✨│ai-help

[hexed cave]

nvm nvm

antivirus stuff stopping me

[odd shale]

MVSEP, right?

[hexed cave]

no

Ease US

https://multimedia.easeus.com/vocal-remover/?utm_source=aitoolssearch&utm_medium=youtubevideo&utm_campaign=bella-202312-vr

but I can download uvr

[odd shale]

Nah, i suggest you to use MVSEP

[hexed cave]

oh-

I can try uvr but idk how to use it lmao

[hexed cave]

oh that website

isn't there like

a queue u have to be in

[odd shale]

You can find it on mvsep.

[hexed cave]

zugzug I tried downloading uvr

didn't work

but u have to wait for a queue

I sees

it didnt work

oof

weird

[rare goblet]

Ayo? @wraith kayak level 5 !!!

[humble jetty]

rvc not running

i see the pop up text where it says press anything to conutine

[hexed cave]

I see it now

[rare goblet]

Ayo? @hexed cave level 3 !!!

[hexed cave]

there is a lot to download

I'll

pass on uvr

and just stick with mvsep

[half cove]

id recommend just using the main download links

[humble jetty]

why my rvc aint poopin up

[rare goblet]

Ayo? @humble jetty level 1 !!!

[devout turret]

hello, where can i find the rvc google sheet where every model was uploaded? thx.

[hexed cave]

I did use the main download links

got an error after the set up was finished

[half cove]

uvr problem

[proper shale]

that isnt updated anymore soo

use https://weights.gg

[hexed cave]

ah-

so its not a problem on my end?

[half cove]

dont know

ive never had problems

[hexed cave]

huh

[hexed cave]

@half cove

this is the error I get

[half cove]

i dont know about uvr errors

[proper shale]

maybe bcuz space on folder name? ehhh

[hexed cave]

its strange

now im waiting my ass for instrumentals and vocal separation

with mvsep

[brittle wing]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces
• GPT-SoVITS (demo), by litagin Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[rare goblet]

Ayo? @brittle wing level 2 !!!

[hexed cave]

I just want a clean vocal n instrumental separation

unfortunately that's not the case cause uvr is not working for me

I found another cupid cover that didn't include the background voices

its not bad but I want to find a way to get an even cleaner vocal n instrumental separation

[devout turret]

thx

[rare goblet]

Ayo? @devout turret level 1 !!!

[hexed cave]

but I can't cause uvr doesn't work for me so I'm kinda upset TvT

[sour merlin]

i used mangio RVC web and it shows two errors where it will make the sound it creates what can I do

[hexed cave]

mangio rvc wont work for me

cause my security finds it as a virus or anything malicious

[sour merlin]

what do you use

[humble jetty]

how do we make our own models?

[hexed cave]

I used a different one

I tried using kits AI

but u are limited

so I was told to use this instead

https://github.com/Tiger14n/RVC-GUI/releases/tag/Windows-pkg

I have a pretty meh computer

with a gtx nvidia 1650 gpu

[half cove]

thats old

dont use

[sour merlin]

i tried that

[rare goblet]

Ayo? @sour merlin level 1 !!!

[half cove]

also dont use kits.ai

that shit poopy

[sour merlin]

which one should I use

[half cove]

you can just use google colab

https://aihubdocs.github.io/en/rvc/cloud/inference/ilaria-rvc/

for inference

(making covers)

[sour merlin]

thank you ;''

[split latch]

where do I export huh....

I do not see 3 dots

[hexed cave]

I see

[half cove]

-links

[azure marsh]

Suggestions for @hexed cave

🖥️ Local stuff

• 🍏 Applio, by IA Hispano GitHub
• Mangio-RVC-Fork, by Mangio621 Huggingface
• RVC Studio, by SayanoAI Huggingface
• AICoverGen, by SociallyIneptWeeb GitHub
• Replay, by Replay Team Website
• Original RVC, by the RVC-Project team GitHub
• GPT-SoVITS, by RVC-Boss GitHub

Need some help? 🤔

You can find more info on the #1159513888199540817 channel. If you can't find your answer, feel free to ask for help in #✨│ai-help. Credits to Faze Masta and Antasma for compiling these links.

More commands: -audio, -colabs, -overtrain, -realtime, -rvc, -help

[half cove]

use original rvc

we also have a guide for it https://aihubdocs.github.io/en

[hexed cave]

thanks

original rvc

is written in chinese-

[split latch]

@odd shale yo I finished the training and it's AWESOME

[odd shale]

You're welcome

[half cove]

most of it should be english

[split latch]

if there is a lot of stops in an audio what can I do for that

[odd shale]

Poop being poop

[hexed cave]

mothafacka

is mangio any good?

[half cove]

by that i meant the webui

[hexed cave]

im downloading mangio

[rare goblet]

Ayo? @hexed cave level 4 !!!

[hexed cave]

original vc just too much chinese

[sand iris]

mate

[young halo]

yo, there's a model that i made, but i'm pretty unsure if i might upload it or not, can you take a look of it?

https://huggingface.co/ivoryprincemilo/WILLOW/resolve/main/WILLOWsmith.zip?download=true

[sand iris]

the links are explained in plain English

The UI is in plain English as well

mangio is like deprecated by this point, and I wouldn't recommend it

[hexed cave]

ugh

[sand iris]

Are you wondering about the quality or something?

[young halo]

yes, and since i got the model maker role, i can't upload it to #1159514067187277865

[hexed cave]

-links

[azure marsh]

🖥️ Local stuff

• 🍏 Applio, by IA Hispano GitHub
• Mangio-RVC-Fork, by Mangio621 Huggingface
• RVC Studio, by SayanoAI Huggingface
• AICoverGen, by SociallyIneptWeeb GitHub
• Replay, by Replay Team Website
• Original RVC, by the RVC-Project team GitHub
• GPT-SoVITS, by RVC-Boss GitHub

Need some help? 🤔

You can find more info on the #1159513888199540817 channel. If you can't find your answer, feel free to ask for help in #✨│ai-help. Credits to Faze Masta and Antasma for compiling these links.

