✨│ai-help

[latent kettle]

It's free but requires your mobile phone number for verification purposes.

[woven anvil]

Bleh, oh well I'm desperate at this point.

[latent kettle]

Training locally will be a good option according to me.

[woven anvil]

Okay then I'll trust your expertise and we'll go for locally.

[latent kettle]

So there is no other option.

[woven anvil]

Lovely

Haha

[latent kettle]

So now you have to download RVC

[woven anvil]

If we are going to do this locally I'll have to set up the other PC, so I have to register a Kaggle account and Install RVC, which version?

[latent kettle]

If you want to do locally, no kaggle account is needed. But if you don't want to use any other PC, then use kaggle instead

For local Version, you have to download RVC, for kaggle you will get a notebook which you have to execute (no need to download anything in kaggle)

[woven anvil]

But you said local = better, so what version do I need to install on my other PC?

I came across a few.

Some I coulnd't figure out how to install.

That's how I landed on the one that opens Gradio.

[latent kettle]

There are many forks.

1. Mainline (original RVC)
2. Applio (modified RVC with advance features)

[woven anvil]

Are these two difficult to install?

[latent kettle]

No.

Applio comes in .zip you just have to extract it and it is ready to use

[woven anvil]

To give you insight into the walnut that houses my brain, this was already too difficult for me: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI

[latent kettle]

This is outdated BTW

[woven anvil]

No additional scripts to run and stuff just unzip and click to use?

[latent kettle]

Yes.

[woven anvil]

Oh had no idea 😅

[latent kettle]

Waana continue with Applio ?

[woven anvil]

Uhm is that the better one of the two or something?

[latent kettle]

What ?

[woven anvil]

I'll install whatever you think is best if its applio I will install that

[latent kettle]

Applio is good, user friendly. But both use same algorithms. So no need to worry about quality

For my personal choice, I use applio

[woven anvil]

Anything else setup related that I have to do?

[latent kettle]

No.

[woven anvil]

Cuz if it's just insttalling applio I'll focus on that and if its done and setup I will come back.

[woven anvil]

Okay I will set everything up, hopefully you're still around later 😬

May I tag/ping you?

[latent kettle]

https://huggingface.co/IAHispano/Applio/tree/main/Compiled/Windows

[woven anvil]

Thanks.

I will be back later then, the other PC is sitting on the floor so I first have to physically set everything up.

[verbal oyster]

can someone help me idk how to open rvc

[azure marsh]

Hey, son! Please use the command !howtoask to increase your chance of getting help by structuring your question in a way others can understand better. Also make sure you're asking in the right help channel:

• General RVC help: #✨│ai-help
• W-Okada / Realtime RVC: #🔍│help-w-okada
• AI image related: #🔍│help-ai-art

[verbal oyster]

please

[latent kettle]

Which rvc are you using

Mainline or applio?

If appio. Runapplio.bat

[verbal oyster]

nevermind i figured

but thanks

[crisp hound]

can i use applio as realt time

[latent kettle]

No. Applio is different thing. For realtime w-okada is best

[crisp hound]

thanks

[low shard]

Do u need help installing Wokada

[torn nebula]

hey i was installing the thingy here but like it's been so long and it still didnt do done any ways i can fix this?

it just stops

around here

and it doesnt load the voice changer itself

[hallow thistle]

For MMVC and W-Okada, go to #🔍│help-w-okada. This #✨│ai-help channel here is for RVC.

[torn nebula]

oh mybad

[latent kettle]

Hello everyone

[latent kettle]

I want to use applio from my pc to my mobile

On local host

I don't want a public link.

@hallow thistle what are you typing? It's too long 😪

[hallow thistle]

You can tweak some command line for Gradio to let it run on public link. However, one of the only ways to let Gradio to share link over LAN instead of internet is to plug LAN adapter from your phone to your PC, but I find this method to be impractical.

[latent kettle]

Bhot my pc and my mobile are connected to same network

[hallow thistle]

I think this way is a bit better than connecting your phone to a LAN adapter on the same network. But you might wanna do some coding a bit.

[latent kettle]

In app.py ?

[hallow thistle]

More likely.

[latent kettle]

Where should I add it

Please tell me @hallow thistle

@simple ore how do I access applio from my LAN devices

[simple ore]

edit run-applio.py and add --share ?

[latent kettle]

Then ?

I want that I can access it from my phone but without internet. My phone and PC are both connected to same network

[woven anvil]

Okay I think it's installed, I should be ready to go now.

[simple ore]

then you just run it and the gradio interface should be available from the network?

[woven anvil]

Does Applio need to be connected to Discord?

[wispy lodge]

You can change launch_gradio in app.py to the following, so it will allow network connections without creating gradio link:

def launch_gradio(port):
    Applio.launch(
        favicon_path="assets/ICON.ico",
        server_name="0.0.0.0",
        share="--share" in sys.argv,
        inbrowser="--open" in sys.argv,
        server_port=port,
    )

[woven anvil]

I need to type that into CMD?

[wispy lodge]

That was in response to other message

[woven anvil]

Oh oops.

How do I fix this?

Ohw Tensorboard doesn't work either : /

[low shard]

ye RVC and wokada are different programs

[latent kettle]

Don't want to use on network. Lemme explain. I'm sitting in Same room same house and we both are connected to same wifi (without internet) and I want to access my local host from my phone (as we do in apache server for web servers) without internet. How do I access Applio GUI in my phone. I mean I'll be the client but the Applio will be hosted by my pc in my own LAN network

[latent kettle]

Lemme try

[simple ore]

"from the network" = from the local network, if that's not clear

[latent kettle]

I want to access localhost:6969 on my mobile.

[latent kettle]

Okay. now how do I access it on my mobile without using internet.

My pc will act as server and I will be a client

[wispy lodge]

You need to check IP address of your PC and use it on mobile, like http://192.168.0.2:6969

[latent kettle]

Ohh that's what I want

A huge thank you 😆 you are the only who understand what I want to say

A huge thank you again

A question can I turn off my cellular data. Because my phone is connected with hotshot with my PC. If I do that, will it work?

[wispy lodge]

If you mean whether you will still be able to access Applio on your server, then yeah. It won't affect this in any way

[woven anvil]

Can someone help maybe?

[azure marsh]

Hey, ◂Maartje▸! Please use the command !howtoask to increase your chance of getting help by structuring your question in a way others can understand better. Also make sure you're asking in the right help channel:

• General RVC help: #✨│ai-help
• W-Okada / Realtime RVC: #🔍│help-w-okada
• AI image related: #🔍│help-ai-art

[woven anvil]

I think I'm in the right channel.

!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.

[woven anvil]

Here's also the screenshot for Tensor.

I got the advice to use Applio, downloaded it try to launch it but I get these two errors. I just want to create a voice model.

[latent kettle]

It works. Thank you. Can I normal use it for training and other things?

I have changed the code as you said

Will it impact on applio ? Do I have to change that code again and again?

[low shard]

it's not an error, that tensorboard will show data only after u actually start training

[wispy lodge]

No, this needs to be changed only once. If you update applio, then you'll need to apply this code again. Everything else should work as usual

[low shard]

did u actually start training after getting the dataset and putting the right settings

[woven anvil]

I don't know how I can't open Applio.

