportError: cannot import name 'stop_infer' from 'tabs.settings.sections.restart' (/Users/name/Desktop/AI/Applio-3.2.8-bugfix/tabs/settings/sections/restart.py)

[simple ore]

no, it is fine

whatever is happening is not zluda

[glacial pollen]

well, then that's some abnormality

[simple ore]

like maybe he did not slice anything

and trying to extract features on 20 min long tiles

[gloomy cairn]

All i did was download zip chmod +x run-install.sh
./run-install.sh
After installation, run:

chmod +x run-applio.sh
./run-applio.sh

and it returned those errors. Could i get help on this?

[glacial pollen]

@formal wind did you slice stuff

please tell me you did

[knotty moth]

20 min file unsliced?

[simple ore]

did you screw with /assets/config.json?

[gloomy cairn]

no I don't know how to touch allat so i just downloaded zip unzipped and ran the codes in the instructions

if i did tho how can i fix it?

[simple ore]

then you screwed something up

because I've seen this error recently and that was from someone fucking up the config

[gloomy cairn]

How did he fix it?

[simple ore]

not messing up with assets/config.json where he forgot to put a comma

[gloomy cairn]

i dont even know where that is tho

[formal wind]

I did lol

[glacial pollen]

welp

[gloomy cairn]

{
"theme": {
"file": "Applio.py",
"class": "Applio"
},
"plugins": [],
"discord_presence": true,
"lang": {
"override": false,
"selected_lang": "en_US"
},
"flask_server": false,
"version": "3.2.8-bugfix",
"model_author": "None"
}

yea i def didnt go in this file and mess with anything

[simple ore]

that looks fine, so something else is messed up... did not unzip properly or something

aka missing i18 files with translations

[gloomy cairn]

is there some other way to unzip properly?

[simple ore]

is that linux?

[gloomy cairn]

no its a macOS

I did unzip Applio-3.2.8-bugfix.zip

[simple ore]

well, duh

[gloomy cairn]

lol

[simple ore]

https://docs.google.com/document/d/10ZyhkaJGdoa_M7QVUmlNSe94brAYG1BusKS4EHTrtwc/edit?pli=1&tab=t.0

delete what you done and do it properly

[gloomy cairn]

thank you!

[simple ore]

dont expect to be able to train

[gloomy cairn]

I wont, im just gonna use it for speech to speech

and for that I would find one from the voice models right

[simple ore]

start with the start, what GPU, did you follow the amd install guide to the letter?

[simple ore]

yeah, inference runs on CPU though, so it may be a bit slow

but it is what it is

[formal wind]

I'm not doin' local applio

I don't got the brains to keep tryin' that

[simple ore]

why does you error mentions windows local path then?

[formal wind]

No Like I mean I dont wanna attempt local anymore

I was trying to fix local

[simple ore]

what's your GPU again?

[formal wind]

AMD Radeon RX 6600

[simple ore]

8GB ram should've been fine

as long as you follow applio's installation instructions properly

but colab should be just as fast

so not much point to run locally, you can train stuff in the cloud and play some games while it is happening

[formal wind]

Bet thanks!

[trail kiln]

any fixes for "failed to load asio driver / error 0 ????? im trying to start the rvc and get the voice changer done and i have selected both flex asio as inputs and it says in cmd "failed to load asio driver / error 0"...... cant seem to get it working

[analog obsidian]

use this channel - > #🔍│help-w-okada
help-rvc is for rvc, not wokada

[trail kiln]

oh mb

[gritty merlin]

what is the lastest version of rvc?

[deep nova]

anyone got any methods to reduce chopyness as some words arent said although im using a high end gpu

[analog obsidian]

#🔍│help-w-okada

[glacial pollen]

Increase the chunk, extra and fiddle with noise supressors ( on off, or change type ) also having a proper distance from mic ( and do not use index

Also As Lyery said, use the w-okada channel

[queen nimbus]

how long on average does it take to train a model on roughly 300 lossless wav files

[glacial pollen]

a wave file can have varying length

[queen nimbus]

20 seconds each rouhgly

[glacial pollen]

aside, you can't really estimate it

sadly

This is not linear like that, unfortunately

[queen nimbus]

i just wanna know if it will take less than a hour or more

:p

[glacial pollen]

As I said, you can't estimate it at all, not even a chance

In machine learning you use the metrics to know when the training's " done "

In case of rvc / applio, you use the tensorboard

Aside, using sets above 1 hour has almost no point unless you're absolutely sure there's a shit ton of variety in there and I mean it, meaningful variety and diversity

Else you're exposing the model to a risk of biasing towards data of similar patterns and that in consequence, can lead to overfitting / bad generalization = bad model

[queen nimbus]

i see

[glacial pollen]

Fear not tho, the docs ( and so, the guide ) has some nicely explained stuff

matter of checking them out

[brittle wing]

Yo, how to enable FP32 on Cordane Fork?

[analog obsidian]

enabled by default

dw

fp16 got removed (in the latest version)

[stoic forge]

hey, new here. might be a stupid question. Is there any way to use trained beatrice v2 models like you can with rvc models?

[analog obsidian]

yes, with a vst

[brittle wing]

Oh okay, thanks for the info! So FP32 is enabled by default in the latest version, no need to change anything.

[stoic forge]

sorry can you explain what a vst is

[analog obsidian]

beatrice is not meant to be used in a webui like rvc

but more in a DAW

like fl studio

https://twitter.com/i/status/1700696685884969096

Beatrice作ってる人@声質変換VST (@prj_beatrice) on X

超軽量・超低遅延なAIボイチェン「Beatrice」正式リリースしました！🎉🎉
CPUシングルスレッドでみんなつくよみちゃんになりましょう！！
https://t.co/dtkPZO0hqa
#つくよみちゃん

X · Sep 10, 2023, 2:24 AM

^

thats the only way to use beatrice

outside original w-okada

every beatrice info available is in japanese tho

[stoic forge]

thank you so much

[analog obsidian]

https://prj-beatrice.com/

good luck!