More commands: -audio, -colabs, -overtrain, -realtime, -rvc, -help

[hexed cave]

so much fking shit holy by that I mean gb's

[young halo]

the model itself is an updated version of another model that i already uploaded, with a longer dataset and more epochs, but there's something of it that doesn't feel good for me

[sand iris]

the two first lines

"for Nvidia GPU owners"

"for AMD GPU owners"

(sidenote: the amd package can only do realtime, because they removed the file conversion module for some reason)

[tepid grove]

how to create my own model?

[hexed cave]

alright

I have it downloaded

[reef marten]

hey uh

what should i use?

[sand iris]

Mangio if it works

Like if it works it works, no need to change it

[reef marten]

ive used mangio for a long, long time

[sand iris]

else, unless the main thing, or applio, whichever works

[reef marten]

could you link me up?

[hexed cave]

the tutorial I watched for making ai covers was like 2 months ago n they mentioned mangio

[sand iris]

Things are rather brittle, so to be frank pick whatever has what you need & still works

There haven't been any significant improvements, only convenience features

[hexed cave]

I got the nvidia gpu owners thing downloaded

[sand iris]

Yup

So now the only tricky part : you have to extract it with 7zip

anything else fails (the windows default thing and WinRAR both error out)

[reef marten]

i watched a tutorial for ai voice covers like more than a year ago and i downloaded mangio then and have been using it since

[hexed cave]

wut-

[sand iris]

you downloaded a .7z file right?

[hexed cave]

yes

[sand iris]

https://www.7-zip.org/

That's the only software that manages to unzip the file

Other options don't work for some random reason

(7zip is a great tool to have anyways)

[hexed cave]

I download the 64-bit yes?

well

[sand iris]

yup

extract that in any folder you want

[hexed cave]

alright

[brittle wing]

For each voice examples. I'm making 5 audio examples of the voice but how long should each of them be to get good results?

[hexed cave]

it extracted

pretty quickly

[odd shale]

You can split them in 10 or 5 secs clips

An dataset should be at least from 10 to 30 mins long to get good results

[brittle wing]

Oh thank you very much for this.

[glad veldt]

is there a kaggle version of rvc disconnected?

[hexed cave]

I got em extracted

[sand iris]

launch go_web

[hexed cave]

aight

is this what im supposed to see?

[sand iris]

yes

[hexed cave]

alright

well

what am i supposed to do?

[sonic agate]

how many epoch for a 59 min dataset?

[rare goblet]

Ayo? @sonic agate level 3 !!!

[sand iris]

what do you want to do ?

gonna head to bed, falling asleep as we speak lol

[hexed cave]

i want to make ai covers-

but there's a guide so

[molten pecan]

https://aihubdocs.github.io/en/ check the epochs page :3

[brittle wing]

How do I make a PTH? I made 5x 10 second audio clips & made them into datasets & thats about it. I downloaded the local RVC aswell

[molten pecan]

Check the Training chapter

Scroll down

Also, turn on Night Mode for a better experience =3

Aaand also, you don't need to split the dataset with modern versions of RVC. 1 audio file is enough

[brittle wing]

Makes sense haha & Thanks will try this out now

[split latch]

how good is my audio cleaning?

[hexed cave]

its sounds really smooth

[split latch]

thanks for letting me know, original voice had a lot of background noise so I had to be sure lol

[hexed cave]

I don't hear any breaths, static or anything unusual

[hexed cave]

:D

[brittle wing]

Tried to find the training part but it's only really explaining about over training etc

[molten pecan]

Sounds good. Only thing is, seems like the quality is compressed

Were you using MP3 or OGG?

[split latch]

no I downloaded using stacher to wav

but maybe that is noise reduction in audacity

[molten pecan]

Go to the Mainline page and go to the Training chapter

Seems like you went to the Epochs page

[molten pecan]

You downloaded from YouTube?

[brittle wing]

Yeah found it now many thanks!!

[split latch]

yeah but I do not use paid downloaders only free ones

maybe because a free downloader isn't that high quality

[molten pecan]

I see. Just a heads up, YouTube compresses the quality by default

[split latch]

the creator does not have spotify sadly

[molten pecan]

Once you upload something in it, a lot of the audio quality will be gone

[split latch]

I will see if she has soundcloud or smth

[molten pecan]

Spotify also compresses a lot, so not much difference hehe

[split latch]

XD

nope. no soundcloud either, I mean does it affect it that much?

here is another sample

[hexed cave]

wait wait question

[brittle wing]

So what I ment previously before is I have 5x 10 second examples & then it created 1 whole dataset zip file. This is okay? @molten pecan

[hexed cave]

is the software im using right now good for ai covers?

[dawn nebula]

yeah

its mainline rvc

[hexed cave]

alrighty

im looking at the guide on github

lets see if I can make a good Hu Tao Cupid ai cover-

(im worried the background noise will ruin it)

[split latch]

@molten pecan worked on the audio a bit, how about now?

[dawn nebula]

this seemed nice

[dawn nebula]

use uvr to separate the vocals out

[hexed cave]

i cant use it locally

[rare goblet]

Ayo? @hexed cave level 5 !!!

[split latch]

nice thanks

[hexed cave]

so im using it through colab

[hexed cave]

[dawn nebula]

use mvsep.com

-audio

[azure marsh]

📚 Audio Guides

• Perfecting Audio Isolation on Low-End Rigs: A Practical Guide, by Litsa The Dancer and Faze Masta Google Docs
• Gathering and Isolating Audio, by SCRFilms Google Docs
• Creating Datasets for RVC using iZotope RX, by LUSBERT Rentry
• Vocal Mixing Tutorial, by Roomie YouTube

🛠️ Tools

Audio Separation/Isolation

• https://mvsep.com/
• https://multimedia.easeus.com/vocal-remover/
• MDX-NET Colab

More commands: -colab, -uvr, -karafan, -overtrain, -help

[dawn nebula]

@hexed cave read this

[hexed cave]

where is that

[hexed cave]

ah

it takes a long time

[odd shale]

Hey bud

[split latch]

yo wassap

[molten pecan]

Oh yeah this is way better

Nice job

[odd shale]

Can you send me the dataset of your girl model via DM?