[low shard]

could u elaborate more

like what's in the cmd, and what's ur pc gpu

and which tut are u following

[woven anvil]

This is what I get if I try to launch it.

[woven anvil]

I'm not following one, tried to but in the first few minutes everything already looks completely different to what I have.

It was said that Applio is user friendly, but is there maybe something super duper user friendly maybe this still is just too complicated I don't know?

I have my voice samples clean and ready to go in WAV format, I just need a program where I can get to the training part. Was halfway there with colab but that one is broken now.

[low shard]

but is there maybe something super duper user friendly maybe this still is just too complicated I don't know?
Nope, AI is not meant to be as friendly as installing a Steam Game, it's computer science research still, also running it locally is harder on cloud
Ofc Applio tries to be as user friendly as possible
The literally EASIEST thing in the world would be maybe Weights.gg (cloud) or Replay (local version of weights.gg basically), but I have never tried Replay, but they both use RVC

[wispy lodge]

Try opening http://127.0.0.1:6969 in your browser

[low shard]

mmm it seems to be running on your browser actually if u open 127.0.0.1:6969, but I have never seen the issue above before, and wasn't able to search for it neither in this server nor IAHispano

not sure if that matters

maybe @simple ore knows to what it's related

[woven anvil]

YES!
Now it's open, do I need this link every time?

[low shard]

yes

however idk why it says that issue above, not sure if it matters

what's ur pc gpu?

[woven anvil]

Quadro P4000

Is there a recommendation you can give to how much training I should do?
Like is there a average min/max.

Last time I did 500, would that have been enough?

[low shard]

is it the 8gb vram one

[low shard]

there isn't

you gotta follow the tensorboard

there isn't any average

[woven anvil]

It should be 8

[low shard]

https://docs.ai-hub.wtf/rvc/resources/training/#epochs--overtraining

[low shard]

should be good enough iirc

[woven anvil]

Well let's see how far I get, at least now I have buttons I can press.
Thank you so much, Dr Jr thank you as well with your help earlier : )

[low shard]

yw

@ionic geyser this is the right channel to ask how to make ai covers

• Cloud (remote good pc, easier and faster than ur PC but it's limited):
  • Ilaria RVC Zero: fastest and simplest that you can get for free
  • Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
  • Applio Colab: max 4 hours, not granted, of GPU

Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero

I gave you only cloud ones since u showed me u got integrated graphics

[ionic geyser]

what are the ones for good pcs

[low shard]

u told me u got integrated graphics

it's useless to do it locally

else show me ur task manager

maybe u got gpu 0 and gpu 1

[ionic geyser]

locally is okada?

[low shard]

bc u said u got a gaming laptop

[ionic geyser]

how do i check that

i got gpu 1

[low shard]

what? wokada is realtime voice changer, it uses RVC but for realtime

RVC does AI Covers and train models

Locally = runs on ur pc
cloud = runs on remote good pc

[ionic geyser]

here look

[low shard]

show me the gpu 1

the gpu 0 is integrated which is not good, maybe gpu 1 is better

[ionic geyser]

@low shard

[low shard]

scroll a bit more down

I needa see the VRAM

the dedicated gpu memory

[ionic geyser]

@low shard

[low shard]

that's not that good

it's a old 3050, laptop gpus are weaker, and it got only 4gb of vram

[ionic geyser]

.okay it still works good for me

so what do i do

i run most programs good

i need to make music with ai

[low shard]

AI takes more computing, your PC isn't that good

if u really wanna do it locally:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC

but i would suggest u cloud

ur pc is gonna be kinda slow

[ionic geyser]

and how do i do it locally

is there a tut

whats the difference between cloud and local like

isnt it the same quality

[low shard]

no yt tuts

only written ones

I literally gave u the link, it's an hyperlink (blue text that are a link when u click them)

[ionic geyser]

is it hard to install

[low shard]

all yt tuts are old

[ionic geyser]

is it hard to install

@low shard

and whats the diff isnt cloud the same quality as local

[low shard]

same quality,
BUT:
Cloud = runs on REMOTE GOOD PC, meaning it willl be faster, but has limited gpu time
Local = runs on ur pc gpu, meaning it will be slower as ur pc gpu isn't that good

[low shard]

have Patience

i'm writing

😭

i can't type in a sec

[low shard]

u just gotta read, ofc this isn't downloading a Steam Game

[ionic geyser]

so no diff

even if u have a good computer

[low shard]

read it all, there's differences man 😭

don't just read the first 2 words

[ionic geyser]

the diff is itll run slow on my pc okay but itll run quick on a good pc just like a colab that id use

so theres no diff

[low shard]

it would be way better on a good pc, bc u aren't limited by a gpu time

and it would be faster

[ionic geyser]

but u dont need unlimityed gpu

u just get back on

later

[low shard]

u can be disconnected randomly on colab

[ionic geyser]

and u can make a song

[distant musk]

But with a GPU its WAY faster

[low shard]

btw this is only for converting voice

like AI Covers

it doesn't make songs like Suno

[ionic geyser]

if youre making a song you shouldnt expect to finish it fast

i wana make a song nick

[low shard]

it doesn't make the instrumentals

it can only convert the voice

[ionic geyser]

i know ill get the beats

[low shard]

like make an ai cover

[ionic geyser]

from yt

i just need to convert my voice

[low shard]

alr if u want to make ai cover it's fine

[low shard]

u want an ai cover right?

[ionic geyser]

i want to make my own cover with an ai voice

yeah

own song with ai voice

[distant musk]

[ionic geyser]

what

[low shard]

oh with your own voice?

[ionic geyser]

yes

i convert my voice

to a singer

[distant musk]

Uh

[low shard]

like, drake voice to yours right?

for an ai cover

[ionic geyser]

no like my voice to drake

[low shard]

alr

ye that works

[distant musk]

No. His voice to someone elses

[ionic geyser]

thanks

[distant musk]

Yea he said it already

[low shard]

u would prefer that, u can actually get randomly disconnected on google collab

bc it gives MAX 4 hours daily

of NOT GRANTED GPU

[ionic geyser]

4 hours is alot of time

for me

[distant musk]

Not for AI stuff

Its nothing

[ionic geyser]

well i can make a song

in that time

and if anything finish it the next day

done it before

[low shard]

alr ur choice then

[low shard]

^

[ionic geyser]

thanks

[low shard]

weights.gg is the easiest

[low shard]

yw

[ionic geyser]

it publishes the song

i dont rlly want that

[low shard]

no the songs are private

[distant musk]

It doesn't

[low shard]

unless u share them

[distant musk]

Its private

[ionic geyser]

ill use illaria theres advanced options

if theres one with tempo

like adjusting voice tempo

that would be amazing

bpm

[low shard]

with that, u also need to separate the vocals and instrumentals of the song manually https://docs.ai-hub.wtf/rvc/resources/dataset-making/#vocal-isolation--cleaning

[ionic geyser]

ill be making the song mysel

ill extract the voice alone and convert it

i wish there was a tempo adjustment for the voice to match the bpm like old collabs had

[low shard]

yes, that guide shows u how

[ionic geyser]

im getting gpu task aborted and i just started collab

what does that mean

[low shard]

show a screenshot

if you are using Ilaria RVC Zero, that's not colab....
that's hosted on HuggingFace Spaces, with a ZeroGPU

[ionic geyser]

so what do i do bro

[low shard]

show a screenshot

of when u get the error

[ionic geyser]

@low shard

[low shard]

don't ping me everytime

i'm testing myself

no need to ping

[ionic geyser]

oh thanks

[low shard]

the space was preparing, refresh the page

[ionic geyser]

whats better qualith that or apolio or is it same quality

[low shard]

Applio gets more updates,
But the quality should be the same between all RVC forks (Mainline, Applio & Ilaria RVC Zero)

there hasn't been a new model architecture since almost 2 years

[ionic geyser]

thanjs alot sir

[low shard]

yw

[ionic geyser]

thanks

[low shard]

do u need anything else or have other issues or can i mark this convo as solved?