[formal wind]

Wait should I be looking at loss_avg or loss in tensorboard

[analog obsidian]

avg loss

ignore old loss graphs

[formal wind]

Bet

[karmic olive]

📚 Documentation

AI HUB Docs

https://docs.ai-hub.wtf

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[radiant jay]

hello I'm sure I do everything right, but ı get a mistake in the way. Can you help me ? " mv: cannot stat 'MMVCServerSIO.py': No such file or directory
/content/voice-changer/server/HVoice.py:3: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
from distutils.util import strtobool
Traceback (most recent call last):
File "/content/voice-changer/server/HVoice.py", line 10, in <module>
from downloader.SampleDownloader import downloadInitialSamples
File "/content/voice-changer/server/downloader/SampleDownloader.py", line 12, in <module>
from voice_changer.RVC.RVCModelSlotGenerator import RVCModelSlotGenerator
File "/content/voice-changer/server/voice_changer/RVC/RVCModelSlotGenerator.py", line 4, in <module>
import torch
ModuleNotFoundError: No module named 'torch'
WARNING:pyngrok.process.ngrok:t=2025-03-12T02:21:03+0000 lvl=warn msg="Stopping forwarder" name=http-33487-51a86839-104e-4c55-9bd3-f2b82bb15f8f acceptErr="failed to accept connection: Listener closed"
--------- SERVER STOPPED! --------- "

[knotty moth]

syntax error moment

[feral marsh]

generating sola buffer?

my voices just suddenly dont work? i havent used rvc in a week and they just wont work. no audio when i select models

i also checked my mic , it works

[analog obsidian]

#🔍│help-w-okada

rvc is not w-okada

[feral marsh]

ik.. i use rvc

realtime?

[analog obsidian]

#🔍│help-w-okada

rvc stands for retrieval based voice conversion, not realtime voice changer

w-okada is a software that allows u to use rvc models in realtime

so this channel is for rvc (not the voice changer)

[feral marsh]

OH

DANG

okok

ty

[rich marsh]

which colab with web ui + vocal remover from youtube is greatest. mine i saw this llast year doesnt work anymore 😦

[knotty moth]

rvc or vocal remover? the latter baked in rvc is nothing but a bloat while there are better alternatives:

• [UVR5 UI](#📰│dev-updates message)
• [MSST colab inference (tweaked version)](#📰│dev-updates message)

[rich marsh]

i was adding my ai vocals on a song. i was putting song Let's say Metallica - Enter sandman , and i was adding Elvis presley voice model, colab were downloading the youtube metallica song+removing vocals+adding elvis voice on it. it was great but now doesnt work

[knotty moth]

colab has banned the downloader, so you'd better do it manually first

[rich marsh]

but i cant even open webui think gradio or ngrook thing. i wish problem was only downloading from youtube 🙂

[low shard]

share a screenshot of the issue

and what google colab link do u use

and what’s ur pc gpu

[rustic dome]

@cosmic spire

[cosmic spire]

what

[rustic dome]

i need jelp

help

[cosmic spire]

bro

[knotty moth]

Don't ping anyone without reason

[rustic dome]

i just said help

[rustic dome]

can u help then

[knotty moth]

no

!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.

[rustic dome]

can u help then

[knotty moth]

https://tenor.com/view/how-no-sloth-about-chill-gif-24912171

[hallow thistle]

How about I telling you how to behave?

Describe your problem what you gonna do about RVC. If you wanna get help about W-Okada the realtime voice changer, go to #🔍│help-w-okada.

Don't ask someone to ask you back. That's the bad thing you could do in a help support.

[uneven relic]

Hi, on applio, I need to merge 2 voices but when I put the two pth files of the templates and click on fusion it gives me an error

[hallow thistle]

Can you send the terminal part of Applio? because I can't read French.

[uneven relic]

I don't understand, what do you want to read?

[hallow thistle]

Um, damn it. The screenshot of cmd of Applio you're running.

[uneven relic]

oh, ok

[uneven relic]

do you have to put only the pth files to merge 2 models?

[hallow thistle]

That looks like to be a lot of problems there.

[uneven relic]

I tried several times

[trim wave]

Does someone know why i don't have the faster whisper v3 turbo asr mode in my ui?

just have this

@glacial pollen

[hallow thistle]

Opera GX spotted.

[trim wave]

oh

which browser

should i use

@hallow thistle

[hallow thistle]

Use Google Chrome, Mozilla Firefox or Microsoft Edge instead.

[trim wave]

and so do i have to delete all i did until now

in the repo

@hallow thistle

[knotty moth]

not related to his issue anyway

[trim wave]

Can you tell me whats the problem so pls?

[knotty moth]

you'd better search the GPT sovits support server

[trim wave]

i'm here now do you know if i have to check "choose audio" or no?

[trim wave]

did that

new problem on top

[lament eagle]

I NEED line of code or that 1 file that can fix the split bug infer for Applio Kaggle, or if anyone here know it feel free to send

[knotty moth]

im sure it is already in the main branch

[trim wave]

by any chance you dont know if i have to check yes in "choose audio" (image on top)

[brittle wing]

Yo, I need some help. Do you know which one is the best between HiFi-GAN, MRF HiFi-GAN, and RefineGAN for RVC Wokada?

[analog obsidian]

hifigan

refinegan got deleted in the latest applio update due to issues

gives models a robotic/metallic sound

[simple ore]

not deleted

just hidden until pretrains are done

blaise wanted to make a release with other fixes

[knotty moth]

yea, still I feel refinegan has room for improvement

[simple ore]

what is actually gone is 44100Hz sampling rate

unfortunately refinegan gen had some bugs

[upbeat carbon]

https://tenor.com/view/venti-genshin-impact-ventispin-spin-gif-25191045

how do you do pitch extraction?

[glacial pollen]

wdym how

There is a button for that within the ui

First the dataset has to be processed and after that, you'd extract the pitch

( pick rmvpe ) and that's it

[upbeat carbon]

i can't show the images dang it

[glacial pollen]

You'd need level 10

give me a sec

[upbeat carbon]

can you show me please?

[glacial pollen]

I'll do it instead for you

[upbeat carbon]

thank you

[glacial pollen]

So, which part of it is unclear?