[molten pecan]

Yup that's fine

[split latch]

sure

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

https://docs.aihub.wtf/v/brazil

🇪🇸 Spain

https://docs.aihub.wtf/v/spain/

🇫🇷 France

https://docs.aihub.wtf/v/france

🇩🇪 Germany

https://docs.aihub.wtf/v/germany

🇯🇵 Japan

https://docs.aihub.wtf/v/japan

🇰🇷 Korea

https://docs.aihub.wtf/v/korea

🇵🇱 Poland

https://docs.aihub.wtf/v/poland

🇷🇺 Russia

https://docs.aihub.wtf/v/russia

🇮🇹 Italy

https://docs.aihub.wtf/v/italy

[odd shale]

I told him to do that

[molten pecan]

https://aihubdocs.github.io/en/ here's a complete guide regarding mainline & cleaning audios through the cloud =3

[hexed cave]

alright thanks

[molten pecan]

You'll need great specs though

To run UVR locally

[hexed cave]

yeah

I once finished the setup

installation for uvr

and I got an error

[dawn nebula]

if you log in, you dont

[hexed cave]

is it as good as uvr?

[brittle wing]

What you reckon is a good setting to produce some decent quality voices? I've found the determined of the sample rate which is 48K btw!

[dawn nebula]

its literally using the same uvr models

[hexed cave]

ah

sick

[dawn nebula]

sample rate should be 32k

[brittle wing]

Oh I went through an article and followed the instructions to find the sample rate for my audio. Apparently choosing the wrong one can cause issues on it's final render.

[dawn nebula]

show me the spectrogram of ur audio

[brittle wing]

No problem!

[molten pecan]

I showed what to do in case the frequencies don't reach the top, dw

[proper shale]

seems like a decent 40k model... but I'd use 32k

[molten pecan]

Why?

[proper shale]

because 32k is downright better

for every voice

💀

[molten pecan]

Because of the pretrain?

[proper shale]

no

because less noise

[molten pecan]

Ah

But why not 44k instead?

Does it still have noise?

[proper shale]

44k. on rvc?

[brittle wing]

32k it is then! What about the other settings?

[molten pecan]

Yeah

Or 40k

[molten pecan]

They are explained in the mainline training chapter

[proper shale]

ig nothing else you can change, everything's fine

[proper shale]

i wouldn't use it

[brittle wing]

Yeah it just says process data. Don't I have to do anything with EPOCHS lol

[molten pecan]

You have to keep reading

[proper shale]

oh yeah maybe set that to 500

[molten pecan]

There's more below

[brittle wing]

I'll do just this lol

Sorry sorry

[proper shale]

rest is fine

[molten pecan]

Omg why do I make guides for people to just read half of them

[proper shale]

nvm actually change the save small finished model to weights option to yes

[brittle wing]

Because we rush 👁️

I'll learn from my mistake

We just wanna make it work out great first time lol

[molten pecan]

Also, are you looking for a quality model?

Or just making one casually?

[brittle wing]

Like pretty realistic yeah mainly. Just doing it for my voice changer!

[finite galleon]

Always happends when making guides

[molten pecan]

@proper shale why did you tell him to use 500 epochs

Don't listen to MJ

[finite galleon]

Heck even if you link a guide they will still ask the same question

[molten pecan]

He's a ding dong

As AI Hub's owner, use TensorBoard instead

[brittle wing]

Don't worry i've not done this yet haha

[rare goblet]

Ayo? @brittle wing level 4 !!!

[molten pecan]

You'll be taught how later, dw

[dawn nebula]

yeah use 32k

[molten pecan]

I'll go buy ice cream

[dawn nebula]

more stable and u cant hear any difference

[molten pecan]

Tell me your fav flavor

[dawn nebula]

ice cream?

[brittle wing]

Process data just quickly does something then an orange border shows outside of it without no text...

[brittle wing]

Alrighty lol

We got a problem. No such thing as 32k here 😉

[finite galleon]

Did you choose v2 architecture

[brittle wing]

No clue what I did. Just followed the guide thing and downloaded

[half cove]

ok so

@brittle wing

just click v1 and then v2

to load it

rvc webui is weird so it doesnt preload it

[brittle wing]

Yeah I see it now 😄

[half cove]

np bbg

[finite galleon]

🤨

[brittle wing]

apparently my directory has an error 💀

[half cove]

[finite galleon]

How the hell

How can a folder have an error

[brittle wing]

It's a zip file maybe?

[half cove]

wat

[brittle wing]

Not sure. I click on copy path and removed the text and pasted then process file

Then get that issue

[finite galleon]

Your path is probably not valid

[brittle wing]

I'll try put it in a more convenient place

[finite galleon]

Or maybe your file name is wierd

Dont use spaces

[brittle wing]

yeah that will be the issue

Should I remove the " when I paste it into the blank text?

[hexed cave]

https://www.youtube.com/watch?v=RxPF0WqjVP0 I need to confirm that this video after separation would work well with my Hu Tao model

or not bc it has background vocals

[proper shale]

...why not?

Genuine question

[proper shale]

yeah remove the "s

[brittle wing]

Okay!

Not sure why my directory is invalid....

There are no spaces either

[half cove]

For background vocals you can use bve

[hexed cave]

bve?

[half cove]

Your training folder doesn't need to be zip

[proper shale]

oh yeah

you must insert the folder patj

💀

[molten pecan]

He's looking for a realistic model for real-time inference

[proper shale]

not the zip file

[proper shale]

and?

like...

[brittle wing]

Oh

[proper shale]

500 epochs is good enough like 💀

[half cove]

You use mvsep right? It should be in uvr then bve

[molten pecan]

It won't be convenient to use 500 epoch like nothing

[proper shale]

id suggest unzipping and keeping the audios in a folder

[hexed cave]

yep

[molten pecan]

It can be better

[proper shale]

yeah exactly

[hexed cave]

well I can't run uvr locally on my pc

[proper shale]

they can continue training later

lmfao

[hexed cave]

remember? I get the error

[molten pecan]

What if they overtrain?

Overall it's more convenient to use TB from the get go

[hexed cave]

idk how to stop this error every time I run it locally

[brittle wing]

Alright i'm back on track now lol

Thanks guys

[molten pecan]

https://aihubdocs.github.io/en/ check the vocal isolation page. Use the local version of UVR

[proper shale]

then they can just use a previous model

lmao

[molten pecan]

That'll work

[proper shale]

why did you immediately go "DONT LISTEN TO HIM"

like if i had just said "GO FOR BATCH SIZE 20"

[molten pecan]

I was joking, I wasn't being that serious

[proper shale]

😭

[brittle wing]

Yall put in 1 for GPU?

