✨│ai-help

[lucid creek]

ita just 2 h

[stiff goblet]

these are my charts , training for 3 hours

[orchid flint]

Question, is 2 and a half hours of data for a model too much? I tried making a Richtofen model and it sounds like shit. Granted I'm not German but its very rough. btw I'm using it with W-Okada so im not sure if that is a factor or not.

[stiff goblet]

It's way too long

[orchid flint]

Ok, just stick to an hour then?

[stiff goblet]

I don't know who Richtofen is, but I think it is overkill to use more than 30 minutes of dataset for a voice model.

[orchid flint]

I use an hour and a half sometimes and it sounds good, first time I tried with a 2 hour dataset

[crude flame]

anything after an hour you just get diminishing returns

[orchid flint]

Ok I see, makes sense

[stiff goblet]

I have no idea actually, never trained that long dataset tbh. You should try shorter, good quality datasets with lower batch sizes imo

[orchid flint]

yeah, it is good quality, I the dataset is ripped right from black ops 3

[stiff goblet]

If the character does not have a versatile voice and speaks in a monotonous range, it is pointless to keep the dataset long.

[orchid flint]

Yeah but this character is very expressive.

[simple ore]

a lot of dips on normal g/total indicates a lot of silence in the dataset

fm going up very fast, so most likely this is gonna be a trash model

[stiff goblet]

He/She was using Applio i guess? Maybe silence training files effecting this situation. Or not?

[analog obsidian]

too much silence in the dataset bad
mute files are fine as long the dataset has the silence truncated

[simple ore]

it is too much to see two silent files showing up in the last epoch so often

[vale heart]

What AI, preferably local, will allow to convert subtitles with timestamps into AI voice?

[stiff goblet]

i don't know why but my feature matching is always going up no matter which model i train

[orchid flint]

Bro, RVC hates me 💀

[analog obsidian]

it should be slowly going up, but not like in your case, usually fm does that when the dataset is super small

[stiff goblet]

My dataset is 13 minutes. And batch size is 4

[analog obsidian]

yep thats a small dataset, so makes sense why fm is raising like that

[stiff goblet]

Thank you. I think a clean vocal performance of 13 minutes of training will be enough. There is nothing I can do for "fm" lol. Avg/g/total will be the chart I will consider

[lucid creek]

strange the dataset dont have silence a lot

[lucid creek]

my is 10

[lucid creek]

if there how i can remove it 🙂

its in overtrain right

[orchid flint]

Eh, kind of hard to tell. You could always train for a bit longer

[viral mason]

idk what bro is using so I cannot help

[lucid creek]

like what

[viral mason]

what are you using to train? it looks different to what I use

[lucid creek]

apillo with of pretain and 10 min dataset

[vale heart]

guys, anyone text to speech ai? with timecodes 00:00:18,000 --> 00:00:21,000 please

[viral mason]

Applio on what? Collab or kaggle

[lucid creek]

kaggle ofc

[lucid creek]

how its looks like

[simple ore]

1s sample, really? 🙂

[lucid creek]

man my pc is trash i took 1 min for this

[turbid ridge]

Quick question even at 250 my rvc models are all quiet

[lucid creek]

yes is too trash

[gilded flicker]

I am running a local ai on a 3060 ti and a ryzen 3 series cpu what can I rent to render faster someone recommend me vps at a good price na servers I wanna be able to open windows on vps and run my own programs

[junior gull]

Allegedly, i cant really confirm just what i was told by someone else, the errors that were resulting from this was like a domino effect that also led to my virtual environment loading incorrectly and processing quietly defaulted to cpu based without failing or warning me. When I ran it all correctly and with cuda support it was blazing fast lol.

[simple ore]

quiet has nothing to do with epochs

if your dataset is too low volume, then the results of inference will be quiet

[simple ore]

1000 epoch is 5 minutes is "you're training on an empty dataset"

there's no way in hell to train a meanigfully good dataset this fast even on the most advanced AI accelerator

[vestal helm]

I js farted

[visual compass]

Does anyone have tips on speeding up the process of building a dataset for a voice model? Specifically sifting through an entire stream or episode's worth of audio to find and isolate where the voice you're trying to clone is located? I figure there's a better way than manually going through it and comparing it to a transcript to easily locate them.

[next vortex]

I have some problem when I try to use RVC on Colab

[hallow thistle]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

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

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

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

__**BE PRODUCTIVE.**__ 🤝

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

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

[hallow thistle]

Which RVC Colab notebook?

[next vortex]

W-Okada one

But I cant send screenshot here

[hallow thistle]

RVC and W-Okada are two different programs.

What is your PC GPU?

[glacial pollen]

yeah, ig this will never end
this infection will just keep on spreading and spreading

[next vortex]

I dont have

[glacial pollen]

rip

[hallow thistle]

The original Colab W-Okada is broken. Using Detris' fork W-Okada on Colab with free plan can get you terminated.

[glacial pollen]

I mean... It can be run on cpu but

the delay.. oh boi

[next vortex]

Yeah I already tried

[hallow thistle]

Use Kaggle instead, but you'll have to register with your phone number on this one. https://www.kaggle.com/code/suneku/voice-changer-public

[next vortex]

i dont know how can I run it without GPU

[hallow thistle]

Also, the guide for fork W-Okada. https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork

[glacial pollen]

I mean, you can just simply do it:

but whether you should ( due to delay ofc and stutters ) that's another story

better if you follow what Namari says, less hassle for nothing beneficial really.

[next vortex]

Do u have some voice changer else

I mean, ofc, free one

[hallow thistle]

W-Okada is a free and open source program. Detris' fork W-Okada is the one I use.

[next vortex]

What do u mean "can get me terminated"?

[simple ore]

at best you google account will be banned from colab

at worst bye-bye google account

[bleak torrent]

hallo

Is the colab/kaggle version of w-okada no longer working?

[latent kettle]

how to fix this ??

[latent kettle]

Will it cause any impact or not ? Should I keep it as it is or there is a need to fix it ?

[hot dome]

If there is some people who need someone to make them thumbnail i can help im a thumbnail youtube maker , mp

[simple ore]

you resumed the training, but since you had some progress after the previous save logged, the jump back got logged twice

does not really matter

[latent kettle]

Actually due to power cut my pc went off

That cause the issue

[simple ore]

same thing

[latent kettle]

After this the data I see. Is accurate ?

[simple ore]

it is just the log got double entries for 18.5k (saved epoch) - 19k (when you lost power)

[brittle wing]

can someone explain why there is an paywal on voice.ai and how to remove it and its not on browser

[simple ore]

you see, when someone decides to make money off the fools who just dont know any better, they add a thing called paywall to a free software

and then the dangle 'you can do it for free*" (* not really)

[brittle wing]

so it cant be removed

[simple ore]

for the money

[brittle wing]

ok

[simple ore]

if you got a decent gpu you can run the same thing locally on your PC, or on google colab/kaggle for free

[knotty moth]

you still want to pay for that garbage?

[lucid creek]

i mange to delete all the silence from dataset this time its so good now ? { i use appilo on kaggle}

[vague edge]

when i try to delete a voice model i get this pop up and i have to reload page

[knotty moth]

close the voice changer then delete the folder in model_dir

[vague edge]

alr ill try that now

[viral mason]

What is that

[uncut rampart]

**in the colab version of Aicovergen i get this for the Run webUI command
**
Timer: 00:02:14Traceback (most recent call last):
File "/content/HRVC/HRVC/src/webui.py", line 10, in <module>
from main import song_cover_pipeline
File "/content/HRVC/HRVC/src/main.py", line 22, in <module>
from rvc import Config, load_hubert, get_vc, rvc_infer
File "/content/HRVC/HRVC/src/rvc.py", line 5, in <module>
from fairseq import checkpoint_utils
File "/usr/local/lib/python3.11/dist-packages/fairseq/init.py", line 20, in <module>
from fairseq.distributed import utils as distributed_utils
File "/usr/local/lib/python3.11/dist-packages/fairseq/distributed/init.py", line 7, in <module>
from .fully_sharded_data_parallel import (
File "/usr/local/lib/python3.11/dist-packages/fairseq/distributed/fully_sharded_data_parallel.py", line 10, in <module>
from fairseq.dataclass.configs import DistributedTrainingConfig
File "/usr/local/lib/python3.11/dist-packages/fairseq/dataclass/init.py", line 6, in <module>
from .configs import FairseqDataclass
File "/usr/local/lib/python3.11/dist-packages/fairseq/dataclass/configs.py", line 1104, in <module>
@dataclass
^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 1232, in dataclass
return wrap(cls)
^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 1222, in wrap
return _process_class(cls, init, repr, eq, order, unsafe_hash,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 958, in _process_class
cls_fields.append(_get_field(cls, name, type, kw_only))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 815, in _get_field
raise ValueError(f'mutable default {type(f.default)} for field '
ValueError: mutable default <class 'fairseq.dataclass.configs.CommonConfig'> for field common is not allowed: use default_factory
Timer: 00:02:16

can someone help?