[upbeat carbon]

huh!?!?

mine looks alot diffrent

wait

which voice changer are you using if i may ask?

cuz i'm using the realtime voice changer client

mabye a diffrent one idk

[knotty moth]

because ur asking it here instead of #🔍│help-w-okada

[upbeat carbon]

oh mb

[upbeat carbon]

btw is it a diffrent voice changer or something?

[analog obsidian]

is rvc
rvc and w-okada are not the same thing

w-okada is a software that allows realtime rvc inference

rvc is for training ai voice models
w-okada is for using them in realtime

rvc can also do local conversion, speech to speech (not realtime)

[upbeat carbon]

Ohhhh thanks cuz i'm using the w-okada one

Also do you have a link for rvc?

[analog obsidian]

then your question has no sense because w-okada does not extract the pitch of an audio

it estimates the pitch in realtime

you want to train a model?

[upbeat carbon]

Well i just want to extract the pitch of the models so i guess yea

[analog obsidian]

you want to extract the pitch of a model???

thats... nonsense

🦈

for training a model you first need a dataset, then in the preprocess steps you do f0 estimation, which gets saved in the model's logs folder
but is not what you think, the pitch saved are just random numbers together

[upbeat carbon]

Yes..

[pastel oak]

Can you use different words to describe what youre trying cause that doesnt make sense

[upbeat carbon]

Who? Me?

[pastel oak]

ys

[upbeat carbon]

I'm trying to extract the pitch cuz some models like minos prime need it

So i just don't extract it and instead i just set the number?

[pastel oak]

que pasa

[upbeat carbon]

?

[pastel oak]

can you give a visual example

or audio

of what the issue is and how it should sound like

[upbeat carbon]

Basically it requires rmvpe or crepe for the extraction thingy

[pastel oak]

Ok that is something you select

Those options are all avialable by default on wokada

[upbeat carbon]

Yup that's right and i indeed selected it

[knotty moth]

it's pitch estimation for both realtime & non realtime rvc

[upbeat carbon]

Oh okie thanks!

[analog obsidian]

use rmvpe

woa so i finally understand

he was trying to ask which f0 estimation to use

🦈 🔥

[wise iris]

Hey, where can i find a tutorial to use the ai voices ? I dont want to train i just want to test the models that re already created

[teal ridge]

Is this really supposed to happen when I started training this? (I am using the no-UI Applio Colab notebook).

It even saved my index file a bit early.

I just quit the training because it's not showing the epoch numbers from it.

[odd shale]

There you have a guide buddy. It got all you need to know.

-rvc

[karmic olive]

📚 Documentation

AI HUB Docs

https://docs.ai-hub.wtf

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[wise iris]

thankss

[odd shale]

You're welcome.

[teal ridge]

Has anyone got the same problem or should someone fix it?

[simple ore]

No it is not, but it looks like tensorboard's dependency got updated and it is not compatible with what applio had installed

[teal ridge]

Well sh_t.

[simple ore]

make a new code cell

and run it

maybe that may fix it

@teal ridge

[radiant jay]

Can someone help me please

[fervent rover]

Is The RVC Mainline Colab Fixed?

Just Asking

[radiant jay]

yardımcı olacak bir türk yok mu ?

[heady plover]

Been thinking, what are the best tools to separate a voice from music for something like a podcast? annoying when training

spleeter is a pile of slop

[knotty moth]

spleeter and demucs are too ancient, I'd recommend some best mel roformer models in applications such as:

• [UVR5 UI](#📰│dev-updates message)
• [MSST inference colab](#📰│dev-updates message)

[hallow thistle]

Please speak English.

!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.

[heady plover]

oh ohly shit there's a Pinokio image for one of them, thanks!!

[fervent rover]

Is The RVC Mainline Colab Fixed?

Just Asking

[glacial pollen]

elaborate

!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.

[glacial pollen]

You need to describe what you use, when it happens and stuff

Don't provide vague descriptions of issues

we can't read minds, unfortunately

[knotty moth]

im sure even chatgpt is more likely to hallucinate when trying to answer such vague question

[glacial pollen]

In any case...
Given you mention onnx which applio / rvc doesn't support, you're talking about w-okada.

Now.. whether you mean the core components or the voice models themselves, you need to get such things.
rmvpe ( f0 extractors in general ) along with hubert / cvec should be bundled, so like yea

You must refer to voice models
They can be either a pytorch format ( pth ) or onnx

Those can be acquired from #🔍│find-models #1175430844685484042 or weights.com

[hallow thistle]

You've asked the same question for the third time already.

[knotty moth]

bruh

[hallow thistle]

Shit.

I don't know about the mainline RVC Colab. But I can only think of Applio the RVC.

[fervent rover]

Well I just wanna to know if anyone has fixed it The RVC Mainline Colab already so I can go back to training the voice models, I apologize for that

[glacial pollen]

just use alternatives

kaggle or whatever

[nocturne patrol]

i went to upload one of the voice models i found here and it says i couldn’t upload it because the file was not “onnx” or “pth”

[glacial pollen]

and what is the model you're trying to upload then

extension

what is it?

[hallow thistle]

[glacial pollen]

I have a feeling you're just dragging a zip in

[knotty moth]

that topic should be discussed in #🔍│help-w-okada

if it is voice changer

[nocturne patrol]

okay well i don’t know

it’s

[glacial pollen]

#🔍│help-w-okada go in there

[nocturne patrol]

the gumball one

i’ll try and find it

okay

[hallow thistle]

For W-Okada the voice changer, go to #🔍│help-w-okada. This channel #✨│ai-help is about RVC programs.

[fervent rover]

No Thanks, I either prefer to just again play the waiting game because I came impatient so yeah, I learn my lesson for that, But Thanks👍🏻

[hallow thistle]

RVC in this context doesn't stand for realtime voice changer.

[knotty moth]

doesnt mean you would spam the same shit

[fervent rover]

Yeah I know

[hallow thistle]

What are you talking about? I was pointing how you asked the same question for third time.

[glacial pollen]

Well either way.. if he wants to wait then so be it

but imo it's a waste of time. Colab's like a bpd person

[hallow thistle]

Well, that's pretty much it if he knows how to code.