[proper shale]

im taking everything seriously sorry

[hexed cave]

main download link?

[molten pecan]

You don't sound okay MJ

Been stressed over school?

[hexed cave]

school stress? no way

[proper shale]

yup

[proper shale]

yes, and also no

there's more to it

[hexed cave]

oh no

[molten pecan]

Whatever is going, take care MJ

[brittle wing]

Makes sense?

[proper shale]

yep

[molten pecan]

Seems good

[proper shale]

idk

[brittle wing]

Alright

[proper shale]

i shouldnt be talking about my problems in HELP RVC lmfao

[hexed cave]

true

xd

at least u got support

[odd shale]

What's the problem bud?

🥺

[brittle wing]

Here is the real deal. What you guys reckon will be suitable for a somewhat realistic voice result?

[odd shale]

Not always.

[odd shale]

Set it to 25 or 30

But that's just for pth checkpoints.

That won't improve your model.

[half cove]

thats the total epoch

[brittle wing]

Thats not the pth checkpoints

[half cove]

😭

[hexed cave]

poopmaster

[half cove]

yea?

[odd shale]

I'm stoopid

[hexed cave]

im currently downloading

the uvr

[half cove]

[hexed cave]

I ran as administrator

[brittle wing]

No problem lol

[odd shale]

Put 500 epochs.

[half cove]

i dont know about uvr problems

[hexed cave]

.exe is installing the uvr

[brittle wing]

Thanks!

[hexed cave]

prayers to not failing this time

[odd shale]

UVR has a installer

[hexed cave]

u mean dis?

I ran that as administrator

[odd shale]

Yes.

Install it

[hexed cave]

I am rn

[half cove]

before that had problems

what did you do before?

[hexed cave]

yup

uhhh

I just followed the instructions

FUCK

it happened again

no idea why it isn't working

hell

even the cloud version has problems on my end when installing

[hexed cave]

for this I clicked on main download link n not the mirror

[brittle wing]

Was it right to put tensorboard inside of RVC folder? it's not showing anything when I'm training?

[hexed cave]

idk why it no workie

if I can't use uvr then I'll have to stick to mvsep as my last resort cause uvr not working

meaning I can't use cupid bc that has background noises and such. No idea what other song I should use other than cupid

[brittle wing]

Looks like my .PTH model didn't appear in the weights folder. My INDEX one did though... any help on why this could be a problem?

[dusty rampart]

no

[dusty rampart]

did you click train model?

[brittle wing]

Ofcourse yeah 👍

[dusty rampart]

did the console output it was trained

and saved?

[brittle wing]

Gives me the index but not the PTH

[zealous copper]

hello! i was starting to get into doing ai voicing, but don't know where to start. can someone redirect me in the right direction?

[dusty rampart]

those are 2 different buttons

train index and model

mark the last one as yes

[brittle wing]

And yes I did click train model

Alright thanks

[dusty rampart]

check pinned on #✦│chat

[brittle wing]

Does not appear still even after it's finished

[rare goblet]

Ayo? @brittle wing level 5 !!!

[dusty rampart]

what parameters did you set?

bc how tf did you train the model that fast

send a screenshot of the console

[brittle wing]

Sent ^^^

[dusty rampart]

console is the CMD

if you're doing it local

otherwise check the colab tab

[brittle wing]

Yeah I am, I know what you mean now

I can't download the pth file and what's the reason of this?😭

1 sec

[dusty rampart]

check if you're still connected to runtime

since you're using disconnected

just use the cell to save to drive

[fleet stirrup]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces
• GPT-SoVITS (demo), by litagin Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[hexed cave]

-links

[azure marsh]

🖥️ Local stuff

• 🍏 Applio, by IA Hispano GitHub
• Mangio-RVC-Fork, by Mangio621 Huggingface
• RVC Studio, by SayanoAI Huggingface
• AICoverGen, by SociallyIneptWeeb GitHub
• Replay, by Replay Team Website
• Original RVC, by the RVC-Project team GitHub
• GPT-SoVITS, by RVC-Boss GitHub

Need some help? 🤔

You can find more info on the #1159513888199540817 channel. If you can't find your answer, feel free to ask for help in #✨│ai-help. Credits to Faze Masta and Antasma for compiling these links.

More commands: -audio, -colabs, -overtrain, -realtime, -rvc, -help

[brittle wing]

Yeah but how check of If I'm still connected to runtime I'm not familiar with this yet

[dusty rampart]

your GPU is too potato to train

[dusty rampart]

top of the page should have a T4 icon

and show "connected"

[brittle wing]

So I can't do anything about this is what your saying?

[dusty rampart]

use rvc disconnected

[brittle wing]

I just can't make models

[dusty rampart]

-rvc

check pinned message on #✦│chat

the docs one

[brittle wing]

Ok

[brittle wing]

Okay but pls do screenrecord of the tutorial to understand more?

[dusty rampart]

you should see something like this

just use the save to drive cell

and download it from your drive

[brittle wing]

Okay

[hexed cave]

how do I start making an ai cover with this

I have the separated music and model

I put in the model .pth and .index file

[ornate hatch]

I have a problem....

[brittle wing]

On tensorboard, how to know if i'm overtraining?

[hexed cave]

where do I put the separated vocals?

[rare goblet]

Ayo? @hexed cave level 6 !!!

[brittle wing]

hey guys i downloaded the realtime voice changer client but its not working can anyone help me

[hexed cave]

@half cove sorry for pinging ya but I just need confirmation as to what the audio file means like the vocals of the song I chose? Cause I have the vocals and it has special characters such as -, _, [ so do I just remove them from the file name?

[half cove]

anything thats ascii you can

dont use spaces

[brittle wing]

Pretty sure the training is doing good from here right?

[thorny cargo]

its on the page

theres a link

[brittle wing]

Should I only stop when it sky rockets?

[hexed cave]

they're mp3's

[brittle wing]

#✨│ai-help message

[half cove]

mp3 is fine

[brittle wing]

My training ok?