[viscid moss]

Are u using the original AICoverGen Colab?

AICoverGen it's outdated

[uncut rampart]

yes

[viscid moss]

makes sense, dat colab isn't working

[uncut rampart]

what can i do now to generate ai covers?

@viscid moss

[viscid moss]

U have 2 ways:
If u want to do everything automatically like AICoverGen, use RVCAICoverMaker.
If u have some time and u can do it manually using UVR5 + Applio

[uncut rampart]

can you provide any link for RVCAICoverMaker

[viscid moss]

https://github.com/Eddycrack864/RVC-AI-Cover-Maker-UI

[uncut rampart]

thank you

[uncut rampart]

can you also provide me with a article or a tutorial for creating own ai cover voice models?

[viscid moss]

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[vague edge]

whats the best chunk and extra setting for rtx 4060 i tried looking on website but couldnt find 4060 settings

[arctic gate]

what do you guys use for voice chagners?

[warm steppe]

guys

[vague edge]

?

[vague edge]

W Okada

https://docs.aihub.gg/

[warm steppe]

how can i get what voice changer i want through weight site?

i have w okada

but i wanna have peter griffim like my voiuce chamher

ansd i dont find it in voice models

[quasi condor]

are u typing fast or are u nervous?

[warm steppe]

fast

mb

[vague edge]

https://www.weights.com/home/models

[warm steppe]

yeah

[vague edge]

you press the three dots

[warm steppe]

but how can i make it into the w oakada

[vague edge]

and press download

[warm steppe]

oakda

oooohhhhh

[quasi condor]

u can also find peter griffin models here https://discord.com/channels/1159260121998827560/1175430844685484042

[vague edge]

true

[warm steppe]

oh ty

wait

what 3 dots?

[quasi condor]

some of the models of peter griffin on weights is shit cuz they just make the model without cleaning it

[warm steppe]

oh ok

[lucid creek]

anyone anybudy 🫠

[quasi condor]

that's good as long it doesn't go down or up sometimes the tensorboard gets a little confusing

but u gotta test the model if its good or nah

[lucid creek]

iknow

[leaden kestrel]

Any turkish girl models?

[lucid creek]

nah

[lucid creek]

how 😉

[arctic gate]

its this server thing right?

[mystic tangle]

how do i make the perfect dataset?

[latent kettle]

https://docs.aihub.gg/rvc/resources/dataset-isolation/

Here you go

[clever burrow]

you guys use the rvc gui thing for ai covers right?

[lucid creek]

yes

[kindred ledge]

help me choose gpu, currenly i use rtx 3050 mobile for rvc, voice good. But now i want to build and change gpu to rx 6700 xt. Will the sound be different?

[simple ore]

sound wont be different

6700 is not a big upgrade

[glacial pollen]

no

rvc-gui is heavily outdated and nobody should use it

[lucid creek]

its good ?

[wheat tapir]

what do you use/recommend?

[glacial pollen]

well, applio / my fork of it

or either original rvc but no point in that

whereas rvc-gui specifically, yeah, outdated ( last update it had was 2 years ago so that should give you an idea of how old it is

tl;dr. What should you choose?

If you want advanced features and are yourself rather advanced / want nightly features n stuff, my fork
https://github.com/codename0og/codename-rvc-fork-3

If you want simplicity and something that just works, og Applio
https://github.com/IAHispano/Applio

[golden walrus]

sire, i want to ask if WavLM has a good outcome :<

[glacial pollen]

wavlm is not ready

[wheat tapir]

thank you!

[pallid patio]

Hi! Good morning/afternoon/evening! Can someone help me create voice models? A simple explanation or, if possible, a collab that does it would be very helpful!

[odd shale]

Here, you can start by reading the docs (Check the first guide)

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[kindred ledge]

Will the sound break up?

[simple ore]

it does not really matter what device you're using to train, math is math

[fresh storm]

in my voice changer have cpu, gpu0, gpu1, gpu2, gpu3, what i choose ?

[knotty moth]

neither of them

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

(assuming you have AMD gpu)

[junior gull]

hmmm is there way to verify?
The applio ouput looked very similar to RVCv2 in size and the wav samples also sounded very similar. Both had pitch training too. I thought it was a little weird how stupid fast it was going....but it appeared all the training files were there.

[balmy girder]

im having trouble with training a model

it keeps saying "list index is out of range"

how do i fix it?

[simple ore]

does it sound like your voice model? how many steps per epoch are shown in the log?

[somber cobalt]

do i need to have amd igpu for it?

[knotty moth]

the radeon RX one, not igpu (aka. radeon graphics)

[somber cobalt]

rtx won't work?

[knotty moth]

it should be the Nvidia version https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#download-nvidia-on-windows

[somber cobalt]

but that version is of 7 dec 24

[knotty moth]

it's the latest one

[somber cobalt]

ok

[knotty moth]

for RTX 50-series you'd need this one https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#download-nvidia-rtx-5000-series-on-windows

[somber cobalt]

no i have 30 series

do i need igpu for it?

my cpu doesn't have one

[knotty moth]

it doesn't matter as you won't use it

[somber cobalt]

so its basically the same as the old voice changer?

[knotty moth]

dec 2024 is still not too old, just before 50-series launches

[somber cobalt]

no i was talking about the okada voice changer

i used it in the past

[somber cobalt]

i am downloading it from releases

not from the link in the guide

is that ok?

[knotty moth]

it should be the same I suppose

[somber cobalt]

in releases it has cuda part 1 and part 2

[knotty moth]

all parts are needed to extract

[somber cobalt]

ok

[uneven harness]

Im using RVC AI Cover Maker and for some reason it wont convert. ill upload the music via drop, select the model, click convert and it says error.

this is the precompiled version

[uneven harness]

nvm

fixed it

:3

[pastel oak]

@fallen wing whats your gpu

[fallen wing]

Nvidia 4050

Which do i download i may be stupid

[pastel oak]

Then download first nvidia

The 2nd says rtx 5000

You have 4000

[fallen wing]

After downloading how do you set it up i havent used RVC for like 2 months or 3

[hallow thistle]

Download NVIDIA on Windows. Your laptop has NVIDIA GeForce RTX 4050 Laptop.

!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.

[fallen wing]

After downloading i assume i extract it?

[pastel oak]

Read audio setup

[hallow thistle]

If the zip finished downloading, use 7zip or WinRAR to extract it to a folder.

No need to ask every step on this one.

[pastel oak]

Ye you asked questions that get answered by scrolling for 1 second on the guide

If anythings still unclear ask

[fallen wing]

My ass be anxious cause i accidently killed my laptop one time from something similar to this not rvc tho

[pastel oak]

Theres nothing risk killing your laptop on the guide, else itd mention risky steps

[fallen wing]

I read but traumatized fr

[low shard]

if there's any issue on the program's end, you could tell @viscid moss btw

[low shard]

what did u possible read last time 😭

[fallen wing]

It wasnt rvc

Something else fr

[copper terrace]

I got this and im confused

idk what to do

[latent kettle]

What you want to do

[hallow thistle]

What kind of fork RVC WebUI is this? I use Applio the RVC fork.