[glacial pollen]

one time it's aweee uwu, one time it's shitty shit

[glacial pollen]

true

[formal wind]

"Shitty shit"' is goin' in my quote book.

[glacial pollen]

lol

[rancid ridge]

Bro , can some one help me please ?

[full aurora]

Is vonovox legit? https://github.com/dr87/Vonovox

[simple ore]

yes, but some features are behind a paywall

[full aurora]

ty!

[low shard]

Yes, it's another fork of WOKADA, it's around the same performance as the deiteris fork

Also it's better u talk about this in #🔍│help-w-okada

Btw Vonovox is Nvidia only

[full aurora]

Sry didnt know it was part of wokada

[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]

Please be specific on which problem you encountered about an RVC program.

[hallow thistle]

I don't know about the Vonovox, but Detris' W-Okada is better at doing realtime voice changer.

[pastel oak]

I mean we talked about it no? 💀

Or did i skip my explanation by accident

[full aurora]

U said u had someone making a new once but never mentioned a name

[blazing solar]

-colab

[karmic olive]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• Hina's Mod AICoverGen WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Hina's Modified Original W-Okada's Realtime Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab
• 🆕 Music Source Separation Training (Inference), by Jarredou & Makidanye Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

More commands: -audio, -hf, -local, -kaggle, -realtime, -rvc, /help

[blazing solar]

-rvc

[karmic olive]

📚 Documentation

AI HUB Docs

https://docs.ai-hub.wtf

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[low shard]

be aware of #📰│dev-updates , many colabs might be broken

[glass igloo]

Hi. What is the best way to improve the quality of the model without increasing the dataset? For example, is it possible to train the model to a specific voice? I mean that the model would learn to convert specifically my voice to the target voice. Or maybe it is possible to add text to the incoming voice, so that the model would understand the words better? At the moment my trained models often have problems with hissing, buzzing and whistling sounds.

[low shard]

RVC is Speech To Speech

and no you can't train a voice model to adhere specifically to your voice

get better cleaner dataset and look at https://docs.aihub.gg/rvc/resources/training/

[calm basin]

Hello, I got this error:

D:\RVC1006Nvidia>runtime\python.exe gui_v1.py
2025-03-13 18:26:47 | INFO | faiss.loader | Loading faiss with AVX2 support.
2025-03-13 18:26:47 | INFO | faiss.loader | Successfully loaded faiss with AVX2 support.
2025-03-13 18:26:47 | INFO | configs.config | Found GPU NVIDIA GeForce GTX 1050 Ti
is_half:True, device:cuda:0
Input device: 7:Microphone (MICUSB) (Windows DirectSound)
Output device: 16:Speakers (Realtek HD Audio output) (Windows WDM-KS)
cuda_is_available: True
Exception in thread Thread-1:
Traceback (most recent call last):
  File "threading.py", line 980, in _bootstrap_inner
  File "threading.py", line 917, in run
  File "D:\RVC1006Nvidia\gui_v1.py", line 653, in soundinput
    with sd.Stream(
  File "D:\RVC1006Nvidia\runtime\lib\site-packages\sounddevice.py", line 1800, in __init__
    _StreamBase.__init__(self, kind='duplex', wrap_callback='array',
  File "D:\RVC1006Nvidia\runtime\lib\site-packages\sounddevice.py", line 898, in __init__
    _check(_lib.Pa_OpenStream(self._ptr, iparameters, oparameters,
  File "D:\RVC1006Nvidia\runtime\lib\site-packages\sounddevice.py", line 2747, in _check
    raise PortAudioError(errormsg, err)
sounddevice.PortAudioError: Error opening Stream: Illegal combination of I/O devices [PaErrorCode -9993]

[river acorn]

Guys, is there a good step by step guide to get applio rvc to work with the rtx 5000 series cards?

[gloomy cairn]

why is my locally installed applio taking forever to convert?

[low shard]

What's your PC GPU and how long is the file

[gloomy cairn]

its a mac pro

and the file was like 2 minutes long

[low shard]

Follow to download it as said it in https://docs.aihub.gg/rvc/local/applio/ , but after you extracted the precompiled, go to the path in Windows explorer, write "CMD" and press enter, then in CMD write env\python -m pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

This should work

[low shard]

Uhh I'm not sure if that still supports MPS, what version are you using

Also which m

[gloomy cairn]

its m2

[low shard]

Alr, what applio version did you download

[gloomy cairn]

3.2.8

its the newest version

[low shard]

The one from our docs that has a link for the latest huggingface stable release?

[gloomy cairn]

i got it from here: https://github.com/IAHispano/Applio/releases

is this not it?

[river acorn]

Thanks for the reply! 🙂
I also asked in the Applio Discord and the answer I got was very similar, but the thing is, after trying to install the new "nightly" version, It just gave me a bunch of "Requirement already satisfied" answers.
But from what i could see they are/they were a bunch of cu121 and not cu128 "files". So I first had to run: env\python -m pip uninstall torch torchvision torchaudio
And after that I did run the command to install the new nightly "version" and now it seems to be working! 🙂

[low shard]

Oh alright, goodluck

[gloomy cairn]

is there anything i could do to optimize the converting speed @low shard

[pastel oak]

You are using 2 different inputs and output audio devices. Your first is windows directsound the second wdm-ks, both have to be the same. Use MME on both

[simple ore]

also need to uninstall old torch torchvision torchaudio using pip uninstall

but that's it

[deep nova]

anyone know how to fix cable output only being in input section

in rvc

[bronze pier]

i've been away from the ai voice stuff for a few years

as of now whats the easiest way to train a voice model

[pastel oak]

The cable labelling is weird. Cable output is in input, cable input is in output, you can ignore the labelling they are all in the correct section

[pastel oak]

locally with rtx nvidia or cloud

[low shard]

thx, I saved it as one of my copypastas

[low shard]

what's ur pc gpu

[low shard]

U can train on AMD too btw, just not as good

[simple ore]

on 7900xtx? pretty good

[low shard]

yeah also depends on the AMD gpu

how's it going after the switch btw?