[hexed cave]

so copy the vocal audio

[half cove]

the path of the audio

[hexed cave]

aight

"C:\Users\Leslie\Downloads\ruthbdandelionsaudiomdx23c7vocals.mp3"

somethng like this?

[half cove]

yes

[hexed cave]

thank you poop master

[brittle wing]

Same, I also have a problem with that Gradio as well 😭

[brittle wing]

Is this considered good training by far?

Or I'm I over training?

[half cove]

wrong graph

[brittle wing]

Which one?

[half cove]

no wait

its right

whats ur smoothing at

[brittle wing]

Max 0.999

[half cove]

its 2am i can barly see

[brittle wing]

When do I only stop training lol

[hexed cave]

oh damn

[half cove]

when it go up

also just test different checkpoints

[hexed cave]

any of the inferences settings should I change?

[half cove]

use rmvpe

[hexed cave]

alright

the settings on the left are fine right?

I used rvmpe before

[half cove]

dont use ur transpose at 1.2

[hexed cave]

should it be at 0?

[half cove]

use a full number

[hexed cave]

cause the original vocal is already female

[half cove]

then yea

[brittle wing]

I think it was reasonable to stop training when that went up alot more then it should right?

I got it's PTH now so cool stuff!!

[hexed cave]

can someone do a quick explanation whats artifacting?

[brittle wing]

We can run that cell after trained the model right?

[hexed cave]

it sounded robotic at that part

[brittle wing]

With my results, I think it sounds okay but just abit robotic. Maybe I should resume & leave it abit longer

Very happy overall

[hexed cave]

I need people's opinions on how to fix it

the song im using is by Ruth B. - Dandelions

the results sounds a bit breathy and robotic

[thorny cargo]

Hi, bit of an inexperienced question but I tried to load a voice model and index file into the okada software, but this was an earlier point in the training that's within the log's. it's at 300 epochs and I wanted to give it a brief test before I let the training continue however it would not load properly into okada.

I tried with both the "D" file and "G" file

[half cove]

d and g are training files

[covert turret]

you should use model.pth not the D or G ig

[thorny cargo]

o

so I oughta wait for the final one

[rare goblet]

Ayo? @thorny cargo level 1 !!!

[dawn nebula]

wait for the final one

[covert turret]

yeah, or use a saved state epoch count

[thorny cargo]

right on

[hexed cave]

it still sounds robotic

[dawn nebula]

or stop training and convert

[hexed cave]

at the end

when I converted it

[half cove]

could just be a shitty model

[hexed cave]

true

should I get the Hu Tao model here?

[thorny cargo]

there is a setting that gives the choice "whether the save only the latest .ckpt file to hard drive space" are these checkpoints earlier iterations of the model?

or is it something else

AH i found the earlier iterations in the weights folder, my bad for not seeing it

time to cook

[covert turret]

real

[brittle wing]

Do you guys reckon this is a stable training progress? I feel like I wanna leave it until it rockets back up.

[covert turret]

looks fine to me

[brittle wing]

Perfect!

[covert turret]

just make sure it's still going downwards

[rare goblet]

Ayo? @covert turret level 5 !!!

[hexed cave]

is a Hu Tao model at 400 epochs n rvmpe a good enough model?

[brittle wing]

Yeah am refreshing every 1 minute to make sure it's looking good.

[half cove]

Haven't tried the model so I can't tell

[brittle wing]

I've seen it go up abit but then go down. Would it be a big concern if the graph suddenly showed a medium spike going up?

[hexed cave]

I'll let you know then

[half cove]

More epochs doesn't mean better

[hexed cave]

ah

[brittle wing]

My epoches are 500

[thorny cargo]

I oughta get tensorboard to actually work so I can maximize the potential of the model

[covert turret]

it can be a concern when it goes wildly up or stagnates, otherwise you're fine

[brittle wing]

Thank for letting me know about this. I did this first time before and kinda panicked because I assumed it was spiking up, but it wasn't. The model sounded okay, but abit robotic. So i'm on round 2 lol

[thorny cargo]

Is it slow to get an epoch every 4 or so seconds?

i don't know what to expect

first time training lol

27 min dataset

[covert turret]

yeah lol that happened to me a lot, if it keeps going down you're golden essentially

[brittle wing]

This terrifies me now haha

[covert turret]

lol

[brittle wing]

When do you only stop the training? like when it goes past 2 numbers on the left side?

[thorny cargo]

wtf am I looking at

[covert turret]

when it keeps going up, i have it checkpoint every 10 epochs for that very reason, i prefer to have a checkpoint that is closer to a good point before overtraining

[ornate hatch]

Could someone help me please?

[covert turret]

i guess it is erroring out due to the file being mp4?

[brittle wing]

There is a step I read on the main logs, that you choose the second to last PTH because the last one is the overtrained one

[ornate hatch]

But it gave me the same error with a mp3 file

[brittle wing]

Really hitting the GOLDEN so far though

[hexed cave]

it still sounds robotic

crapp

[covert turret]

the model.pth is the last saved epoch, there are other .pth's like model_e400_s12349.pth that basically let you know what step did that checkpoint was at and stuff

[ornate hatch]

It sometimes happens with me

[thorny cargo]

i just set epochs way up. not sure what to expect

[covert turret]

it's still going lower, so not too bad

[hexed cave]

ima just change the song

[brittle wing]

So I'm assuming the lower it gets, the better the quality of the model?

[hexed cave]

I can't use uvr cause im imcompatible with it

so i used mvsep

[thorny cargo]

is this the graph that matters?

[brittle wing]

Yeah correct. Put smooth on 0.999 so you can really analysis it! 😄

[ornate hatch]

It also gave me ERROR when I tried with an MP3 file

[covert turret]

That's correct for the most part, but not always

[thorny cargo]

and i suppose its doing fine considering it displays a similar trend

[ornate hatch]

What's that???

[brittle wing]

Oh scary. I'm doing this to use fro my a.i voice changer lol

[covert turret]

g/total and d/total are

[ornate hatch]

@covert turret why does it give me error?

[covert turret]

idk, haven't really used ila rvc tbh

[ornate hatch]

Oh, alr...

[hexed cave]

usually it would be the music sometimes. When I see Ai covers, they would take covers of the song n separate the vocals and instrumentals

[thorny cargo]

so when the model i am training is on the cusp of overtraining i simply click stop training?