[ionic geyser]

no thanks

[low shard]

so I can mark this as solved right?

[ionic geyser]

wait i downloaded the model into apolio but i dont see it when i click the model

voice models is empty

[low shard]

apolio
it's actually Applio

Could you tell me step by step what you did?

[ionic geyser]

i went to downalod

put the index and pth files

and i went back to where u use the auio to make a song

i cant find the model

[low shard]

did u click 'refresh models'?

[ionic geyser]

its empty

/content/program_ml/logs/model

it says its here

yeah i did

[low shard]

send me the model download link to be sure it's working

also show me a screenshot of ur Applio

[lavish lintel]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 97!

[ionic geyser]

https://www.weights.gg/models/clm734eri1laocctcdovppjdn

[low shard]

ye that's the model

it's put

it shows the path of the modle .pth

[ionic geyser]

do i put the index

[low shard]

not the name

[low shard]

yes

[ionic geyser]

how

its a download

i already put it when uploading

[low shard]

click on 'Index file'

[ionic geyser]

nothing shows up

[low shard]

btw all models on weights.gg are renamed to model.pth and model.index

the .pth is the actual voice, the index is the accent

[ionic geyser]

oh

well i dont see it

on index

?

[low shard]

u sure u uploaded it just like in https://docs.ai-hub.wtf/rvc/cloud/applio-colab/#inference-

[ionic geyser]

i downloaded the files theres no link

a hugging face link doesnt exist

[low shard]

ik, click the "manually"

did u put first pth, then index?

[ionic geyser]

i put both same time

[low shard]

u should put the pth first

THEN put the index

maybe retry that

@ionic geyser btw the quality between Applio & Ilaria RVC Zero should be the same, dw

it's still just RVC

and also Weights.gg AI Covers are private

you can use that too if u want, there are advanced settings

[ionic geyser]

the same ones or extra ones?
thanks alot btw bro!

[brittle wing]

What happened wuthRVC V3?

Over Trained? Or Keep on going?

Batch Size 4 + 30+ mins datasets + Pretrain KLM 48k X3 4.3

You replied the wrong person

[latent kettle]

@brittle wing I'm sorry to bother you

[brittle wing]

No problem

[latent kettle]

Seems over trained I guess

[brittle wing]

Damn that's a lot of overtraining...

[latent kettle]

RVC V3 ?

[brittle wing]

YEES

[latent kettle]

Do it exist?

[brittle wing]

It's in development

[latent kettle]

Oh wow, how do you know btw ?

[latent kettle]

Over training Started here

[latent kettle]

@brittle wing

[brittle wing]

#🔊│ai-development

[brittle wing]

Thank you.

I’m going to redo it because I do notice my dataset has a mixture of 40k and 48k.

I’m going to have to resample everything to 40k, maybe it won’t over train too much

[latent kettle]

Increase your batch size to 8

If you have dataset of 30+ minutes

[brittle wing]

I’ll keep that in mind.

Should I resample everything?

I noticed my dataset is mixed with 40k+48k using KLM 3 48k pretrainers

[patent hare]

Hey, how do we make models now that the google collab methods been busted?

[latent kettle]

Yes you can. Check your dataset sample rate in spek

[latent kettle]

Use kaggle

[drowsy eagle]

Gpu 0

Gpu 1

[latent kettle]

??

[pastel oak]

show name of the gpus

[pastel oak]

stick with gpu0

[latent kettle]

Why his GPU 0 is working even he have plugged in another GPU ?

[pastel oak]

i need context

[drowsy eagle]

so what can i use to train voice models with like programs?

[latent kettle]

Use your nvedia rtx 4060

[pastel oak]

https://docs.ai-hub.wtf/

rvc -> local -> mainline or applio

people prefer applio these days since it gets updates

[latent kettle]

Training or RTX 2060 is possible?

[low shard]

it doesn't exist

RVC v3 will never officially exist

there's just some EXPERIMENTS like codename fork

but it's not v3, the original creators don't care much about it anymore

[proven hill]

not even officially

[low shard]

you meant unofficially?

[proven hill]

yes sorry was about to correct

every fork is just a ui update

[low shard]

actually, @simple ore and @glacial pollen are working on updates

[pastel oak]

yea just not sure how well/fast

[low shard]

they added 44.1k sample rate and other things iirc

[proven hill]

i bet everything it will not change much

[low shard]

check #🔊│ai-development

[proven hill]

no access

[pastel oak]

no acess gamers

[latent kettle]

Someone Is asking me. Should I say him to train locally..

[analog obsidian]

tldr they changed everything, the gan, the discriminators etc

[low shard]

https://docs.ai-hub.wtf/rvc/local/codename_fork/

[low shard]

be sure to get the 'ai-testing' role from onboarding

[proven hill]

how big of a difference it makes? because its probably close to nothing

no shitting on others project its just that i dont see any real improvement

[low shard]

they are currently working on it, surely not 'nothing'

[proven hill]

well good luck

[low shard]

ye goodluck to them

i don't work on that

I think it will surely take a lot to make an 'rvc v3', but hey better than nothing

[latent kettle]

Is RVC V3 is coming?

[low shard]

let's see how it goes

[low shard]

no

officially it's 99% never coming

the creators left rvc to rot

they work on gpt-sovits

noobies and codename are working on an experimental fork

[analog obsidian]

pretrains are still being trained, but the early results they got is the fixing of mirroring frequencies
their project is still not finished tho

[latent kettle]

Gpt so vits. Text to speech thing ?

[proven hill]

i hope true pretrains and not fine tunes

[low shard]

yes

[analog obsidian]

true pretrains from scratch

[latent kettle]

Okay. Someone Is asking me 2060 is good for training with 6 gb vram ?

[proven hill]

no

[low shard]

ehh the 6gb of vram aren't the best

I think doable as @quasi dagger trained (super slowly) even on a gtx 1650

but ye he should be able tho

[latent kettle]

Okay. So using kaggle will be a good option?