[copper terrace]

I want to use real time voice changer

[copper terrace]

idk I just found link and downloaded 😅

and sorry for the late late replies

[simple ore]

obviously some ancient shite

older than dinosaurs in AI terms

[copper terrace]

xD

[copper terrace]

can u give me the latest version, if you have it

or a link to a guide

[simple ore]

what's your GPU?

[copper terrace]

1650 (laptop)

[simple ore]

with 4GB VRAM at most you'll be able to replace voice in small files

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

[copper terrace]

alright

[copper terrace]

thank you

[tender socket]

is there a guide here on how to make an AI voice model for RVC

[analog obsidian]

just in case, he sent you the actual rvc, which is not for realtime

[copper terrace]

How do I use for realtime then

I need to record voice then covert in rvc?

[analog obsidian]

oh

so u wanna do local conversion

[copper terrace]

I'm confused

I'm an actually ai noob lol

[analog obsidian]

you want to change the voice of a recording? use applio

you want to change your voice in realtime to use it on disc/games? use deiteris w-okada

[simple ore]

realtime in games on 1650... i'm ded

[analog obsidian]

[copper terrace]

Going afk for a bit*

[analog obsidian]

yea he cant do that with a 1650 laptop gpu

[simple ore]

i mean.. if you play competitive Solitaire

[copper terrace]

yeah I wanna do that

[copper terrace]

yeah its not a good gpu

[copper terrace]

can u give link for that

I'm try both and see

[analog obsidian]

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

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

[analog obsidian]

first link

you may not be able to use it while playing games, your gpu is too weak

[gloomy mortar]

bc i dont have a really good computer, ive tried this voice changer on kaggle. the thing is, with this voice changer apon clicking any of the models i get this error "TypeError: Cannot read properties of null (reading 'modelSlots')
TypeError: Cannot read properties of null (reading 'modelSlots')
at i (https://85c5fd18-860e-42eb-8752-dc96044fe4e4.hrzn.run/index.js:2:1305771)
at Object.updateServerSettings (https://85c5fd18-860e-42eb-8752-dc96044fe4e4.hrzn.run/index.js:2:1306003)"

then while trying to intalise again i get this: "TypeError: (0 , Zz.removeDB) is not a function
TypeError: (0 , Zz.removeDB) is not a function
at https://85c5fd18-860e-42eb-8752-dc96044fe4e4.hrzn.run/index.js:2:3293292
at m (https://85c5fd18-860e-42eb-8752-dc96044fe4e4.hrzn.run/index.js:2:1536451)
at Generator.<anonymous> (https://85c5fd18-860e-42eb-8752-dc96044fe4e4.hrzn.run/index.js:2:1537797)
at Generator.next (https://85c5fd18-860e-42eb-8752-dc96044fe4e4.hrzn.run/index.js:2:1536880)
at e (https://85c5fd18-860e-42eb-8752-dc96044fe4e4.hrzn.run/index.js:2:1543652)
at s (https://85c5fd18-860e-42eb-8752-dc96044fe4e4.hrzn.run/index.js:2:1543855)" any fixes? im using hrzn

this is w-okawda's voice changer btw

and yes i do have the modded version

but i also wanna try this out

[low shard]

let's talk in https://discord.com/channels/1159260121998827560/1373937000750649404

[low shard]

yes, but what's your pc gpu first?

[potent bone]

should i use codename fork instead of applio?

[peak path]

hi
should i use .index file in my model zip file ?
i use only the final .pth and add it to my zip
Replay software can import only .pth file without error.
i want to know if i don't use .index, can effect in my final result?

[regal fog]

help, an error pops up
[Voice Changer] Web Server Launch Exception, DLL load failed while importing beatrice_internal_api: The Dynamic link library initialization program (DLL) failed.
[VCClient] wait web server...10 http://127.0.0.1:18888/
[VCClient] wait web server...20 http://127.0.0.1:18888/
[VCClient] wait web server...30 http://127.0.0.1:18888/

[simple ore]

only if you want to tweak 20 things at once and experiment

[potent bone]

do u think its worth it?

[simple ore]

have you ever trained a good voice model with old RVC/mangio/applio?

it is not noob friendly after all

[low shard]

please elaborate:

• ur pc gpu
• what u want to do
• the tut link

[magic hare]

I'm using RVC and stuff, as a voice changer for VRChat, whenever I load a model, I get this error: Error:'NoneType' object has no attribute 'host_api'

[low shard]

please elaborate:

• ur pc gpu
• the tut link

[magic hare]

The what link?

[low shard]

the tutorial link that you're following

[trail dirge]

do i have to use RVC GUI for files or can i just use wokada?

[simple ore]

if your wokada has input=file, output=file, then yes you can use it

or you can just use the cloud option on huggingface with illaria rvc

[trail dirge]

is there command to see the RVC GUI link thing like with - realtime?

[simple ore]

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[smoky linden]

RVC i think

[junior magnet]

Hi, is it normal for the audio to take so long to convert?

[junior magnet]

Is the collab working?

[turbid juniper]

does anyone have this kind of problem when merging models?

[smoky linden]

anyone now how to install rvc voice changer

[knotty moth]

try using Applio for merging models

[turbid juniper]

oh ok I've been using the older versions, do you have a link for that one?

wait I think I found it

thx

[low shard]

be sure to be using wokada deiteris fork

[low shard]

what’s ur pc gpu? What do u want to do?

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. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.

Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)

[low shard]

elaborate

what’s ur pc gpu? Which collab link? How much time did it take?

[turbid juniper]

I also use that but sometimes it's slow, so I run w okada locally if that happens

[low shard]

what? Wokada has 2 main versions

[low shard]

Wokada has 2 main versions:

• Original made by Wok
• Deiteris fork (modified version) made by Deiteris

each version has it's own updates

the latest deiteris fork has way better performance and quality than the latest original

wokada deiteris fork got more performance

be sure to not use original wokada

it is local for both

[low shard]

be sure to not use rvc gui

[mystic tangle]

what settings should i use in UVR vocal remover if im trying to make a dataset?

this is what i have currently

its supposed to be a talking model

[copper terrace]

how to I use gpu?

I thought I downloaded the GPU version but its using my CPU

[mystic tangle]

how do i trim audio in audacity efficiently? its hard to get the exact trim

[low shard]

what’s ur pc gpu? What do u want to do? What tutorial link are u following?

[mystic tangle]

is this the right place to ask my question

[copper terrace]

followed this

the first link

[low shard]

what’s ur pc gpu? What do u want to do?

oh u got a gtx 1650?

[low shard]

seems like wokada deiteris fork latest version, be sure ur nvidia drivers are up to date

[low shard]

yeah

[copper terrace]

yes

[copper terrace]

okay

[low shard]

lmk

[copper terrace]

working now, thx

[stiff goblet]

still learning ?

[sly topaz]

I can't send files to make ai covers

[icy ember]

Hey all, i have a problem with the voicechanger from w okada.
It runs and works perfectly, but i want to use it when i play a game eg. Red M, i have a huge delay, the voice laggs and i cant use it really.
If i tab out of the game and switch to any other programm such as discord or stuff, the laggs and delay are gone.
Can anyone help me ?

[quasi condor]

what's your G P U?

[broken arrow]

hey guys i have a 2080 ti and have tried using a voice changer to troll but quite a few number of times now people call it out very quickly. Is there any tips i can do to make it sound more natural? idk what much else to say should i provide a demo on how it sounds like?

[icy ember]

3060 Nivida (12GB)

[icy ember]

.

[quasi condor]

maybe its because ur playing red m?

are u playing it on high settings?

[icy ember]

I do have higher settings, but the porblem also appears if i play other games too.
I can have the game open and just switch to a diffrent window, then the voicechanger works fine, i can show you if you have time

[low shard]

Set every single game settings to the lowest, every single graphics and everything

Show a screenshot of ur wokada while rubbing the game

Be sure to close useless programs in the background

[icy ember]

On it!

[trail dirge]

is it bad or something?