[rare goblet]

Ayo? @thorny cargo level 2 !!!

[thorny cargo]

[silver dirge]

guys im new to this which collab is better in cloud?

[thorny cargo]

[hexed cave]

how hard is it

to make an ai cover with a japanese model

[sour mason]

when I run the batch file to run tensorboard, I get this

Launching TensorBoard...
'tensorboard' is not recognized as an internal or external command,
operable program or batch file.

and then press any key to continue to shut it

[dusty rampart]

ye

[hexed cave]

I accidentally closed the

console

without closing the web

on my browser

how do I download my exported audio? The three dots are not there

@half cove

[half cove]

It's 3am now

[half cove]

Yea?

[hexed cave]

oh god

3 am

thats my horror time

anyways I can't download my exported audio

[half cove]

Try use it on another browser like Chrome

Just copy the localhost link

[hexed cave]

alright

it'll save my stuff right?

[half cove]

It will refresh

Just take a image

I mean you already did

[hexed cave]

yeh

I should be fine

yippie

I got it

[thorny cargo]

should I keep it goin lol

[hexed cave]

poopmaster

[thorny cargo]

its already past 400 epoch

[hexed cave]

it doesn't sound soo bad

thanks for your poopmaster

[thorny cargo]

poopmaster?

[hexed cave]

its not bad

[azure marsh]

This interaction has expired, use the command /guides realtime if you wish to see it again.

[thorny cargo]

lol

[hexed cave]

lol

[ornate hatch]

Oh, what did you use to make it?

[hexed cave]

I used

[rare goblet]

Ayo? @hexed cave level 7 !!!

[hexed cave]

-links

[azure marsh]

🖥️ Local stuff

• 🍏 Applio, by IA Hispano GitHub
• Mangio-RVC-Fork, by Mangio621 Huggingface
• RVC Studio, by SayanoAI Huggingface
• AICoverGen, by SociallyIneptWeeb GitHub
• Replay, by Replay Team Website
• Original RVC, by the RVC-Project team GitHub
• GPT-SoVITS, by RVC-Boss GitHub

Need some help? 🤔

You can find more info on the #1159513888199540817 channel. If you can't find your answer, feel free to ask for help in #✨│ai-help. Credits to Faze Masta and Antasma for compiling these links.

More commands: -audio, -colabs, -overtrain, -realtime, -rvc, -help

[hexed cave]

original RVC

for the conversion

[ornate hatch]

Oh, ty

[hexed cave]

mysep for the vocal and instrumental separation

cause I couldn't use uvr

[ornate hatch]

Alr ty. I already have them separted

[rare goblet]

Ayo? @ornate hatch level 5 !!!

[hexed cave]

be sure to use uh covers that are clean n dont have much background noises

thats what worked for me

n good luck c:

[ornate hatch]

Okayyy, thank you :))

[hexed cave]

[hexed cave]

btw the Hu Tao model isn't by me lol

[finite galleon]

Is this copyrighted

[hexed cave]

I took a cover of the song

and separated the vocals n instrumentals

[finite galleon]

Pretty sure you arent allowed to share ai covers with the instruments

[hexed cave]

ah alright

[finite galleon]

Because we dont want ai hub 3

[hexed cave]

thanks for letting me know c:

[ornate hatch]

@molten pecan donde descargo el audio?

[molten pecan]

te lo dice la guía

está en song_output

lee bien porfi 🙏

[ornate hatch]

Bueno, Gracias. Es que no me aparecia lo mismo

[molten pecan]

ups es verdad. escribí la guía mal. ahora la corrigo

[ornate hatch]

Ya no me aparece más 😦

[molten pecan]

ve a content

[ornate hatch]

Ah, okk no te preocupes

[rare goblet]

Ayo? @stuck ingot level 1 !!!

[ornate hatch]

Bueno, y dsps!

?*

[molten pecan]

a fix

[ornate hatch]

Oka

[molten pecan]

y ahí debería estar las carpetas de la app

[ornate hatch]

Ok, gracias

[molten pecan]

de nada :3

[stuck ingot]

hi, you know why I can't make a model a with rvc, and the other means are paid, could someone help me? I would like to make a model of RomanHorse.

[molten pecan]

https://aihubdocs.github.io/en/ everything you need to know is here

lmk if you need help!

[stuck ingot]

Thanks !!

[molten pecan]

yw :3

[ornate hatch]

Perdón, no encuentro fix

Perdón si t molesto

[molten pecan]

ah, pues no hace falta. estás en la carpeta base correcta. el audio de salida estarán en la carpeta song_output

output significa salida

(los resultados)

[ornate hatch]

Ok, clickeo en song_output y despues?

[molten pecan]

mantén presionado el primer archivo y luego toca Descargar

[stuck ingot]

I did everything as the tutorial said, but it didn't work.

[molten pecan]

please give more context

!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.

[stuck ingot]

Ok, basically, I did everything, I took extracts of my own voice, I did everything in order, but at the end when everything is "ready" it doesn't do anything at all, it doesn't nothing happens on the page (sorry for my english, i am French)

[molten pecan]

which RVC are you using exactly?

and in what step are you on?

[stuck ingot]

The Mangio-RVC-Fork

[molten pecan]

so, are you trying to make a voice model?

or apply a voice model to an audio?

[stuck ingot]

Make à voice model

[molten pecan]

alright

so, are you getting an error?

[stuck ingot]

Well actually, it doesn't give me any error message, or nothing happens, it's as if the button doesn't work

[molten pecan]

which button exactly?

[stuck ingot]

Train model

[molten pecan]

did you check the command tab?

the black one

[stuck ingot]

yes i cheked

[rare goblet]

Ayo? @stuck ingot level 2 !!!

[molten pecan]

and what does it say?

[stuck ingot]

but nothing is displayed

[stuck ingot]

Nothing

[molten pecan]

nothing at all?

no text?

[stuck ingot]

I do this on the phone I don't know if it changes anything

yeah

[molten pecan]

uhm you can't run Mangio on the phone

is only available on desktop

[stuck ingot]

eh, but yet it works, I think it's because I put the thing, display like on a computer

[molten pecan]

what's your GPU?