[low shard]

even if ofc a better gpu would be suggested, like the min requirement in the docs is rtx 20 ti

[proven hill]

why not weights at this point

[latent kettle]

You are right. Weights.gg can be helpful too. I just forgot about that. Thanks ilaria

[proven hill]

no prob

[low shard]

he could try locally but I'm not sure how good that will run

anyways ye

ofc weights.gg is even easier than kaggle

his choice

[latent kettle]

You are no longer admin ?

[low shard]

ye since a lot of time

[proven hill]

since a year

[quasi dagger]

Even though it has 6gb VRAM I only managed to train at batch size 4

[low shard]

ye surely doable

[west summit]

Hi
I want to make song from ronaldo voice how can i make?

[low shard]

What's ur PC GPU?

[west summit]

rtx 3070

[low shard]

is it laptop?

how much vram (dedicated gpu memory) does it have

[west summit]

pc

[latent kettle]

Well what if I increase batch size to 8 ? Maybe bigger batch size = less vram consumption?

[west summit]

8 gb

[latent kettle]

Use applio or mainline RVC

[quasi dagger]

On Applio you can only set it to the maximum your GPU allows, so if you set to 8 it would go back to 6

[low shard]

good enough for inference

[west summit]

how to use it or anyguide?

[low shard]

Your Nvidia GPU is good enough

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• Cloud (remote good pc, easier and faster than ur PC but it's limited):
  • Ilaria RVC Zero: fastest and simplest that you can get for free
  • Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
  • Applio Colab: max 4 hours, not granted, of GPU

Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero
easiest local: Applio

I would suggest locally if u gonna use it more than once, if it's a 1 time thing, u can just use weights.gg

[latent kettle]

@low shard suggested you

[proven hill]

is ilaria rvc zero still untopped for inference?

[wispy lodge]

It's the opposite. Larger batch size - more vram is used. With 6gb vram, you can fit batch size 4-6

[quasi dagger]

6 it's very pushing the limits on a 6gb VRAM

So I only managed to do 4 batch size

[west summit]

thanks for you time
If i have any question in future i will ask

[low shard]

alr

[low shard]

wdym with 'untopped'? like speed?

[latent kettle]

Okay. So if 8GB vram can I set batch size to 10 or 12

[quasi dagger]

8

[latent kettle]

Yw : )

[proven hill]

like being easy

[low shard]

Weights.gg uses multiple A100s iirc, the only difference between it and ilaria rvc zero in inference, is that it's easier and it has queue

[low shard]

nah weights.gg is easier for that

[proven hill]

well yea tbh it has already the models

[low shard]

it automatically separates instrumentals and vocals

[proven hill]

but isnt it paid?

[low shard]

it's noob-friendly

[latent kettle]

Okay. I never knew that

[low shard]

it has a paid plan, but 99% of the things are free

except u got some little queue

that's all

[proven hill]

how can it be a sustainable business then

[low shard]

and on the paid plan it allows u to make videos

[low shard]

people don't like waiting + weights.gg is rich asf

[lavish lintel]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 98!

[quasi dagger]

Also skip queue for inference

[proven hill]

seems shady on my end

[low shard]

u can talk with River or Bea if u really think that

[proven hill]

i talked multiple times with river

[low shard]

I just know that they rich asf

like

[quasi dagger]

Not sure about the rich part as I saw they were cutting the free training a while ago due to the constraints

[proven hill]

but offering a free product with nothing back in returns means the user is the product

[low shard]

that was months ago, but it got added back

• lora training free too

• image creation free too

[proven hill]

how can that be free

[quasi dagger]

It has limitations for daily usage like other platforms like Leonardo ai for instance so it's not like 100% free 24/7

[low shard]

Queues

[proven hill]

still

[low shard]

yea true

[proven hill]

its a cost for them

[low shard]

I don't really think that's how Weights.gg is going, they pretty chill and have the money for it

else they wouldn't be giving away a free vr & nintendo switch #1317718319628943360

it's to get more popular i guess

[quasi dagger]

Yeah, so more people can go with the weights premium subscription

[low shard]

ye

[proven hill]

mh

[low shard]

if they were evil they would have already did the kits.ai way

[latent kettle]

Nintendo switch

[low shard]

it's not that hard to just shorten alot the limits

[quasi dagger]

Actually @proven hill weights is now owner of AI HUB

[low shard]

yup

[proven hill]

sadly

[latent kettle]

How will they choose a winner?

[proven hill]

closed source won again 😔

[low shard]

but they don't bc they care about the users here, last time people complained about no free trainings so they added them back basically

[quasi dagger]

Blaise stopped working on Applio as well

[proven hill]

mh mh idk its still strange

[proven hill]

i knew before everyone

[latent kettle]

Do Weights.gg own Ai Hub ?

[quasi dagger]

Technically yes

[proven hill]

blaise told me he wouldve stopped months ago

everyone simply found jobs and got tired of this

[latent kettle]

But how and why ?

[proven hill]

money and more money is the answer

jk

[low shard]

I remind everyone this is a help channel

[latent kettle]

🥺 seriously

[proven hill]

smh

[latent kettle]

Oops

[quasi dagger]

Btw you also have by Weights in your username

[low shard]

trying to revive the server since it's dead

[low shard]

events works atleast

even if we prolly need smt better

[latent kettle]

I just followed the trend. I thought people are doing this to be cool.

[low shard]

it's a meme of the name change

[quasi dagger]

No, they were mocking the server name change

[low shard]

as many complained about it before, and now I don't see anyone talk about that much

I just do it bc it's funni tbh

idrc if it's named ai hub or ai hub by weights or wtv

[latent kettle]

Even I asked you "should I change my name and add by Weights " you said yes

[low shard]

ye we were joking around lol

also let's talk in #✦│chat

[latent kettle]

I'm going to change my name again. I was unaware about that

[brittle wing]

Thank you

[low shard]

more than mocking, just joking

[night rune]

Is that good enough?

[simple ore]

go to Scalars tab

set smoothing to 0.95, do other charts (fm, mel) agree?

[simple ore]

scroll down

and expand losses

[night rune]

I don't know much about it (Tensorboard)

[simple ore]

how big is the dataset and batch size?

[night rune]

batch size 15
What do you mean by “how big is the dataset”?

[simple ore]

how many hours are in it

[night rune]

01:16
Minutes:Seconds

[simple ore]

i mean batch 15 is a weird number, suitable for 10+ hour sets

not for 75 seconds of audio

[night rune]

How much is it supposed to be?

[simple ore]

2 or 4

[night rune]

f*ck

[simple ore]

there's way too much variety between steps on those charts

[night rune]

So, what's the best one to put in? 2 or 4?

i'll set 3

[simple ore]

set it to 4

[night rune]

jk

ok*

another three hours of training

let's go

[fading rock]

hi dose anyone know what software is normally used to make rvc models

[night rune]

And how often do I need to save the models?

[simple ore]

10 is fine, you'll be able to pick something out before runs into a wrong direction

[night rune]

ok, thanks

[night rune]

Btw, what is batch size exactly?