[bronze pier]

7.8gb

[low shard]

what

that seems to be just memory, it could be storage, it could be ram, since you just said the unit for memory GigaByte

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

[simple ore]

6700xt -> 4070TiS is 6-7x faster, can also use a bunch of other things like flash attention, triton, etc

those were ass to install with amd/zluda combo

[low shard]

Nice, I don't think you regretted it at all lol

welcome to the green side

[simple ore]

nope, been running it prety much non-stop for last 2.5 month

with 6700xt I'd given up long ago

[low shard]

🔥

[brittle wing]

is that under rocm on linux
cause im jw how amd would compare to just directly using nvidia on a card with equivalent memory

[glacial pollen]

green ftw

💚

[low shard]

AMD and Intel are good only for cheaper gaming

But for AI Nvidia is da best

[glacial pollen]

Imo AMD cpu x Nvidia gpu is just a godtier

I honestly have had enough of intel, never again

[brittle wing]

i still want to know how rocm compares to natively doing stuff on nvidia

i heard from like one person it works fine but i have no frame of reference

[simple ore]

windows, hip sdk + zluda

[brittle wing]

ah i see

i know nothing of that

[fresh cairn]

anyone know why rvc isnt working on linux on AMD?

[severe kestrel]

can someone help me with the client like my voice is a bit robotic and idk how to fix that

[formal wind]

Ask in https://discord.com/channels/1159260121998827560/1159290161683767298

[bright sparrow]

anyone know how to make it your model/any model speak a different language than english?

like for example it works speaking another language but it doesnt make certain sounds

im using applio

[bright sparrow]

i assume it's got something with embedder model "Model used for learning speaker embedding."

though there's only asian things here

there's custom but idk what to do with ti

alright I got it figured out lol.

for anyone trying the ai to speak a language more like the input file just make sure the "search feature ratio" is set to a low value

[knotty moth]

I'm curious of 9070 XT, was thinking it could have performance (optimization) leap

[hallow thistle]

You mean an index file of an RVC voice model or a TTS?

[bright sparrow]

nah i meant like for example i'd put in an audio of me speaking my language and when the ai'd say it it'd say it but skip over the special sounds like just not pronounce them completely

stuff like ł or ć yknow

[thin edge]

I waited long enough but this error persists “ERR_NGROK_8012”

[hallow thistle]

You can either take the index file out or reduce index ratio down, but the result might be unexpected.

[hallow thistle]

Are you using Applio with ngrok?

[thin edge]

collab

[hallow thistle]

No, that's not what I meant. I know you're running from a cloud service like Google Colab and Kaggle. But is it W-Okada or RVC?

[thin edge]

mainline, rvc i guess

[hallow thistle]

I don't know about the mainline RVC for Colab. A lot of people here complained to me this specific Colab notebook won't work.

[thin edge]

oh so this is a common problem like when they change the python version ?

[hallow thistle]

Likely.

[crystal shore]

Wrong channel sorry

[hallow thistle]

https://tenor.com/view/i-saw-what-you-deleted-cat-gif-25407007

[thin edge]

oh welp, thanks yuuka

[hallow thistle]

You're welcome.

[gloomy cairn]

how long does it take to usually convert like a 3 min audio on applio?

[knotty moth]

depends on your gpu

you can't find the answer till you try it

[thin edge]

I can't put my dataset into this folder.

[knotty moth]

put in anywhere else

[thin edge]

Is that a joke ?

[knotty moth]

does it work or nah?

I don't get why you think so

[hallow thistle]

You think it is a joke?

[thin edge]

because of this

[glacial pollen]

because:

1. You have to first do the preprocessing of the dataset:

2. You have to have the proper model folder selected:

3. It has to contain sliced_audios and sliced_audios_16k folders

Else you'll encounter:

no-feature-todo

It quite literally says " no feature extraction to execute "

[glacial pollen]

because feature extraction is done on samples, which after preprocessing end up in

It's pretty straightforward dude.

[thin edge]

thank you 👍

[glacial pollen]

yw, hope it works now

[merry sand]

Is there a tutorial on how to use this Applio thing? https://colab.research.google.com/github/IAHispano/Applio/blob/main/assets/Applio_NoUI.ipynb#scrollTo=v0EgikgjFCjE

[low shard]

https://docs.aihub.gg/rvc/cloud/applio-no-ui-colab/

[rich moss]

Hey, can someone check this spectrogram from Spek and lmk if it looks good for training a voice model? Or is the quality too low pls?

[latent kettle]

It's good

Select high sample rate when training

48khz

[rich moss]

ok ty

[latent kettle]

Or 44khz

[analog obsidian]

did you upscaled this audio?
(apollo, resemble, etc)

high frequencies look very artificial

its bad to have fake frequencies in the dataset, rvc gets confused during training

it cant work well with synthetic data

better train the original audio (without any sort of upscaling)

rvc already upscales the audio in the training process

[knotty moth]

not exactly "upscaling" but it is sort of pretrain ability. when going further epochs it may slowly reproduces the dataset including the cutoff

[analog obsidian]

temporal upscaling

[knotty moth]

nah DLSS 4 & FSR 4 are better

[analog obsidian]

what about intel xess

[idle crypt]

Hi! how are you guys?

I already trained a model but I wanna increase the training, do you know how I can do that on Google Colab?

[odd shale]

Simply (i'm not sure if you're using Applio colab) don't delete your training files from the file explorer, leave the dataset path empty, put a higher epoch count on "total epochs" and click on train. You can find similar info on the guides.

[trail wraith]

What Python package do I install? its teling me to download 3.13.2 but when I tried using that it didnt work because it wouldn't find the torch location with it, even though I downloaded torch. Am I downloading the wrong version?

Does anybody have a tutorial I can follow? the Youtube tutorials are all outdated

@low shard Here

[low shard]

alr we talkied in #🧬│ai-chat

[trail wraith]

Yes

[low shard]

first of all, elaborate:

• ur pc gpu
• what u want to do
• what OS do you use

[trail wraith]

I would like to install Python using my GPU should I use their newest version that is compatible with my laptop, or is there a specific older version I need to use?