[stuck ingot]

Adreno 660

[molten pecan]

i'm talking about your computer's GPU

[stuck ingot]

oh m'y bad

[molten pecan]

RVC in general can only train in computers with NVIDIA GPU

[stuck ingot]

No, but I don't do that on the computer, since I'm not one, I do that on the phone

[molten pecan]

uhm well, you can't use Mangio locally on the phone. you'll have to use the cloud-based counterpart

use this instead

https://aihubdocs.github.io/en/rvc/cloud/training/rvc-disconnected/

[stuck ingot]

So, I can't create a model on phone?

[molten pecan]

technically you can, but not through utilizing its resources

only through a website, not using the hardware

[stuck ingot]

okay okay, but otherwise I can't see if someone nice could make me the model..

[molten pecan]

oh yeah ofc. here are two ways:

1. ask in #1159289738314919936. Keep in mind, with Paid Requests you'll get answers much faster.
2. check #1191429836321849435 and reach out to a Model Master

[stuck ingot]

i dont have money..

[stuck ingot]

i trying this

thanks you !!

[molten pecan]

alr then

yw!

[hexed cave]

can anyone see if its good? (The model I used wasn't trained by me) ofc I used mvsep instead uvr bc I'm incompatible with it

nvm

ignore my request

[molten pecan]

I recommend the cloud version of UVR instead. it's muuuch faster & convenient

[hexed cave]

alright

the hu tao toxic cover doesn't sound accurate

to the character at all

prob bc of the model im using

am I allowed to share a part of the AI cover without the instrumentals?

[molten pecan]

https://aihubdocs.github.io/en/vocal-isolation--datasets/vocal-isolation/#cloud-version- here's the usage guide

[molten pecan]

sure

[hexed cave]

aight

oh yeah the cloud version

I remember trying that and it went red from what I remember

[molten pecan]

it went red?

wdym?

[hexed cave]

like I got an error

its getting late for me

I'll explain tomorrow

[balmy trout]

so i wanna train a custom model using noUI

i assume i check the Colab section

[molten pecan]

that website is dead and outdated. this is the new one https://aihubdocs.github.io/en/rvc/cloud/training/rvc-disconnected/

hope that helps!

[balmy trout]

so rvc disconnected is the free way of training with collab without breaking the TOS

also question, i understand the latter 3 but what does the first one mean?

do users have a time limit?

[molten pecan]

yup. it's encrypted now

[molten pecan]

it has a usage limit of 4-6 hours for free users

[balmy trout]

well, uh does models usually sound nice with 4 hours of training?

[molten pecan]

well it depends

every model needs a different training time

[balmy trout]

like whats the most average time for AI-Voice training

[molten pecan]

but you can switch from account before you ran out of usage time

np

and also, after hitting the usage limit you'll have to wait 12 hours only

[balmy trout]

ah isee

[molten pecan]

if you scroll down in the doc, you'll see the "retraining" chapter

you can learn more there

[stuck ingot]

@molten pecan it's good ?

[molten pecan]

yeah it's going well

[stuck ingot]

thanks !!

[molten pecan]

remember to check the TensorBoard page if you want a high quality model

[stuck ingot]

okay no problem

I put 500 epoch is it good?

[rare goblet]

Ayo? @stuck ingot level 3 !!!

[stuck ingot]

I can go to sleep for a while 🥲

[molten pecan]

well, there isn't a way to know the right amount of epochs beforehand

check the Epochs page to learn more

it includes info regarding TensorBoard too

[stuck ingot]

what ?

"saving model and optimizer state at epoch 40" i stop it ? it's good ?

[covert turret]

That's to be expected if you told RVC to save a checkpoint

[molten pecan]

everything is explained in the doc. please read thoroughly before asking questions

[young estuary]

Someone know wtf with that error

All my models public

And worked fine

But now no

[covert turret]

which notebook are you using?

[young estuary]

That

[carmine oar]

C:\Users\User\Downloads\Mangio-RVC-v23.7.0_INFER_TRAIN\Mangio-RVC-v23.7.0>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7897
'runtime\python.exe' is not recognized as an internal or external command,
operable program or batch file.

C:\Users\User\Downloads\Mangio-RVC-v23.7.0_INFER_TRAIN\Mangio-RVC-v23.7.0>pause
Press any key to continue . . .

anyone having the same issue once you press any key it just closes?

[covert turret]

oof, I guess gdown isn't working properly anymore again

[young estuary]

oh shit

[rare goblet]

Ayo? @young estuary level 2 !!!

[young estuary]

so no ant other methods?

[covert turret]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces
• GPT-SoVITS (demo), by litagin Huggingface Spaces

More commands: -audio, -local, -overtrain, -realtime, -rvc, -help

[covert turret]

use applio or aicovergen ig

[young estuary]

There any guides for that?

[azure marsh]

🍏 Applio-RVC-Fork

Enhanced RVC Variant: Optimized Performance Through Modifications, Built upon Mangio-RVC-Fork.

[covert turret]

@young estuary there you go

[young estuary]

Thank you!

[covert turret]

не за что)

[carmine oar]

C:\Users\User\Downloads\Mangio-RVC-v23.7.0_INFER_TRAIN\Mangio-RVC-v23.7.0>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7897
'runtime\python.exe' is not recognized as an internal or external command,
operable program or batch file.

C:\Users\User\Downloads\Mangio-RVC-v23.7.0_INFER_TRAIN\Mangio-RVC-v23.7.0>pause
Press any key to continue . . .

anyone having the same issue once you press any key it just closes? i hear its a python issue but idk how to sett up python proporly for this.

[covert turret]

did you click on "add to PATH" when installing python?

that might your problem as it seems cmd cannot locate your python install

[carmine oar]

is the "add to path" something thats found with in the python site prior to downloading?

[covert turret]

no, it shows when you open up the installer

[carmine oar]

arlight, you think it woudl be best to uninstal and reinstal to find this option?

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

https://docs.aihub.wtf/v/brazil

🇪🇸 Spain

https://docs.aihub.wtf/v/spain/

🇫🇷 France

https://docs.aihub.wtf/v/france

🇩🇪 Germany

https://docs.aihub.wtf/v/germany

🇯🇵 Japan

https://docs.aihub.wtf/v/japan

🇰🇷 Korea

https://docs.aihub.wtf/v/korea

🇵🇱 Poland

https://docs.aihub.wtf/v/poland

🇷🇺 Russia

https://docs.aihub.wtf/v/russia

🇮🇹 Italy

https://docs.aihub.wtf/v/italy

[rare goblet]

Ayo? @carmine oar level 1 !!!