[simple ore]

75 seconds of audio = 25 samples x 3s + 2 mute files = 27 samples

27 samples / batch 4 = 7 batches (=steps) per epoch

[low shard]

RVC, what's ur pc gpu

[simple ore]

the training step takes a batch (4 samples), tries to reproduce them, gets evaluated, re-adjusts parameters, repeats

[night rune]

omg

[night rune]

Okay, I think I got it, thanks again

[simple ore]

batch size dictates how big the adjustment is going to be

big variety of samples = bigger the variance between steps

that's why you had crazy 20-45 range for some metrics

at the end of the epoch the learning rate goes 1 notch down (=training speed decreases), samples get reshuffled between new batches and the training contuniues

so it is a balance between having enough adjustment per step vs constantly lowering learning rate

[night rune]

I hate it when that happens

Is there any way I can get this epoch 215?

[simple ore]

dont look at this output

it is just generator loss without fm/mel

[night rune]

Where am I supposed to look then?

[simple ore]

look at the charts

https://docs.applio.org/applio/getting-started/tensorboard

[azure marsh]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab

ℹ️ Note

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

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

[night rune]

Now I have this

[simple ore]

hard to choose, check one close to 1600 steps

[balmy estuary]

Starting training...
Loaded pretrained (G) 'rvc/models/pretraineds/pretrained_v2/f0G48k.pth'
Loaded pretrained (D) 'rvc/models/pretraineds/pretrained_v2/f0D48k.pth'
Process Process-1:
Traceback (most recent call last):
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\rvc\train\train.py", line 497, in run
reference,
UnboundLocalError: local variable 'reference' referenced before assignment
An error occurred extracting the index: need at least one array to concatenate
If you are running this code in a virtual environment, make sure you have enough GPU available to generate the Index file.

how to fix

[night rune]

I think I'll wait for the full training first (500 epochs) and then I'll start choosing

[simple ore]

make sure you actually sliced the audio and not trying to train on 1hour long wav

[balmy estuary]

i did

[simple ore]

show log for extract features step

[balmy estuary]

i'll have to redo it since i relaunched it

one sec

[simple ore]

and/or filelist.txt content

[balmy estuary]

wtf

the filelist is empty?

should i restart from scratch?

[simple ore]

what is in sliced audios / 16k

[balmy estuary]

alot of wavs

481 of them

[simple ore]

are there other folders?

[balmy estuary]

i sliced it with applio

[simple ore]

and what is in f0 and v2_extracted

[balmy estuary]

nothing

[simple ore]

run extract features then

and check the log

[balmy estuary]

Starting pitch extraction with 3 cores on cuda:1 using rmvpe...
Process Process-1:
Traceback (most recent call last):
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\rvc\train\extract\extract.py", line 118, in process_files
self.model_rmvpe = RMVPE0Predictor(
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\rvc\lib\predictors\RMVPE.py", line 460, in init
).to(device)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 1173, in to
return self._apply(convert)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 853, in _apply
self._buffers[key] = fn(buf)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 1159, in convert
return t.to(
RuntimeError: CUDA error: invalid device ordinal
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Pitch extraction completed in 6.72 seconds.
Starting embedding extraction with 3 cores on cuda:1...
Process Process-2:
Traceback (most recent call last):
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\rvc\train\extract\extract.py", line 179, in process_file_embedding
model = load_embedding(embedder_model, embedder_model_custom).to(dtype).to(device)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\transformers\modeling_utils.py", line 2905, in to
return super().to(*args, **kwargs)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 1173, in to
return self._apply(convert)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 779, in _apply
module._apply(fn)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 779, in _apply
module._apply(fn)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 779, in _apply
module._apply(fn)
[Previous line repeated 1 more time]
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 804, in _apply
param_applied = fn(param)
File "C:\Users\iibug\Downloads\ApplioV3.2.8-bugfix\env\lib\site-packages\torch\nn\modules\module.py", line 1159, in convert
return t.to(
RuntimeError: CUDA error: invalid device ordinal
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

i'll delete this after since so big

[simple ore]

wrong device selected?

[balmy estuary]

wym?

anything wrong with this?

[simple ore]

should be 0 to match your 2060

gpu number that is

[balmy estuary]

IT'S EXTRACTING

IT COMPLETED

what do you think would be good for batch size?

it's at 5 rn

[simple ore]

it does not have much to do with VRAM.. other than "will batch x fit into 12GB"

so you got like 20 min audio there

[balmy estuary]

so what do you think would be good for my card?

yes

[simple ore]

use batch 4

[balmy estuary]

do i gen index before or after training?

[simple ore]

it runs automatically, but you can generate it after you start the training

[night rune]

what do you think?

[simple ore]

click the square button under each chart to adjust the fit

[night rune]

right or left square?

[simple ore]

\

[simple ore]

they do look rather terrible, but not much can be done with such small set

you may have better luck using a non-default pretrain like ov2

[night rune]

What is pretrain?

Okay, I found where to read about it.

[night rune]

“works well for small datasets and English language” this line alarms me, as my model is trained in Russian language

[simple ore]

it is okay, just dont use index and you'l have almost no accent

[night rune]

By the way, what does the index affect?

[simple ore]

mostly prononcuation.. english audio + voice model + russian index (1.0) = russian accent

[night rune]

Which version should I download?

[simple ore]

depends on your audio

[night rune]

Ok, I downloaded it all, but I'll probably try it during the day since I'm deep in the night now

[night rune]

Good night and thank you for helping me

[sudden tree]

how can i update rvc if I have an old ancient version easily?

[cloud finch]

    raise LocalProtocolError("Too little data for declared Content-Length")
h11._util.LocalProtocolError: Too little data for declared Content-Length```

anyone know what this error means?

I just dragged in my audio and its outputting this

[knotty moth]

another way to do is to put the input audio in your Applio\assets\audios
or if not sure, try download the latest applio 3.2.8 bugfix

[cloud finch]

Thats what i tried doing

Ill try getting the bugfix

[terse nymph]

https://docs.ai-hub.wtf/rvc/resources/vocal-isolation/
^ this page is down, anyone know what UVR models were recommended for clean vocal extraction?

[sudden tree]

rvc

[hallow thistle]

Their server is up now.

[sudden tree]

how do i update my old rvc build from july 2023

i want rvmpe

do i have to wipe the whole folder and start over?

[hallow thistle]

If you mean the original RVC GUI, this one has been long outdated. The most recent one released around October 2023.

[sudden tree]

oh, can i update it to that one?

I want to use rmvpe models instead of crepe

isnt rmvpe way better?

[hallow thistle]

RMVPE is way better than any other model.

Unless you wanna make your own fork of RVC GUI by yourself, there you can download its source code from GitHub. Otherwise, you can try Applio the RVC GUI to do some audio conversion.

[sudden tree]

ok but wont the latest rvc gui work regardless?

how can i update mine to the newest

idc if its old

[hallow thistle]

The most recent RVC GUI indeed works, but there won't be any official update after that.

[knotty moth]

rvc gui is too old and doesn't have rmvpe which gives better quality
I recommend the much newer one, i.e. Applio

[bold epoch]

hello

sir

i need useful link

[hallow thistle]

For RVC GUI, you'd find some bugs going on.