[low shard]

Are you trying to do anything RVC-related? or just want to install python?

[trail wraith]

Its RVC related

[low shard]

then, elaborate #✨│ai-help message first

[latent rune]

windows 10, nvidia

what do

just voice changer

[trail wraith]

I have tested RVC without python or pytorch, but when I looked at another tutorial it said I needed python and pytorch so I would just like to know which version of Python I need? I am currently running Cuda 11.8 but they want me to install Python 3.132

[trail wraith]

So I dont need to install python or pytorch to use it on discord

?

[low shard]

nvidia is a company that makes a lot of things, which nvidia gpu?

[low shard]

realtime for calls?

[low shard]

you don't need python

ignore everything you get off old youtube tuts

RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models

Wokada = uses RVC for realtime inference

[trail wraith]

So I can just download the voice changer and im ready to go

[latent rune]

I have a 2070super, looking to get rvc for calls

[low shard]

Do yall actually need Wokada or RVC?

[low shard]

then you don't need RVC, you need wokada

go to #🔍│help-w-okada

[low shard]

so, you want wokada, the realtime voice changer for calls?

[trail wraith]

Yes

[low shard]

RVC is not a voice changer for calls

[low shard]

^

[low shard]

tell your pc gpu and OS in #🔍│help-w-okada

[heady plover]

Hi, I just got to this, are there docs or something to tell me which model does what, which is better for an xyz purpose...

[glacial pollen]

Yes. There is huge as hell doc in fact

https://docs.google.com/document/d/17fjNvJzj8ZGSer7c7OFe_CNfUKbAxEh_OBv94ZdRG5c/edit?tab=t.0

[tawny shuttle]

tensorboard loads fine through collab, but the rvc link leads me to an issue,
ERR_NGROK_8012
Traffic was successfully tunneled to the ngrok agent, but the agent failed to establish a connection to the upstream web service at http://localhost:xxxx. The error encountered was:

dial tcp xxx.x.x.x.xxxx connect: connection refused

the local url also does not appear in the notebook either:
RVC URL:
Tensorboard URL:
File URL:
The tensorboard extension is already loaded. To reload it, use:
%reload_ext tensorboard
Reusing TensorBoard on port 8077 (pid 5487), started 0:09:36 ago. (Use '!kill 5487' to kill it.)
Traceback (most recent call last):
File "/content/training/runmain.py", line 3, in <module>
from dotenv import load_dotenv
ModuleNotFoundError: No module named 'dotenv'

very new to rvc so please bear with me

[gentle hollow]

Can someone model this for me

[glacial pollen]

#1159289738314919936 is what you should use

or visit #1191429836321849435

[fresh cairn]

has anyone gotten the RVC realtime to work on arch linux with AMD gpus. i keep getting errors

[knotty moth]

go to #🔍│help-w-okada and check the pinned guide to get w-okada fork with better optimization

[hallow thistle]

To request someone to do a voice model for you, you can create one in #1159289738314919936 or train it by yourself.

[hallow thistle]

For W-Okada, go to #🔍│help-w-okada. If you mean by the realtime mode in an RVC program, that thing is too old.

[olive hill]

What should be dataset for Applio? is there a specific length?

[hallow thistle]

There's no specific length you should use to train a voice model in Applio. Although a good quality audio can be used to train a voice model to achieve a good quality, 30 - 60 minutes audio should good enough.

[olive hill]

Should we chop the audio like earlier? where we used to make chunks of 10 sec audio, i mean 30 mins audio splitted in 10 sec each, or a 30 min single audio file?

[hallow thistle]

I think Applio should have a feature to autochop audio for train. You can read some more about Applio there. https://docs.applio.org/applio

[knotty moth]

5 minutes as bare minimum, the more the better, but more than 1-2 hours usually gives less noticable improvement given the same quality consistency

[olive hill]

can we train in Steps? like i have a bad GPU ( NVIDIA GeForce GTX 1050 Ti (4 GB)) , can i train daily 1 hour-2 hour and continue again?

[hallow thistle]

NVIDIA GeForce GTX 1050 can be used for AI inference. Although AI training is possible for this specific GPU, it would be real slow.

You'd have to open your PC overtime to finish a single training.

[knotty moth]

6 GB is bare minimum and 8+ is recommended

GTX cards are also not recommended due to lack of tensor cores for optimization

[low shard]

elaborate:

• ur pc gpu
• what u want to do
• what guide are u using
• what are u doing step by step

[olive hill]

Then in cloud which is recommeded? i want to do 500 Epochs

15 mins of dataset

[low shard]

epochs is just a unit of measurement of the traijing cycle

more or less don’t mean more quality

you need to monitor the tensorboard

[olive hill]

Which is easy to use? i have used collab in earlier days of RVC

it used to get disconected after certain Epochs or load

[low shard]

that’s bc google colab has a random daily gpu time

which can be max 4 hours

kaggle gives 30 hours weekly for better gpus

but it needs phone number and its harder to use

I would suggest applio kaggle

[olive hill]

Now for 200-300 Epochs with data set of 15 mins i belive it should take 2-3 hours right? which cloud should i use so that it doesnt get disconnected and is easy

[olive hill]

U sure?

[hallow thistle]

Yes.

[olive hill]

Whats this error?

[knotty moth]

invalid authtoken, double check at https://dashboard.ngrok.com/get-started/your-authtoken

then paste it at this highlighted one

[olive hill]

how to upload data set?

[olive hill]

Thanks worked

[olive hill]

Please @knotty moth

[vocal tiger]

https://gofile.io/d/zMigAR

anyone know why

the voloume is weird like that

and how i can fix it

also sent a link to a file cause i didnt have perms to post a vid

[hallow thistle]

Which RVC or W-Okada program are you trying to run?

[unreal granite]

Hello everyone, I'm new in this world of to discovering new things and that attract me a lot, I learned well or bad to start the program, but when I feel I don't know if the setting is correct, I tried it as a joke with friends on discord and they tell me that it is not so clear, I don't know if you eat words, as well as the delay. My setup is a R5 5600X and a RTX 4070s

[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]

Are you trying to install RVC or W-Okada or something?