[covert turret]

i would reinstall just to make sure that's the issue

[split latch]

is go-realtime-gui_new.bat supposed to take long launching?

[lunar bolt]

Could someone listen to some short samples for me, not sure if I should include them in my model training

[thorny cargo]

@proud elbow I am literally snippeting the audio

little by little

since the model i made from this sounded so robotic and stuffy

like there was this lag

i did everything else by the book so im just looking back at my training data to see if that's the issue

None of my experience with stable diffusion training is helping here lmao

[proud elbow]

what training parameters did you use, and have you picked the best model checkpoint?

[thorny cargo]

let me double check

I am pretty sure i left it alone for those parameters

[thorny cargo]

I had exported and sampled the audio to be 48k to i chose that for the sample rate

version v1

pitch guidance on

ID for speaker 0. hit process data

[proud elbow]

this!

try 32k sample rate which is less likely to cause static glitches

[thorny cargo]

these were the only two

[rare goblet]

Ayo? @thorny cargo level 3 !!!

[thorny cargo]

idk why i don't got 32k

also im not sure what to do about these

[proud elbow]

select v2, or select v1 and then v2 in the UI

[thorny cargo]

This one is also within the guide but im sure there are other opinions out there

I set the batch size to 40 because of my 4090. however if this impacts quality in ANY way I do not mind modifying it

in SD having batch size low is actually a beneficial thing, not aware of how it applies here though

i trained the model multiple times. first time at 500 epoch with 8 batch

then 500 with 40 batch

then now im going for 2000

im just trying different stuff

😭

[proud elbow]

regardless of the GPU and VRAM, would recommend using batch size 8, esp practically for 5-10 mins dataset

[thorny cargo]

dataset is 27 minutes

was told that affects epochs

[thorny cargo]

would going lower increase the quality of the model?

or is it negligible

(doesn't matter)

[proud elbow]

would take longer time per epoch and may be less stable, but may be good for very clean dataset

[thorny cargo]

https://tenor.com/view/dumb-and-dumber-lloyd-theres-a-chance-jim-carrey-gif-11098164

ah

forgot

no embed

you get what i mean

after comparing the 500 epoch with 1250 epoch im actually noticing a little cleaner audio

yay

[proud elbow]

there's a rule of thumb to use power-of-two number as batch size tho (2, 4, 8, 16). 16 may be not bad but won't recommend going beyond even for your 4090

[thorny cargo]

thanks for the wisdom amigo

i been putting off making audio models for a long time

[haughty schooner]

How can I continue training a already finished model? Like I have a voice with 250 epochs, but want to put additional audio to the data face and train to 400. how can I do that? and another question: do you mix dialogue and singing voice in one database? I have some high quality studio vocals and wanted to put these into a database, in which I have around 20minutes of relatively clean dialogue. thanks in advance, have a great sunday!

[pastel oak]

Everytime you train a model it creates a D and G path in the logs folder I think it was

On the rvc training tab you just fetch the dataset like you usually would, but before hitting train button you load in the manual G and D path and then hit train

In the command line you should see it starting at 250 Epochs right away, if not you made a mistake

Idk about the second question

[jaunty shale]

trying to convert audio, it gets me this popup when I put audio in and start it. Its not even there for some reason
what do I do?

[brittle wing]

rvc uses the free 100 compute units in like 3 minutes

😔

[glad zealot]

wut?

colab?

[brittle wing]

yah

[glad zealot]

which colab notebook are you using?

lower the batch size

or using online options like google colab

[brittle wing]

https://colab.research.google.com/github/w-okada/voice-changer/blob/master/Realtime_Voice_Changer_on_Colab.ipynb

[rare goblet]

Ayo? @brittle wing level 1 !!!

[glad zealot]

yup that one is outdated

[glad zealot]

https://colab.research.google.com/github/hinabl/voice-changer-colab/blob/master/Hina_Modified_Realtime_Voice_Changer_on_Colab.ipynb

[brittle wing]

oh wait it is?

[glad zealot]

use this

[brittle wing]

it was the official link on the github

thanks mate

[glad zealot]

ye that modified one should also be there but that one is sort of the main version for colab atm?

as long as you are confident on how clean your dataset is, you can go to as low as 4 batch size

BUT needs to be clean

yup you can just lower it then

also loss/d isnt really that important, the 2 important ones are mel and g

as long as they are going down it should be fine

yu that looks fine to me

how long is your dataset?

[severe sand]

Hi quick question, should I delete bad slices from the logs after preprocessing or will that break it?

[glad zealot]

might need to clean it more probably? like you will not be able to hear anything at all other that voice, no echo also, no reverb also ye train for longer, might help

[glad zealot]

i dont think you should touch anything from it, you can just cut it before uploadeding the dataset

train it for longer than probably

you need to edit the code

its in the infer-web.py iirc

[severe sand]

No i mean I'm using mainline RVC and the preprocessing step puts out a few useless <1 second sounds. I would expect that to cause worse results, but maybe I will just try it with and without.

[glad zealot]

nah keep it in

[severe sand]

kk ty

Can I also get these graphs in mainline rvc?

[jaunty shale]

anyone got an idea what fcpe means?

[glad zealot]

not sure but iirc thats a feature extract made specifically for voice changer.. probably wrong tho

[jaunty shale]

if it's for voice changer then.. why in applio?

[glad zealot]

voice models are usable in voice changer

so whatever you train in applio will work on the voice changer

[jaunty shale]

I get this error again

I just updated applio

why does it do that?

[glad zealot]

ye not sure how local or worse applio works

so cant help with that

[jaunty shale]

the output file is not even there

like this is not there

okay so.. I figured out. I had split audio on. without it, it works.

Although, I use split audio for better results

[severe sand]

Am I seeing right that it picks training back up where it stopped last time? like when I do 50 epochs, then another 50

[rare goblet]

Ayo? @severe sand level 1 !!!

[jaunty shale]

okay, this is just getting annoying. It wasn't doing that before.

it doesn't create the output file.