[low shard]

It's extremely outdated and abandoned since 2023

I'm talking about the rvc GUI fork from t1g3r or smt like that

What's ur PC GPU

[low shard]

Any updates

[icy ember]

So it helped but its still not good enough

sadly i need to go now, i will msg you later

[trail dirge]

4070S

[knotty moth]

-gui

[patent trellis]

RVC Easy GUI

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

[knotty moth]

rvc gui by tiger18n is too old and doesn't have rmvpe pitch extractor as the current state of the art

[trail dirge]

ive seen here Apolio being mentioned

maybe thats newer

[low shard]

Send a screenshot later

[low shard]

Applio is a newer updated rvc fork

[low shard]

Nice

read https://docs.aihub.gg

[trail dirge]

alright will check it out later when I get home

yesterday i updated to wokada fork and its waaay better than the one i previously had

[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. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.

Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)

Rvc (like the applio fork) have completely different purposes than wokada

[junior magnet]

I was using this collab

On my phone

And the time it took was like, forever

When i clicked to convert, it didn't convert

[junior magnet]

I accessed the collab through this link, ai hub docs

[low shard]

You can't always use the same Gradio public url, you need to always get a new one by re-running the whole collab

[slim schooner]

is anyone using rvc-project? im not a coder or anything and im literally just a casual person trying to clone a voice. i was using grok.ai to help me out but it seems like the one it asked me to download just runs into problems over and over again, does anyone have one where there aren't any issues at all? grok.ai has been asking me to add debugging scripts all god damn day

would appreciate the help, if any!

copy and pasted from the other channel lol

[icy ember]

Cant msg you, you have to add me

[low shard]

no need to dm, this channel exists for help

[low shard]

lemme guess, you used a youtube tutorial?

this is an old original wokada version which has worse performance and quality

along with bad settings

plus windows users reported that vb audio cable might give issues

forget everything about the tutorial you followed and uninstall the programs you got from it

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

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

[low shard]

read the 1st link, wokada deiteris fork

the only updated guide is the written one

[low shard]

what's your pc gpu and what do you want to do?

[slim schooner]

i got an rtx 4070ti and im trying to clone a voice locally since its "better" i guess? and then use that cloned voice for the wakada voice changer. however i just saw you mention how the one used in yt tutorials is old and has performance issues. so im assuming i should scrap the old wakada and download the newer one?

[low shard]

wokada is for realtime

not for training

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. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.

Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)

inference = use models

inference realtime = use models in realtime

training = actually making the models

do you just need to use the models in realtime voice changing?

[slim schooner]

i know i know, i want to train one and then import it to wokada, i want to train one using rvc project, which was my main issue since the one i have installed has too many issues

[analog obsidian]

the original w-okada perfomance is very bad compared to the modified version
if you wanna train some models i'd recommend you to use the compiled version of applio, then for good results ensure your dataset is clean (remove room reverb and noise), use the right batch size and don't overtrain the model

[low shard]

also local is better because you won't be limited by the free cloud time, since you got a good pc gpu

[low shard]

https://docs.aihub.gg

check applio for training models

and get wokada deiteris fork for realtime inference

[slim schooner]

sounds good, thanks! ill let you guys know if i run into any issues (hopefully i dont)

[junior magnet]

I didn't understand

[low shard]

be sure to always re-run the notebook whenever you need it

don't just use the same gradio public url link

because it will always change whenever you re-run it, as the notebook won't stay running 24/7

[junior magnet]

Okay

Do you know another collab link so i can do a ai cover?

[low shard]

the problem isn't the colab link

the problem is you used the same url public link that was expired

be sure to always redo the same colab process everytime you need to use it

[icy ember]

I did, okay i will delete everything

[icy ember]

I got Cuda, VB - Audio, python

Delete everything ?

[simple ore]

dont use vb-audio for sure

use the virtual cable lite v4.70

[icy ember]

I currently am deleting everything as told 😮

[simple ore]

realtime voice changer does not need python, it comes with its own packaged in, and I think does not require cuda toolkit either

[low shard]

yeah just delete everything

#✨│ai-help message

[low shard]

he used an old ass youtube tutorial

[alpine eagle]

Como hago las piedras de evolucion?

[icy ember]

Final Step on installing...

[icy ember]

@low shard

Voice works. but it cuts at the end.. have not tried ingame yet

[abstract anvil]

Hey, quick question: Let's say you have quite a few models in Okada's directory and you want to organize them better. Can you move models into different numbered folders and have it reflected in the program's slots next time you open it? Or do you have to delete the folders like it says in the guide and start over?

[icy ember]

So ingame i cant rly use it, any setting i should use

[low shard]

this server is english only

[low shard]

set extra to 2.7, also try it in game and show again the program while in game at min settings

[icy ember]

Would you be able to help me live via voice ?

The game is on min settings

[low shard]

i can't really vc

[low shard]

uncheck sup1

put game graphics to minimum

be sure to also play with the pitch and other models, not all models are perfect

[icy ember]

They are on minimum, sup1 is unchecked

Is there a way to creata ticket for a meetup on live supp via vc.?

[low shard]

the settings are good, u could optionally use force fp32 mode on for worse delay but slightly better quality, and also server mode for less delay https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#reduce-more-delay

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

[low shard]

nope

tickets are meant for moderation

it depends on the staff if he can help via vc, which not really everyone can

[icy ember]

forget about the delay, the voice sounds robotic 95% of the time when ingame

[low shard]

then try another model

not all models are good

[icy ember]

Thx for the help, for today i give up

[low shard]

you just need to try other models really, i even gave you some suggested models

[icy ember]

I know, and i will, but for today i got no time left.
You already helped me alot, and it got better, i will get there

[low shard]

alr lmk

[sharp palm]

continue to get "Not enough data present in the training set. Perhaps you forgot to slice the audio files in preprocess?" 22 min dataset, saved as .wav, with the path to the folder in the Preprocess. I even tried uploaded my dataset. What's up with this error?

[low shard]

Perhaps you forgot to slice the audio files in preprocess?

that's kinda the fix

[sharp palm]

I mean, I sliced to 156 files. but it's just one file that is added to preprocess, right?

[sharp palm]

it's been a minute for me since the RVC V2 Disconnected colab days, so I'm trying to catch up lol

[low shard]

is the audio cutting feature to automatic in the preprocess settings

[sharp palm]

Yes

[sharp palm]

and in the tensorboard process when I click preprocess, it shows the 22 mins of data is uploaded. So it has to do something with the slices?

[low shard]

does it show only now or did it also show before that error?

[sharp palm]

It shows after i finish everything else and hit train

[low shard]

you need to make sure preprocess output shows X minutes were loaded, then you need to make sure extract features did extract N segments

[sharp palm]

thanks for your help - one last thing. With the sliced audio, back in the day you'd just upload the .zip with all files. Do the files need to be spaced out in the DAW (Ableton, Audacity, etc) by x amount of ms before export?

[simple ore]

depending on the colab, in some cases you did upload a zip with source audios, in others you uploaded an individual .wav file(s)

you dont need to split audio in audacity

[sharp palm]

then I don't know what the issue is. Any sample datasets on here I can evanluate & see where I'm getting hung up?

[sharp palm]

do I add the sliced audio to the folder of the path I input?

[simple ore]

i'll just give you this picture

if you're using applio colab with UI, you click [x] dataset creation, enter the set name, then upload individual file(s)

then preprocess

if you're using applio colab without UI, you put the files into a folder on your google drive and provide the folder name

everythig is done properly you'll see 'Preprocess completed on x seconds of the audio' and then extract features gonna have ~x/3s number of slices

so for one hour of audio there should be 1200-1300 slices

[sharp palm]

Yep this is where I went wrong. Much appreciated!

[slim schooner]

lol was having this exact same issue

its training now

[junior magnet]

I think the problem is not that the link has expired.