[vocal tiger]

program as in the application itself or the specific voice im using

[hallow thistle]

For W-Okada the realtime voice changer, go to #🔍│help-w-okada. For RVC the voice converter/changer, #✨│ai-help here. Anything else, let me know.

No, I ain't clicking on that link.

[vocal tiger]

thats fine, can i have perms to post a video then?

or at least an audio file

[hallow thistle]

But I'm not a moderator bruh. At least just tell the name of the program.

[vocal tiger]

okay i figured out its rvc, like i thought

[hallow thistle]

Now say the full name of it. "Applio, mainline RVC or Tiger RVC GUI" for example.

[olive hill]

is this normal? or am i doing something wrong? because its being trained super fast and model i check for 100 Epochs its not sounding great

[tame mica]

batch size 15

[fresh cairn]

Oh it changed? I used it on windows awhile ago(a year ish). I kinda out of loop. I'll look it up since I like using it on discord

[knotty moth]

you didn't seem to follow the training guide

• stick on batch size 4 or 8 for most cases
• use the "simple" slicing method
• 5 minutes of dataset is bare minimum to yield good enough results and it shouldn't be only 1 step per epoch

[fresh cairn]

Actually yea, this is the one I am using I believe. If I am understanding the fork anyway, this is it but I can't get it to work when I run the command.

[thin edge]

are these 2 things normal? (using rvc mainline collab)

[simple ore]

you did not slice the audio or your slices are >10s

[simple ore]

no they are obviously not, the requirements have not been installed

[thin edge]

So I need to install py 3.10.12 on my pc?

I've tried installing py 3 10 12 on my pc only because it doesn't use the installer I'm not sure how to open it, there is a folder but I don't see the application in the install folder.

[simple ore]

what kind of linux is that?

[thin edge]

im using win 11

[simple ore]

why is it using unix paths?

you said you're using (using rvc mainline collab)

[thin edge]

yes

[simple ore]

#📰│dev-updates message

[thin edge]

i need to WAIT.......?

[simple ore]

or use the colab that has been fixed

or you can use a local install if you have a decent gpu

[thin edge]

my issue with kaggle, the “datasets” folder in the Mainline web GUI (Kaggle) is locked, I can't put my datasets into the folder

[low quail]

anyone around that knows how to use the current cloud method to make covers?
I've tried like 6 times, yet the song always has some kinks to iron out, can never figure it out for the life of me

sometimes the start of the song, the middle, and at the very end some artifacting happens

or it happens at the start but is absent at the end

or sometimes, it's fine throughout until some straining lines approach

[knotty moth]

#📰│dev-updates message or make creations in weights

[low quail]

I think that's the one I used, even had it segment the audio to avoid it but it happens still
felt out of my element with this so figured I'd ask here

[quaint nacelle]

Hey

[thin edge]

turns out i can use appolio

[quaint nacelle]

Yesss appolio is great

Or the collab I sue is the Rvc ai cover maker works the best for me . Or locally if u have a strong gpu

[thin edge]

nah, i want to train voice

how to use custom pretrain (other pretain #1235952130855010365 ) on appolio ?

[surreal venture]

hii can anyone teach me how to use the latest UVR5 to make ai cover? i can't find tutorial right now 😭 plz

last time i made ai cover was last year and idk whats happening rn but i can't use the old rvc anymore, i just need the simplist way to make ai cover plz

[thin edge]

https://www.weights.com/create this should be the "easiest"

[surreal venture]

tysm!! does it have the same quality as the rvc and other google collab ?

[hallow thistle]

That site provides pretty much the basic stem separation: vocals and an instrumental.

There's a working UVR5 Colab notebook link that being said by a mod somewhere here.

[thin edge]

bro wanted the easiest/simplest one, so I gave it to him.

[hallow thistle]

[surreal venture]

i saw the link but im dumb i didn't see the tutorial, can i work it online with Mac Os?

[hallow thistle]

Google Colab is a website. It will sure work on Safari.

[keen crescent]

-colab

[karmic olive]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• Hina's Mod AICoverGen WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Hina's Modified Original W-Okada's Realtime Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab
• 🆕 Music Source Separation Training (Inference), by Jarredou & Makidanye Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

More commands: -audio, -hf, -local, -kaggle, -realtime, -rvc, /help

[surreal venture]

yeah but i just don't know how to use it ....

just click them step by step or...?is there a tutorial link?

i used to follow youtube video but they are now outdated

[knotty cobalt]

I'm running RVC on my macbook m4 2024. I got it somewhat working, and the WebUI is popping up. But, whenever I try to do anything, my tasks load in the queue indefinitely without completion. Is this a common bug? Anyone have any info on how to fix it?

[surreal venture]

can anyone tell me how to install voice model at this page?

[thin edge]

I don't know, I haven't even used it yet, that's why I'm recommending Weight.

I don't know if I should feel annoyed or happy when the voice model I made in weight and the one I did in collab are barely different.

[carmine hearth]

UVR5 UI does not use a voice model
RVC or Applio might be more in line with what you're looking for.

[low shard]

you don't

that's just for cleaning vocals

what's ur pc gpu

[surreal venture]

ahhh i see

then how can i make ai cover ?

[surreal venture]

yess😭 the quality weight made is kinda......

[low shard]

what's ur pc gpu

[low shard]

@oblique heart

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 by Vidal (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

Here's all cloud options

if you need any help or ask any questions, you can here

[oblique heart]

ty

[low shard]

yw and lmk

[oblique heart]

um it says i can turn on headless mode optionally to run the gpu on all sessions

what does that mean?

im a little confused

[knotty moth]

to keep the session running even when the browser tab is closed

[oblique heart]

oh

the ngrok link leads to an error

error 406

the link to sign up

nvm i got it to work

wait what is the pretrain

is it the AI model i want it to sound like?