[bold epoch]

im coming from a video

[knotty moth]

-gui

[azure marsh]

https://cdn.discordapp.com/attachments/1122285248844144733/1203460490475343953/caption.gif?ex=65d12cec&is=65beb7ec&hm=bd2fb8d010006dd7c6e3c1c67d3ae846fd1478e1a3124c544c31b43086fe54aa&

[hallow thistle]

Useful link of what?

[bold epoch]

to create

rvc ai cover

im on mobile

[hallow thistle]

-rvc

[azure marsh]

📚 Documentation

AI HUB Docs

https://docs.ai-hub.wtf

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

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

[hallow thistle]

If you're on your smartphone, a cloud service is the best and only way to do RVC AI cover.

-colab

[azure marsh]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab

ℹ️ Note

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

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

[hallow thistle]

Google Colab is a cloud service that lets you run Python there. First six links are RVC, but I'd recommend for Applio.

[bold epoch]

i need to install

applio ?

in codes saying install applio

[hallow thistle]

Install Applio there on Colab? Yes.

[bold epoch]

okay lemme test

[sudden tree]

Oh I didn't realize that thank you. Also applio just scares me since it's closed source

[low shard]

Applio isn't closed source

It's open source

https://github.com/IAHispano/Applio

What's ur PC gpu

[bold epoch]

im on ios

sir

[hallow thistle]

RVC, W-Okada, Stable Diffusion are all open source. Unless you've downloaded a proprietary program that does audio conversation similar or like RVC, which is indeed closed source.

[oak edge]

can someone tell which sample rate is good for voice accuracy

[latent kettle]

The sample rate of your dataset

[latent kettle]

IOS >> apple iPhone?

[sudden tree]

hey how come the official uvr 5 new model announcement uses some weird github release not the official anjok07 one?

like eddiecrack somethng

[hexed hinge]

how can i avoid overtraining? using weights.gg

[viscid moss]

My UVR5 UI use audio-separator a CLI-type python package based on the same code as UVR5

[sudden tree]

i see

so you just made a webui version

i decided to just use anjok

how can i seperate main and backing seperately with melband roformer

[viscid moss]

ye i think, although audio-separator has some extra things, likewise UVR5 has other extra things

[sudden tree]

yessir

does anyone know if you need to split inputs to train?

i put 2 10 min files in

uncut

[viscid moss]

Mel-Roformer-Karaoke-Aufr33-Viperx could be useful for that, or BVE models on VR Arch

[viscid moss]

train RVC model?

No needed

RVC do that automatically

[sudden tree]

i only see kim and viper melband roforomer on uvr

how do i find this Mel-Roformer-Karaoke-Aufr33-Viperx

[viscid moss]

That's one of the disadvantages of UVR5, you have to search and add the models manually, and change the configuration files if necessary

[viscid moss]

https://mega.nz/file/qQA1XTrb#LUNCfUMUwg4m4LZeicQwq_VdKSq9IQN34l0E1bb0fz4

[sudden tree]

so what is better in your opinion kim or viper roformer

[viscid moss]

For vocals? I think Kim

[sudden tree]

ok thanks

yeah, i still found kim model in official download section of uvr 5 tho @viscid moss

i wonder what model mvsep uses?

[viscid moss]

lemme see

gimme a sec

I think is the same

https://github.com/KimberleyJensen/Mel-Band-Roformer-Vocal-Model
This model

[sudden tree]

yes, did kimberly really make this herself?

[viscid moss]

idk

[sudden tree]

thats crazy it sounds so clear lol

[viscid moss]

fr

[sudden tree]

one more question where can i find the melband roformer karaoke model?

it seems to be on mvsep but I would prefer to download it locally iykwim @viscid moss

[latent kettle]

How can I download new UVR5 UI ? Do it have any compiled version or I have to do everything manually on CMD

[brittle wing]

is there a way to train on mac yet?

or is there any service which offers me online training with individual settings & unlimuted amount of dataset?

[sudden tree]

lol where did you get this from? I am nervous to download random files?

is it from their huggingface?

[viscid moss]

Here
https://mega.nz/file/qQA1XTrb#LUNCfUMUwg4m4LZeicQwq_VdKSq9IQN34l0E1bb0fz4

[viscid moss]

Nope, they don't publish that model on HF idk why

[viscid moss]

Not yet, instalation is through cmd with an installer i made

https://github.com/Eddycrack864/UVR5-UI?tab=readme-ov-file#getting-started

I recommed cloning and install it on ur Downloads folder

[latent kettle]

When will the .exe will be available?

[viscid moss]

Not sure, I'm not specifically working on that

If u need help to install it lmk

[latent kettle]

Okay. Which models are best for vocal isolation, de eco, de reverb and de noise?

[latent kettle]

Not now, my wifi is not functioning.

[viscid moss]

best extraction method for vocals: unwa's big mel roformer beta 4
de-reverb: anvuew mel dereverb v2
backing vocals: mel roformer karaoke
denoise: mel denoiser

[low shard]

Yeah nvm

• Cloud (remote good pc, easier and faster than ur PC but it's limited):
  • Ilaria RVC Zero: fastest and simplest that you can get for free
  • Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
  • Applio Colab: max 4 hours, not granted, of GPU

Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero

[latent kettle]

And do you know how to remove stereo delay ?

[low shard]

That uses an automatic overtrain detector

[viscid moss]

Hmmm idk

[latent kettle]

I was using MDX23C for vocal isolation, Vr ach- deEco , MDX de reverb normal and UVR denoise is that bad?

Should I remove my old UVR and Go with New one ?

[viscid moss]

MDX23C ye, others are kinda fine

[viscid moss]

Def yes

[latent kettle]

What's wrong with MDX23C ?

[viscid moss]

Low SDR

[latent kettle]

Can you show me how new UVR look like

[viscid moss]

Roformers are better

[viscid moss]

Beta version is identical just code changes that u can't really see

[latent kettle]

SDR meaning ?

[viscid moss]

UVR5 UI is different cause is gradio UI

[latent kettle]

Okay. If I remove old one and follow the link you have sent. Will it install everything automatically?

[latent kettle]

Ohh I see. Just like applio and mainline

[viscid moss]

Yep

[latent kettle]

Will it remains like this forever?

Or it will get a GUI ? Just as old UVR 5 ?

[viscid moss]

I need to install 3 things, Git (to update UVR5 UI), FFmpeg and UVR5 UI. All that is described on the link i send u

[latent kettle]

Okay. What about models?

[viscid moss]

Models will be downloaded automatically when u choose them

[latent kettle]

Okay got it

Thank you

[viscid moss]

Ur welcome

[ionic geyser]

@low shard how did the voice model i send work for u yesterday

[low shard]

What issue are u having now

[night rune]

Okay, I finished training and these are the results

[simple ore]

compare model from 1600 and 4200 steps

both are likely complete and utter trash

[night rune]

wym? What should I do or look at?