[junior magnet]

Look, I followed the step by step correctly, I just clicked on the link, there's no way it could have expired

I just recorded this video

[low shard]

Oh you're not using the t4 GPU

So it's running in CPU

Click the arrow top right

Change runtime type

Set t4

Then re run the colab

GPU is wayyyy faster than CPU

[junior magnet]

Okay, I'll try

Ty

[sharp palm]

I don't mean to be a pain in the ass. but look at attached. 545 slices from 23 min of audio (and they show in the log file, but then look at the bottom I'm still getting the error cc: @slim schooner

[junior magnet]

I changed it and still the audio doesn't convert

😭

[low shard]

Did you click save then re run the whole notebook?

[junior magnet]

Yes

[low shard]

When you change GPU and CPU the whole session restart so you have to do everything again

It doesn't seem in the background that you run the cell again

[junior magnet]

I redid everything

[low shard]

Are you sure you clicked save then did everything again? There should be at least some output in the background and also showing the cell running but nothing is running

[junior magnet]

It's because I took the screenshot when I changed, not after I did it all over again

[low shard]

Could you show how it's going right now?

[slim schooner]

not sure where it went wrong for you cus for me its working just fine it says its training and my gpu is getting worked lmao

[junior magnet]

Wait a minute, I'll record another video

[slim schooner]

id help if i could but im just following the pros here lol

[simple ore]

how many files are in f0, f0_voices, extracted folders?

[junior magnet]

Still not converting audio

[distant turtle]

-colab

[patent trellis]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[low shard]

First time I ever see someone using m4a

Try converting the file to mp3 https://convertio.co/m4a-mp3/

It would have been better if you showed the colab output after that it not being able to show the converted audio file, because here it shows the actual error

But from my guessing it might be because of the extension

Please try converting it, if it still doesn't work, do another video showing the colab output too at the end

[sharp palm]

545 in extracted, none showing in F0 and F0_voiced, 545 in sliced_audios

[simple ore]

try rmvpe extractor instead of crepe

[junior magnet]

Ok

Aaaa so hard😭

I'm terrible at programming

[low shard]

This isn't programming 😭

[slim schooner]

ok training stopped after overtraining was detected, is 321 epochs good for 15 mins of audio?

[low shard]

Programming is the process of creating the instructions, code, for the computer to follow, while this is just inference (utilization of the models) of the RVC program, and AI is a bit harder to use since this one is open source and not meant to be easy to use even though forks like applio make it kinda easier

[analog obsidian]

don't use overtraining detector, it's not accurate
321 epochs sounds a lot for just 15 minutes of data
monitor your model losses using applio's tensorboard, then stop the training when you notice the model starts to degrade

[slim schooner]

makes sense, so it would be safe to assume that its been overtrained and that overtraining results in quality suffering?

[analog obsidian]

if the model sounds ok in multiple epochs but suddenly becomes super robotic and bad after a certain point you can assume the model got overfitted

it's pretty easy to hear those problems

you'll hear the model having a static-electric type sound

[slim schooner]

i hear it, usually at the start or at the end when i say something

thanks man, ill see what i can do 👍

[sharp palm]

Thank you, rmvpe works and fixed the issue. Any idea why crepe would be causing those problems?

[slim schooner]

do you achieve this by supplying more data? like how many minutes of audio did you use to train this voice model? im guessing 15 isnt enough as it isnt anywhere as good as what you have shown here, even the overtrain sounds better than what i got lmaoo

[analog obsidian]

i used 42 minutes of data and a batch size of 8

[simple ore]

cant say without an error dump

[analog obsidian]

this is a 16 minute dataset

^ worse than the 42 minute model, but the quality is ok
so ideally you want your dataset to have good quality (remove noise and reverb) and diverse data

[slim schooner]

adding singing clips boosts perforance i assume rather than just talking clips correct?

[analog obsidian]

makes singing better yeah, these models weren't trained using singing tho, they were trained only using speech

i think it's fine to have both singing and speech in the dataset tho i have never tried it

[slim schooner]

ok and you can do that inside applio or do you need to source that outside of applio?

oh it looks like you deleted your messages lol but you mentioned noise can kill a model so i was wondering where you can do the cleanup and whatnot

say like maybe wind blowing or background music and stuff

[analog obsidian]

lmao i got confused, it wasn't noise that killed my model, i had forgotten that the model that sounds bad actually sounds that way because I was testing shit settings.

more of a me error

[slim schooner]

but noise still can kill a model or no?

[analog obsidian]

yup, the embedder is sensitive to noise

[slim schooner]

and what would i use to clean an audio file? say maybe theres like background music or something? is that the "noise reduction" option in applio?

[analog obsidian]

to separate vocals from instrumentals use the gabox_fv4 model in uvr

for just background noise (pc fans, mic noise, etc) mel denoiser aggr

[slim schooner]

aahh so the first option is only if its like people singing alongside a soundtrack?

and the second one would be as if people were speaking like in a livestream or something and you could hear their ac or something?

[analog obsidian]

can also work if someone is just talking but has background music (like a streamer talking while having some song in the background)

but it cannot remove sound effects, only instrumentals

[analog obsidian]

yes

[slim schooner]

aaa got it, i could probably add like 25 mins of audio then, hopefully this helps out the quality lol

thanks dude, you've been hella helpful

[analog obsidian]

ok so the true answer to this is that it can remove some of the sound effects, but not all lol

there's another model dedicated to remove sfx but the results are veeery bad for rvc

[analog obsidian]

be sure every audio is from the same source

rvc can't deal well with audio inconsistency

[analog obsidian]

so same mic used and same recording place

[slim schooner]

so like if i were to talk in a cheap microphone in a potato phone vs a good microphone? both of these audio clips would contradict because of the audio quality difference?

makes sense

[analog obsidian]

yup

for us it sounds fine but rvc isnt human, so

[slim schooner]

makes perfect sense actually, thanks

[analog obsidian]

[junior magnet]

aaa😭

[junior magnet]

mhmm

The audio has finally been converted

Thank youu sooo muchhh

[worldly magnet]

what the best tts ai i can use my own voices for free

[quiet axle]

hi i was wondering, is it possible to run VC on a phone. Of course not locally. Lets say i start my vc on my laptop, and i host it on my local network. Can i connect into it from my phone?

[low shard]

Yw

[low shard]

What's your PC GPU

[low shard]

The interface will be shown on your phone, but you can't really use it

Phones lack of a VAC

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

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

[quiet axle]

hmm i see, as expected tbh

so theres no VAC for android?

or its straight up impossible

[low shard]

Nope

It's straight up impossible to use it on ur phone rn

[zinc mesa]

One message removed from a suspended account.

[quiet axle]

aight then

[dreamy seal]

how do i create a ai text based that uses gguf models? since i tried using chatgpt and it doesnt work

[junior gull]

lol yeap...it was silently failing and just going through the motions >_<
FINALLY got everythign working for real....took 5 hours....and trained model sounds beautiful!
I've used it for TTS and realtime-rvc. So happy to FINALLY have a success. I ended up getting Applio to work, after I fixed a few things it ran perfectly. RVCv2 still not running right though.
How can I post my trained voice model to the voice-models channel? I've already packed it in a zip included the index and config and a working onnx conversion too.

[slim schooner]

@analog obsidian hey not sure if you're on atm but ive come across this uvr app that im using, i know you mentioned that I should use gabox_fv4 however it's nowhere to be seen or found. perhaps i need to use a different uvr software? or if i could manually download it from somewhere and use it with the current one? same goes for mel denoiser aggr. hmu when you can

[viral mango]

Is it normal for pre-processing a 10 minute data set in Applio to be taking over 35 minutes with a good CPU/GPU?

[simple ore]

no

with simple slicing 50-hour set of 109 files takes just a minute

[viral mango]

Got it. Just restarted and it finished within seconds

[final glade]

Am I only one who struggles uploading model on weights.gg? For some reason it says that verification not completed and my files are corrupted but everything is fine when using in local RVC

[tropic geyser]

Hi I'm using a voice modified from male to female specifically it's called psych2go-By-Dan I'm using it on Fivem the problem is that most of the time it sounds bad and very robotic can someone help me fix it?

[odd shale]

AI-based processes are more resource-demanding than videogames.

If you got a great pc and if you're using W-Okada for voice changing, then it's likely that the issue is on your w-okada settings or the model itself.