[low shard]

it's basically a base for your actual training

without it, everyone would need more than 40 hours for training

but that works as a base for you to train your own model with much less than that much dataset

[idle plaza]

If I stream on twitch (obs open 3500 bitrate) + using w-okada RVC + game running --> will my PC "melt" as in lower the life span of my hardware significantly? I have a 4070 Ti Super OC Nvidia // Ryzen 9 3900x // 32gb ram // lots of fans // and a pretty good cooling system that maintains low temps

[analog obsidian]

rvc is not w-okada
use the w-okada channel for stuff related to it #🔍│help-w-okada

[analog obsidian]

as long your temps are fine, your pc will not degrade as much

what kills components are temps

and unstable voltages

if your voltages and temps are fine, you'll be fine

[idle plaza]

oh woopsie my bad, and thanks for the response

[spiral granite]

I made this little sht!

[thorn abyss]

yo, does anyone knows why training not working on applio colab?

[fair magnet]

How do I make my own voice model with voice clips?

I've searched for methods but they're all outdated

[ruby zinc]

Can an NPU work for rvc

[simple ore]

need to update numpy to 1.26.4

[thorn abyss]

how do i update it on colab?

[sinful ridge]

Okay so I've created an app... for the time being it's using edge-tts to generate the speech output, not too slow, but obviously there's not a lot of control over the voices (though between rate and pitch you can do a lot more than you think).

That being said, it's hard to beat the 2-3 seconds or so it takes to get the audio (depending on length).
I want to do it with cloned voices, trained or otherwise, but it has to be fast, either on CPU or a 3060 12GB.

What are my best options?

[simple ore]

make a new code cell

[plucky jay]

i need help

i tried to do the cover thing

i cant send screenshots

but the output part is empty

im trying to make a cover in rvc colab

applio

and after i hit convert it says file inferred successfully but export audio is empty

@odd shale

[low shard]

!give-media-perms 1h @plucky jay

elaborate:

• your pc gpu
• what guide are u using

also don't ping random helpers

@dusty mortar #🔍│help-w-okada message

How to (unofficially) use Applio for RTX 50 serie cards

Follow to download it as said it in https://docs.aihub.gg/rvc/local/applio/

After you extracted the precompiled, go to the path in Windows explorer, write "CMD" and press enter, then in CMD write env\python -m pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

If you get any already satisfied requirement issue, run env\python -m pip uninstall torch torchvision torchaudio then the command said above

[ruby zinc]

is a 1660 super good enough for training in a decent time? (≤2 hours)

[knotty moth]

being faster than the audio duration or half of it is sufficient imo

most Nvidia gpus should have fulfilled it

[sinful ridge]

and I should be using what to do that? I have voice-changer open now to toy with, but that's assuming I'd be piping the resulting edge-tts audio into it and then to my speakers I guess.

hmm... this is working somewhat..

[hollow shale]

Hello, I'm new in Weights, how can I have an accurate Ai cover? My Ai covers are mostly dry and they sometimes can't withstand high notes

[hollow shale]

They sometimes sing like they didn't drink water

[fast scarab]

Hi everyone, do these pre-trained models work well for French, or are they mainly optimized for English?

[deft hawk]

help, I downloaded a voice changer, configured it, I launch it - I can't hear the neural network RVC voices don't work, only BEATRICE works video card RTX 3050 processor AMD RYZEN 5 6600h vol 0

[plucky jay]

i was using the applio colab i switched to the other one and it worked thank you

[knotty moth]

doing it "realtime" is not a good idea

[knotty moth]

pls go to #🔍│help-w-okada and read the pinned guide there

[main tiger]

um so whenever i uploaded a new ai voice i need to press download embredder before actally dowloading it but when i download the embredder thing it just says wait a momment and ive been waiting for like almost 10 mins ngl

did yall have to do the same thing

or am i buggin bruh

nvm

[mild surge]

Hello guys, there is a model that i use which sounds good, (psycho2go by dan), but it's pronunciation in arabic isn't good, i downloaded arabic dataset,

Is there a way to make this model with the same sound good in arabic ?

[knotty moth]

in talking or singing vocals? depending on how well the input audio articulates, and it may struggle more on the latter

[mild surge]

In talking, but it should be able to handle changing in tune without going all robotic😅

[mild surge]

I tried adding the pth file of psycho2go model in the load model in train tab rvc but it gives me error

[vapid gale]

guys ive got a question i'm beginner with all these ai stuff and i wanted to ask something ive got the vocals of a song that i want to convert it to ai covers but i dont know what to use can someone help me?

[knotty moth]

you aren't supposed to "load" pth files in training

[mild surge]

But i need to get the same voice just need it to learn the pronunciation

[knotty moth]

you don't need the existing model, you need the dataset

if the model was made by someone else, you're cooked

[mild surge]

I guess I'm cooked 😭😭😂

Chatgpt told me I'm not cooked and i can fine tune this model even if i had an arabic dataset for other speaker

[knotty moth]

alternatively, you can try using the output audio inferred with the model as a dataset

though I'm not sure if it's not ideal

[knotty moth]

you prob mean an arabic pretrain, but unfortunately I don't think it exists yet

and to make an arabic pretrain you'd need massive amount of dataset (100h+)

[mild surge]

Then back to being cooked😂

[mild surge]

With the same speaker?

[knotty moth]

several speakers, perhaps around 20-100 speakers (30m-1h each)

[mild surge]

I give up😂

[gritty merlin]

https://prnt.sc/bYroEhdoY4h9

[vapid gale]

i downloaded a voice model but unlike the other models, the one i have downloaded has model.pth and a metadata.json in it what do i do with the metadata.json

[low shard]

nothing, it's just the metadata of the model on weights.com

[vapid gale]

so all i need to do is insert the model.pth in the place that i would normally put the pth files?

[low shard]

!give-media-perms 1h @gritty merlin

elaborate:

• ur pc gpu
• what guide are u following

[knotty moth]

use another model's index file as placeholder and set index rate to 0

[low shard]

you can even delete the model metadaata, all you need is the pth and index if it has one

also u might wanna rename the model

since all models on weights are renamed as "model"

[vapid gale]

i have to use a index file?

or it wont work?

[low shard]

It's optional

[vapid gale]

alr ty

[knotty moth]

it should work in Applio I suppose

[low shard]

What you absolutely need is a pth

pth is basically the model containing the voice
index shortly contains the accent it has been trained on

[thin anchor]

wsp