[simple ore]

just infer a test audio with those and hear the output

[onyx ruin]

Seeking Help with Climate Data Project!
I’m working on a project that uses synthetic data generation to address gaps in climate datasets and improve predictive modeling for climate change mitigation. Here's what I aim to achieve:

Generate synthetic climate datasets (Avg Temperature, CO2 Emissions, Renewable Energy %) using methods like Gaussian Mixture Models (GMMs), Variational Autoencoders (VAEs), and Copulas.
Validate synthetic data with metrics like Wasserstein distance and Kolmogorov-Smirnov tests to ensure consistency with real-world patterns.
Visualize trends, anomalies, and regional variations using ArcGIS/QGIS and traditional plots.
The goal is to create scalable, reliable datasets that aid in better climate analysis and decision-making. I’d love your inputs, advice, or suggestions on how to improve this framework or achieve these goals more effectively.

If you’ve worked on synthetic data, GIS tools, or climate modeling, I’d really appreciate your help! Let’s brainstorm

[lilac yacht]

can someone help me i have dragged the spongebob voice ai zip and it says that i cant upload files like that

[azure marsh]

Hey, djgyrtazao! Please use the command !howtoask to increase your chance of getting help by structuring your question in a way others can understand better. Also make sure you're asking in the right help channel:

• General RVC help: #✨│ai-help
• W-Okada / Realtime RVC: #🔍│help-w-okada
• AI image related: #🔍│help-ai-art

[lilac yacht]

who are you can you help me

spongbob

!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.

[lilac yacht]

i cant put screenshot for spongebob

[sudden tree]

Oh this isn't it? https://huggingface.co/jarredou/aufr33-viperx-karaoke-melroformer-model/tree/main

[low shard]

are you talking about https://voice.ai ?

it's trash, we don't support that

it abuses ur pc power for distributed training

or are u talking about smt else

[lilac yacht]

si

e voice ai

[low shard]

so, u are using the program I said?

oh sei italiano

[lilac yacht]

sono italiano anche io

si e quello li

[low shard]

voice.ai fa schifo, abusa del potere del tuo pc in background

disinstallalo fra, lo usavo anche io prima

[lilac yacht]

lo so ho letto

e che uso

[low shard]

Wokada, specialmente il detieris fork

[lilac yacht]

??

e come voice

[low shard]

parliamone in #🔍│help-w-okada

[lilac yacht]

ai

[low shard]

si

[lilac yacht]

ok

[cloud mantle]

@low shard

[low shard]

yea u need to put the huggingface download link url in that

what model link are u having issues with

[viscid moss]

oh, ya seems like

[cloud mantle]

generally, I need a new model for my song remake and I'm checking what the options are here

[low shard]

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @earnest musk
• https://weights.gg/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• #1159289738314919936
• #1191429836321849435
• make it yourself with our docs guides https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/

[earnest musk]

:wave: @low shard, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[cloud mantle]

I thought you could paste a YouTube link there and it would create a model itself

[low shard]

what? No model training doesn't work like that

Ilaria RVC Zero is made only for Inference (use models) not training (make models)

on RVC you need to download, curately clean the dataset, then train it looking at the tensorboard

it's AI

[low shard]

the only way to do it that ez, is on https://weights.gg

no other way to do it that easy

[cloud mantle]

I used a site outside of this discord and it worked normally but this site didn't have the features I needed

[crude flame]

if only it was that easy 😔

[low shard]

RVC isn't a site, it's a program

You prolly used smt like kits.ai, which still uses RVC in the background on their cloud pc,

AI isn't installing a Steam game

[ornate ore]

I have a question when it comes to RVC. I want to change out my 5700XT for a 4080S Suprim X but I'm using a i7-11700F CPU. Will the CPU affect RVC when it comes to my voice latency or no?

Idk if I need to upgrade my CPU to reduce latency or will all that work be using the GPU primarily?

[crude flame]

cpu has pretty much nothing to do with rvc

its all gpu

[ornate ore]

Okay,perfect

[low shard]

@cloud mantle the only way to do it that easy, is by using weights.gg

what's wrong with weights?

just use since it's the easiest free way, it uses the same program anyways

[cloud finch]

im going to strangle this error

I dont understand

[cloud mantle]

only when generating a song the AI doesn't get into the rhythm well and sings in a different key

[zinc crow]

Ciao, buonasera.
Sono nuovo in questo forum e in questo "mondo" AI...

Dove posso chattare (o aprire un ticket) con uno Staffer?

Possibilmente (se possibile) uno staffer italiano poichè non so altre lingue

[proven hill]

LMFAO

[proven hill]

@low shard

[zinc crow]

Ciao, grazie cara. Gli scrivo in privato o qua?

[proven hill]

aspetta che risponda

cosa devi fare?

[zinc crow]

Volevo sapere come si creava una Singing Voice Cover, prendendo una canzone, lasciando la base così, e cambiando le parole... Vorrei farlo con la voce di Lazza, perchè ho visto che c'è un'anteprima ma non so come fare per farlo cantare ugualmente all'originale.

[zinc crow]

Per il momento ti ringrazio tanto

Attendo sue allora

#1287422682446565578 message

[cloud finch]

DOES ANYONE KNOW HOW TO FIX THIS

ITS BEEN DOING THIS FOR 2 DAYS

[low shard]

cambiando le parole giusto? è più difficile di un AI Cover normale, perchè dovresti:

1. scaricare la canzone
2. separare le vocali e la base
3. cantare la versione con le parole modificate
4. Usare RVC per cambiare la tua voce in quella di Lazza
5. Mettere insieme tutto quanto

o intendi un AI Cover normale senza cambiare le parole?

[low shard]

what's ur pc gpu and what are you using

[zinc crow]

Ciao grazie della risposta. Intendo come ha fatto qua [#1287422682446565578 message] il 3 audio. Però con un altra canzone di lazza. Poi la base la ricollego.

[cloud finch]

NVIDIA RTX 3060

[low shard]

penso che lui abbia fatto cosi, oppure si può usare il TTS (Text To Speech) (da testo a voce) ma non è brava nel cantare

[cloud finch]

Intel i7

[low shard]

mm, what softeware did u download

[cloud finch]

Applio

[low shard]

you might wanna check if your drivers are updated

and that u are running on the latest applio version

[low shard]

what's ur ram, and vram?

[zinc crow]

esiste che tu sappia un tts singer che gli puoi dare degli input (es: canta come questa canzone), però con il testo che scrivo io?

[cloud finch]

The Latest

[low shard]

non credo che esiste, forse Suno o Udio ma non li uso

[low shard]

I found this #✨│ai-help message , maybe that works

[low shard]

yea it's enoughù

[cloud finch]

I did that, didn't work

[low shard]

so, u used that and are 100% sure ur drivers are updated?

[zinc crow]

non hai una piattaforma testata da te che mi puoi consigliare?

[cloud finch]

I'll try updating them but its wierd cause the issue only happens when the file isn't small

[low shard]

sinceramente no, o puoi provare con il metodo difficile con RVC, o puoi testare se Suno/Udio vadi bene

ma di solito non modifico il testo delle canzoni

facevo solo ai cover prima che le eliminassi

[low shard]

what's the vram (dedicated gpu memory) of ur 3060

and is it laptop

[cloud finch]

Its not a laptop, and I dont know the Vram sorry

[zinc crow]

ovvero? che cambia?