[distant turtle]

-colab

[patent trellis]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[tropic geyser]

I don't think he's using owaka how can I show you? I'm not very experienced @odd shale

[odd shale]

In case you're using any other sort of program for realtime voice changing, then discard it immediately

We only provide support for OG W-Okada and deiteris' fork.

All other ones are paywalled garbage.

[tropic geyser]

real time voice changer

come can i download this app? and most importantly a realistic male to female voice? @odd shale

[odd shale]

No need to ping me on every reply.

[tropic geyser]

oh sorry

[odd shale]

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

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

[odd shale]

Read the deiteris guide.

[tropic geyser]

I tried to install it but I don't understand how to open the application

I can't find the folder to open the application
is there a section where I can send you the screenshots? @odd shale

[low shard]

first of all, it's better you elaborate

what's your pc gpu? what tutorial link did you use? are the game settings to the minimum? can you show a screenshot of the program?

!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.

[low shard]

may be better to ask in discord.gg/weights

[olive plover]

Force FP32 mode: on (THIS IS OFF BY DEFAULT!) Turning this on improves stability, significantly reduces glitching/artifacting, increases VRAM usage by 200 MB.

I put it ( on ) will it improve the sound quality?

[crude flame]

Yes

But it’s very minor

[olive plover]

I didn't understand what you mean

[crude flame]

Yes it will improve the sound but not by a lot

[olive plover]

Ok, should I keep it on?

ON

[crude flame]

Yep

[low shard]

btw it will add a bit of delay for that little improvement

[analog obsidian]

i wouldnt call getting stable results a minimal improvement

[low shard]

I didn't say minimal, I wanted to just add that it does add more delay

[craggy saffron]

is there any way to... change... model names... in... okada... it... crashes... when... i.. try... to...

Error
unhandledrejection
no error stack

TypeError: Cannot read properties of null (reading 'modelSlots')

[hallow thistle]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

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

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

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

__**BE PRODUCTIVE.**__ 🤝

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

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

[hallow thistle]

Which W-Okada version are you using?

[low shard]

elaborate:

• your pc gpu
• what tutorial link did u follow

[craggy saffron]

i didnt follow a tutorial i install fork do the settings recomended

my gpu is 3060

rtx

[low shard]

what did you install exactly? send the link of whatever you used

fork just means modified version in IT field

if you make a project named "a" that just prints "a" on github, i can fork it, making a modified version that is called "ab" and prints "ab" to console

that's the meaning of fork

[craggy saffron]

idk

wait im gonna go find

it was something nvidia cuda

[low shard]

send a screenshot of the entire program or the program folder if you really don't remember the link

[craggy saffron]

it open in web

[low shard]

Wokada has 2 main versions:

• Original made by Wok
• Deiteris fork (modified version) made by Deiteris

each version has it's own updates

the latest deiteris fork has way better performance and quality than the latest original

from my guessing, it might be the wokada deiteris fork, is it the version b2332?

[craggy saffron]

it perfectly works

[low shard]

we have our guide for its latest version there https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

[craggy saffron]

Is that the right one

[low shard]

alr it seems the latest version, the issue is in that wokada deiteris fork there's a bug at the moment where you can't rename a model after uploading it, the only way is by renaming the model file then re-uploading the slot

[craggy saffron]

thats no issue then easy peasy

just rename the original file

[low shard]

yup, then re-upload the slot

[craggy saffron]

which one

json is the accent yes

[low shard]

pth, and if there is there, the .index too

json isn't needed at all

you can delete the json

[craggy saffron]

why is json not needed

[low shard]

it's just the metadata of the model got off weights.com

[craggy saffron]

if i get the model i use the whole model blud

[hallow thistle]

Or you can go through inside \MMVCServerSIO\model_dir\ folder, then use Notepad to edit "params.json".

[low shard]

it's not part of the model.. it's just extra info from weights.com which don't impact the model at all...

[craggy saffron]

hmmm

[hallow thistle]

You confused?

[low shard]

it means nothing for the program, it doesn't even utilizes it

[craggy saffron]

oh

[low shard]

but sure you can keep it and not save some storage if u really want lol

[craggy saffron]

its only 3 kb

so

theres like nothing in it

ok it changed the name now

thanks

[low shard]

in rvc context:

• pth files: contain the voice
• added index files: contain the accent
• metadata.sjon file: it's just some extra info about the model download link if you downloaded it off weights.com, it's not needed and won't impact the actual model at all

[hallow thistle]

In a voice model zip file that downloaded from Weights, there may be a json file alongside pth and index files, which is not needed for latter.

[low shard]

i mean sure do whatever, it won't really matter, tho it's usually just suggested to delete the json for saving some space

[low shard]

yw

[hallow thistle]

I'm just too slow to explain.

[craggy saffron]

wait .json is not the accent

oh

[low shard]

nope lol

[craggy saffron]

i thought it was the accent

do you need the accent

on 1 or 0

[low shard]

nope, that's the index

[craggy saffron]

yes do oyu need the index be on

[hallow thistle]

A json file is not what it is. In this case, a json file stores metadata of a voice model.

[craggy saffron]

i know voice models work without a index but is it really necessary

[hallow thistle]

The actual accent file is an index file.

[craggy saffron]

ye i just checked in it just realised that

[hallow thistle]

Not really necessary on W-Okada.

[low shard]

you need to play with the value,

higher value can give you a more similar accent to the trained one but can make it look like autotune

lower value will make it more like your own accent, but won't sound like autotune,

usually it's at like 0.3 or not used in realtime since it's not necessary but you can play around with it

[craggy saffron]

K

so if the model doesnt have a accent file i can just mash potato it with a another accent file from a different model

[low shard]

u can also optionally turn force fp32 mode in advanced settings for a more stable model and slightly better quality at the cost of some delay

[low shard]

if it doesn't have an index file, you can just not upload it, it's not needed

[craggy saffron]

is this right

[low shard]

i mean u can use the index file of another model, but it would be useless

[low shard]

force fp32 mode

[craggy saffron]

doesnt let me

[low shard]

click the stop button then try to change it

[craggy saffron]

to disable or enable

tahnks

so thats it?

[hallow thistle]

That's fast.

[craggy saffron]

yes

[low shard]

yup, u can also optionally use server instead of client to have less delay with more complex steps but u won't be able to use noise/echo suppression

u interested into that?

[hallow thistle]

I forgot where index slider setting located on fork W-Okada. Not gonna open my own W-Okada today, but this is what fork W-Okada looks like. https://cdn.discordapp.com/attachments/1371568582848417834/1372882565949816874/image.png

[craggy saffron]

what does that do

[craggy saffron]

thats the normal okada the one i have is fork

[hallow thistle]

https://cdn.discordapp.com/attachments/1371568582848417834/1372907409030381638/image.png

[craggy saffron]

interesting

[low shard]

it's a setting
server = can have less delay but is harder
client = has more delay but can use noise/echo suppression and is easier

which do you need? you're using client right now

[craggy saffron]

client

red or blue pill be faster but drain more energy be slower but have more energy

[hallow thistle]

No. It's fork W-Okada in that screenshot. Everything is the same, the difference is that it's DirectML variant.

[craggy saffron]

oh

[low shard]

also i noticed that the disable jit compilation setting is off

Disable JIT compilation: off for faster loading speed of the program, on for slightly better performance (10-15 ms) for Nvidia only)

do you wanna leave it off for faster startup speed or set it on for longer start up but a bit less delay?

[hallow thistle]

The normal or original version of W-Okada launches its own separated window. The fork W-Okada should always launch up a browser, so this it is.

[craggy saffron]

what does it chagne

exactly

[craggy saffron]

k

[low shard]

i explained it in the message, or is there a specific part that you didn't understand?

[craggy saffron]

"on for slightly better performance" on what the models or the function of ai

[low shard]

better performance at using the ai models, like reducing the delay of around 10ms

[analog obsidian]

provides a very small speed boost

[hallow thistle]

You got any idea that you still confused? I may explain more about that.

[craggy saffron]

ig ill enable it

[craggy saffron]

no i think i tried the normal version of okada

but i know it was bad at that time

shit

[low shard]

alr, just be sure that ur using vac lite and not vb audio cable and ur all fine

[craggy saffron]

u use vb audio cable for other stuff

i use vac lite for this thing

[low shard]

try restarting the program

[craggy saffron]

@low shard what do you mean by slower start up

its the same speed

[low shard]

like slightly slower to launch the program, but very minimal difference

[craggy saffron]

oh

okay

[low shard]

yeah it's not going to take like 5 mins to start up lol, pretty little difference

[low shard]

u should be all fine now

[craggy saffron]

K

[azure patio]

what would be the best site to train my voice model?

[low shard]

you should first of all check if your pc gpu is good enough to do it locally rather than a site

what's your pc gpu?

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

[azure patio]

i think NVIDIA Quadro P620

[low shard]

u checked it's that in task manager?

if so, then yeah your pc has wayy to low vram, 2gb vram are nothing nowdays

[azure patio]

yeah

[low shard]

Train (make) RVC Models on cloud:

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

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

3. Be sure to know about the tensorboard

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

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

You can use either:

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

[azure patio]

ah great 🥲

[low shard]

there's cloud, remote good pc

[low shard]

it's from 2018, now idk what u do with your pc, but AI is hungry for processing power and vram

[azure patio]

yeah definetly, its just my laptop that i use it sometimes

[low shard]

just like how content creation like video editing and games also need a decent pc most of the times

[low shard]

yeah, just to let u the market is shifting away from 8gb vram gpus lol

[broken urchin]

on w okada deiteris fork i get an average perf of 37 - 42ms so how low can i lower the Chunk setting?

[low shard]

close every single program in the background, set game graphics to the minimum, show a screenshot of ur wokada

[broken urchin]

no no, without any game open

just discord and wokada

[azure patio]

damn thats crazy

[low shard]

alr, show a screenshot of ur wokada

[broken urchin]

ok

[low shard]

yeah things change so fast, missing like half a year on the ai field is like missing 10 chapters atp for example

always new models racing with eachother

now iim not saying every single program and model but you get the point, the majority is, just like chatbots

[broken urchin]

i cant share screen

no perms i think

[low shard]

!give-media-perms 1h @broken urchin

retry

[azure patio]

have not checked it that much recently, but i was wondering whether voice models for examples are progressing much, i remember my models from 2023 being not that bad compared to now

[low shard]

unfortunately original rvc devs left the project to kinda rot to work on another TTS project instead, gpt-so-vits, which is different than RVC which is STS, so now rvc is mostly being experimented by the community

[broken urchin]

there

[low shard]

We are having improvements but mostly about performance

[broken urchin]

so how low can i lower the Chunk?

[azure patio]

ah i see, too sad for us 🥲

[low shard]

You might wanna check #🔊│ai-development

There's some of our engineers experimenting

Right now the most up to date fork is applio with the best performance improvements

[azure patio]

appereantly i have no access

[low shard]

get the ai testing roles in the guides at the top

[slim schooner]

hey guys, just wondering if i could get some help here, i originally pinged Lyery but they may have missed my message. anyways, i was running some tests on this uvr software and it seems to always produce a slight "static/hiss" noise throughout the whole clip. its not very noticeabale unless you listen very carefully but like people have said rvc aint human so theres a chance it will pick it up and ruin the whole model. Lyery mentioned on using gabox_fv4 and mel denoiser aggr for separation of vocals from instruments or background music and background noise like ac, fan, etc, respectively. however i can not find these two models. my question is, do i need to download a different uvr software or can i download those models from somewhere for me to use with the current one?

[analog obsidian]

join the audio separation discord (google "audio separation discord" since we cant share invites here)
then download the latest uvr beta gui there, if you got more questions regarding separation models ask them in that server

[slim schooner]

thanks Lyery, i'll do that 👍

[analog obsidian]

https://huggingface.co/jarredou/aufr33_MelBand_Denoise/tree/main
https://huggingface.co/GaboxR67/MelBandRoformers/tree/main/melbandroformers/vocals (download fv4, fv5 is noisy and bad for rvc)

hmm iirc latest uvr beta can download them from the gui itself
gonna check that rn

[slim schooner]

sounds good, i may not have downloaded the beta as i was not able to find those two there, but if the beta has 'em ill download that version instead

[analog obsidian]

i personally use the colab version of uvr because the conversion can be quite slow with big datasets (anything above 30 minutes)

[slim schooner]

ahh makes sense, i decided to split them into clips instead to make it easier somewhat.

[analog obsidian]

ok none of these can be downloaded using the gui

[analog obsidian]

ideally you want to merge all of the files into one

because for rvc training you have to split the whole dataset into 3s chunks

the process is like this:
merge all files into a big one singular file
truncate silence
simple slicing in applio using 3s and 0.3s

[slim schooner]

i could use adobe premier to combine them all in one right?

audacity is a bit weird and i dont like its ui

[analog obsidian]

you have to use audacity to truncate the dataset silence

otherwise results are going to be bad

rvc kinda hates silence

[slim schooner]

i could do that, i meant as just mainly the combining part, like combine all the files into one export that and import the single file into audacity?

[analog obsidian]

yup

installing the uvr models is easy

be sure to download the ckpt file alongside its yaml

then just follow the gui instructions

remember to export the file as wav 32bit

[slim schooner]

alright so i combine all files into one using adobe, export that as wav 32bit, and do the cleanup in uvr and then use the cleaned clip in applio?

[analog obsidian]

1. combine all files into a big singular one
2. clean the dataset
3. convert to mono and truncate silence
4. in applio select "simple mode slicing" and use the default settings of 3s and 0.3s
5. f0 extraction
6. train

[slim schooner]

got it

thanks, you've been hella helpful ngl 👍😁 hopefully it goes smoothly

[analog obsidian]

when cleaning the dataset don't use any type of eqing or compressors

[slim schooner]

gotcha, ill keep note of that

[analog obsidian]

you can use this truncate silence settings, work fine for me (those are the codename settings, i just borrowed them )

[slim schooner]

niceee this is gonna be helpful, thanks for that 👍

[azure patio]

does anyone knows what is causing this? i am copying the path of my audio file and im assuming its not finding it, when i try to pre-process

[simple ore]

what path do you give to preprocess?

[worldly magnet]

3060ti

[harsh nexus]

I want to do the voice of Mario Bros.
I want to download it for an animation.

[brisk trail]

Hello dear peoples, I have a question, So I'm working on an Alex Mercer voice model and I want it to be as high quality as possible, it's dataset length is: 00:30:58
What's the best batch and epoch for it?

[brisk trail]

Forgot to mention, I'm a newbie at training models and my goal is to become a model maker, this is my first model.

[north galleon]

i think weights is down again

or it’s at least having technical difficulties

i keep getting no healthy upstream errors

[slim schooner]

i went ahead and downloaded gabox and i get this error, not sure what went wrong here, sorry for the amount of screen shots lol, what would the fix be here? anyone know what went wrong? not a coder so whatever happened i have no idea 😭

[analog obsidian]

have you installed the beta roformer version?

[slim schooner]

dang forgot about that, lmaooo let me do that

[restive kiln]

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

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

[restive kiln]

How do I fix this? It does not show my mic

[low shard]

What browser are you using

[restive kiln]

Edge

[low shard]

Be sure you gave it microphone permission, you may also want to try if it works on chrome or firefox

[restive kiln]

oh god.

I thought I allowed mic.

I pressed block

[azure patio]

I just copy the path of the dataset file

[slim schooner]

just wanted to let you know i got it working and god damn, it's clean asl, no static, no background music its freaking crazy lmaoo

also just one question, i split the clips (beforehand like i said) but some clips have music while the ones that dont have any music do have paper flipping, would it still be fine for me to combine all clips and run them through both cleaning methods? or should i clean them separately and then combine them in adobe and then run that single file through audacity for truncate silencing and then export it to applio for training? lmk what you think

[empty pecan]

hi sorry if this was asked a dozen times, but how/where can i use the Cloud RVC?