[low shard]

show me a screenshot when going into gpu 0 or gpu 1 in task manager

[analog obsidian]

this occurs when starting training?

[cloud finch]

It occurs when I convert it to ai, im not trainging

training*

[zinc crow]

Ho provato già Suno e Udio l'ho visto ora ma Udio non fa caricare audio nella versione free, mentre Suno capisce nulla e rileva le canzoni di altri artisti perciò non va come input.

[low shard]

che semplicemente potevo cambiare la voce dandogli come input le vocali della canzone, ma nel tuo caso o le canti tu o usi un TTS (che cantera male di solito)

[analog obsidian]

have you tried multiple files or only the same one?

[proven hill]

@zinc crow il metodo migliore è cantare tu stesso, poi fare il cambio dalla tua voce a quella di lazza (se ho capito bene) e poi metterci sotto la base

[cloud finch]

Only that one

[analog obsidian]

try other one

[cloud finch]

alright

hold on

[low shard]

mmm, forse potresti chiedere anche a @civic meteor , ma al 99.9% lui avra fatto come ti ho spiegato io con RVC

cioè cantando

[cloud finch]

I tried a differen't one and it worked

the one Ive been trying to do is 30 minutes

and the one that worked was 19

Is my file too long?

[analog obsidian]

which pitch extractor are you using? rmvpe or crepe? which format is the audio you want to inference?

is it wav?

[cloud finch]

Wav, rvmpe

[analog obsidian]

is "split audio" option in the inference being used?

[cloud finch]

yes

[analog obsidian]

alr i believe the audio file might be too long and you're probably getting out of vram

[cloud finch]

Really?

It happens at the very start

[analog obsidian]

open task manager and check your main gpu vram usage (the 3060)

[cloud finch]

this one?

[analog obsidian]

now try to inference the audio that gives you that error

yup

[cloud finch]

okay

so if it reaches 12

thats bad right?

[analog obsidian]

yea it means you're getting out of vram

[cloud finch]

highest it got was 11.2

So, i dunno if thats it

[analog obsidian]

doesn't look like a oom problem
might be related to how applio handles very long* audio files

only an applio dev can really help you in this

[cloud finch]

https://tenor.com/view/cry-izuku-midoriya-boku-no-hero-academia-angry-sad-gif-5928195

[analog obsidian]

@simple ore he has been trying to inference an 30 minute long audio file in applio but he gets cudnn non contiguous input
he tried another audio file and it worked just fine
what do you think its going on here? he's not getting out of vram either

split audio option is checked

[simple ore]

does split audio actually split it into multiple chunks? or does it not find an appropriate silence gaps to separate?

[analog obsidian]

he gets cudnn error as soon he tries to inference that specific audio

the other audio he tried was 19 minutes instead of 30

[simple ore]

so what's in the log?

does it say 'split into x chunks'

[analog obsidian]

can you try to inference your audio file again but this time show us what do you get in the cmd window?

the one that gives u the cudnn error

[analog obsidian]

@cloud finch

[simple ore]

normally with 12GB you can infer ~15min audio

[analog obsidian]

oh so it was indeed a vram error

i suspected that first

[brittle wing]

i need help with rvc

[hot belfry]

whats a good free rvc?

[brittle wing]

i just installed it again after long time, i started go-realtime-gui and it opened but when i click start it crashes, no not really it just gets stuck and dont work then

[low shard]

You shouldn't use the go realtime for realtime voice changing

[brittle wing]

what else then?

[low shard]

Wokada deiteris fork (more optimized) > Wokada > RVC go-realtime

[low shard]

Wokada deiteris fork, let's talk in #🔍│help-w-okada

[brittle wing]

ok G

[low shard]

RVC is open source and free

Tell me ur PC GPU and what are u looking for

RVC = Retrieval-based-Voice-Conversion, NOT Realtime Voice Changer

[red cliff]

Anybody know exactly what the OG pretrain is? The devs say VCTK but I tried training my own with VCTK as a test, and while the output sounded decent, it came out more muffled/less high-end-y than the OG pretrain.

also the RVC devs say is the readme that they will be adding more data to the pretrain so I can't tell if they did or not.

[simple ore]

it is 109 speaker set using vctk corpus 0.80 or 0.92 mic1

[red cliff]

Ah I wonder if I included other mics

[exotic timber]

help pls

i have the model downloaded from this server but idk what to do know

[bleak pasture]

Hi, where can I train RVC voice models? I use Android. Can someone tell me where I can train voice models?

[glacial pollen]

Unfortunately, if you intend to use the Android device itself, that's not possible. As for which websites ( such as weights.gg and such ) support android devices, I've no clue

[hallow thistle]

If you want to train a voice model using Termux on your phone, no. But you can go for a cloud service like Google Colab or Weights instead.

[low shard]

!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.

[bleak pasture]

well thanks

[glacial pollen]

Try weights website I suppose

if it doesn't work in mobile mode, switch to desktop mode ( if still doesn't, rip. Maybe try google colab but again, I never needed to use it so don't know if it works

[bleak pasture]

Ok, and what collaboration would be Google and Weight? I was not told that on that page the voice models do not give good results when they are trained there.

[hallow thistle]

When did I said there was a collaboration between Google and Weights?

"Colab" and collaboration aren't the same word.

-colab

[azure marsh]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab

ℹ️ Note

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

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

[glacial pollen]

^

[hallow thistle]

First three links are RVC Colab notebooks that can do training. But I'd suggest to go for Applio as it has an easy GUI to use.

[bleak pasture]

and which ones are for training models?

[hallow thistle]

For Weights, I suggest it would be better to have a desktop PC so you can do drag and drop audio file into Weights easily.

[hallow thistle]

Please read my message above.

[bleak pasture]

The applio one doesn't work, it doesn't open

[hallow thistle]

Can you send a screenshot here when it won't launch?

[bleak pasture]

But I don't have a PC, I live in Latin America where everything is horrible. I only have Android, otherwise I wouldn't be asking how I train from Android.

[hallow thistle]

Again, can you screenshot where it won't launch? Just so I can verify about issue that made Applio not working.

[bleak pasture]

And which of those that are there are useful for training?

[zenith hedge]

Was this ever solved?

[knotty moth]

even if you have a 5090 and 128 GB ram, it is strongly recommended to split into at most 6 mins each, or using applio's split audio feature
also before using split audio feature, you should denoise and noise gate the input audio first, as there couldn't be a voice sentence longer than 1 minute without silence gap

[old yarrow]

does anyone know like if theres any ai that can find a song because websites dont wok for me

im having trouble finding this song and its not known so idk if anyone can help so is there like s ong finder ai or anything

[simple ore]

regular shazam dont work?

[old yarrow]

no

[simple ore]

upload it to soundcloud, if it is copyrighted you'll know soon 🙂

[old yarrow]

its unreleased and rlly hard to find so idk if it will get copyrighted

[cloud finch]

Wierd thing is though is that the other 2 files I do were around the same length, 29 -32 minutes

and they worked fine

[cloud finch]

But it only happens with this specific file

its so wierd

[glacial pollen]

Try to cut the file into 3 more-or-less even parts and infer each separately ( without slice audio option )