[simple ore]

you need to provide a path to the folder, not to the file

like G:\training, not G:\training\character.wav

[azure patio]

aah i see, thanks

[spare marlin]

Also guys how to make ai cover the rvc v2 thing got shut down in google collab cuz no longer free

[low shard]

also use rmvpe not rmvpe_onnx

[low shard]

what's your pc gpu first of all?

[low shard]

you should first of all check your pc gpu to check if it's good enough for local

did you check it?

because it's pointless to use cloud when you got a good pc

[empty pecan]

i have an AMD gpu, is it ok but just slower?

[low shard]

which amd gpu?

there's many

[empty pecan]

AMD ryzen 5 7535U with radeon graphics

thats whats listed on my device manager like 15 times

[low shard]

that's not a CPU, not a GPU

[empty pecan]

oh sorry...

[low shard]

the CPU is the Central Processing Unit, while the GPU is the Graphics Processing Unit

[empty pecan]

AMD radeon(TM) graphics?

[low shard]

that's extremely weak integrated graphics, do you got any other gpus?

like a dedicated amd gpu

[empty pecan]

no idea where to check

[low shard]

or is it an old laptop

[empty pecan]

yes

[low shard]

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

it's better to check

[empty pecan]

i have no idea what im looking at, but it says the dedicated gpu memory is 0.8/1.0 gb

same gpu

[low shard]

just send a screenshot of your task manager

!give-media-perms 1h @empty pecan

[low shard]

nvm, your pc is extremely way too weak for local ai usage

[empty pecan]

mhm!

[low shard]

Train (make) RVC Models on cloud:

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

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

3. Be sure to know about the tensorboard

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

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

You can use either:

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

[low shard]

here

gave the list to u

[empty pecan]

thanks

imma try to use my brain now

[upbeat carbon]

Quick question: should i use seed vc instead of realtime voice changer?

Cuz idk

[low shard]

what's your pc gpu and what do u want to do exactly?

[upbeat carbon]

My gpu is like 6gb i think and someone said that it's weak anyways i've wanted to run seed vc but i've been getting countless issuses and new errors

[low shard]

i mean the gpu name

also, you want to do realtime voice changing for calls/games?

[upbeat carbon]

NVIDA I think

[low shard]

which?

[upbeat carbon]

Basically how it works is that you put an audio.mp3 and you can use that audio to talk

[upbeat carbon]

3060

[low shard]

in rvc context:
Inference = using the models
training = making the models

there's inference on pre-recorded audios like ai covers, then there's realtime inference which is used for discord voice channels for example

which do you need?

RVC = Retrieval-based-Voice-Conversion, it's the best quality few shots training Speech To Speech AI

[upbeat carbon]

But in there you don't need a pth and a index file, anyways can i sent a video so you can understand what i mean?

[low shard]

so you got the rtx 3060 with 6gb vram?

[upbeat carbon]

Yup

[low shard]

seed vc is 0 shot, i have seen that project before, but 0 shot is also lower quality compared to few shots

are you sure you want to use seed vc compared to rvc?

[upbeat carbon]

Not sure..

[low shard]

also, i asked you if you wanted to use ai for pre-recorded audios or realtime for calls, because there's different programs

id need to at least know this

[upbeat carbon]

Uh i just want it for talking and mabye games too

[upbeat carbon]

Oh sure! I want to!

[low shard]

soo, you want it for realtime in calls and games?

[upbeat carbon]

Yup

[low shard]

alright,
RVC = Retrieval-based-Voice-Conversion, the best Few Shots Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.

Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)

Your pc gpu is good enough, you can do it locally, idk who told you it isn't

What you seem you need is Wokada deiteris fork for the best realtime inference, which you can get by reading https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

if you want, you could optionally also use https://github.com/Plachtaa/seed-vc which is less used though, since yes you don't have to train models since it's 0 shot, but at the same time it will be lower quality

0 shot is kinda like a student who read 10 pages once, and few shots is like a student who read those 10 pages multiple times, ofcourse the 2nd student should perform better at the test

personally, i'd suggest you wokada deiteris fork, but it's your choice

[upbeat carbon]

Oh okay! Can you tell me why you recommend deiteris? I'm curious

[low shard]

well i kinda explained it at the end of the message

a properly trained rvc model should have better quality compared to another ai that doesn't train

-# ofcourse, if the model is trained on garbage, the model will be garbage

[upbeat carbon]

Oh okay mb i thought it was just some kind of uhh joke

[low shard]

no it was a comparison to explain you the difference between zero shot and few shots ai trainings

[upbeat carbon]

Also by training models you mean uploading a pth and an index file right?

[low shard]

by training a model i mean the process when a user prepares the dataset, uses the program to train on it using a pretrain, and at the end gets the .pth and .index which are the weights files you find in models trained by other users in #1175430844685484042

[upbeat carbon]

Oh okay! Tysm!

But how do i sound accuretly to the character that i want?

[hallow thistle]

Uploading pth and index files into RVC isn't training. It's more like you upload files of an already trained RVC voice model into RVC, so the program can do AI cover and so on. The actual training is when you upload an audio file which we called it a dataset into the program, and when everything all set you then click start training there.

[upbeat carbon]

Thanks for explaning! Thank you so so much!

[mortal topaz]

Is it possible to reduce artifacting somehow?

This is what the model Im using right now sounds like https://youtu.be/hckM9f1NyAI

using this model #1226933547940446251 message

Like, can using an embedder help?

[deep nova]

Anybody got a good video to help me create a good rvc model for the first time

[low shard]

All video tutorials are outdated

There's only written guides

What's your PC GPU

[deep nova]

rtx 3050

idk if its good enough

[worldly ibex]

is there a model i can run locally that can work from youtube? i dont wanna seperate the voice and shit

[eager shuttle]

The biggest problem I have is installing I have the files, but I always have trouble with the torch section especially with having to command prompt it never works

[potent bone]

i literally know nothing about making voice models should i use applio or codename fork? and also is there a guide that teaches me how or what to do?

[viral mason]

If I apologize to Weights will they be nice again and verify all my models

[tight ether]

applio. code's fork might be a bit complicated because it has a lot of parameters

[viral mason]

Applio is pretty good, I could teach you the kaggle applio

I have no experience with codename tho so can't help there, dm me whenever you're ready

[odd isle]

whats the diff between applio and kaggle applio

[viral mason]

idk but it's just what I use, kaggle applio

I don't think there is a difference

[stiff valley]

guys i need help there's an echo where it seems like the voice changer is picking up the changed voice, and applying the changer again. Is there a way to fix that?

[low shard]

how much vram?

[low shard]

be sure to not use video tutorials

what's your pc gpu?
show a screenshot also of the program

[low shard]

applio is basically the most up to date easy to use, codename is based on applio and has some extra feature, for beginners it's better applio, what's your pc gpu?

[low shard]

elaborate

what's your pc gpu

what do you want

what do you want to install

be sure to not use video tutorials

[potent bone]

3060 ti

[low shard]

applio is the program itself, kaggle is just the cloud computing service that can be used to run the code for people with a bad pc

[viral mason]

what he said

[low shard]

did you even check your pc is good enough before using applio kaggle lol

this is why u should always check ur pc before using cloud

[low shard]

nice, good enough, check https://docs.aihub.gg/rvc/local/applio/

[stiff valley]

bro it started working now thanks, 4060 btw

[viral mason]

no I just use it because it's fast and simple to use

I dislike local because it makes my vr run slow

[low shard]

do u may want me to check your settings? I can suggest you the best ones for quality and least delay

[stiff valley]

i think i cant

[low shard]

i mean, cloud got limited free time and can break easily because kaggle can update its dependencies whenever they want

[stiff valley]

where can i send it

[low shard]

!give-media-perms 1h @stiff valley

try now

[low shard]

lemme guess, you used a youtube tutorial?

[stiff valley]

yessirski

[low shard]

this is an old version of original wokada

and the settings are completely wrong lmao

[low shard]

never do that for ai programs

they are outdated asf