✨│ai-help

[brittle wing]

So I can apply the effects by myself afterwards

[red kayak]

yes u could

[brittle wing]

Of course effects can make it sound realistic

[red kayak]

yeah

[brittle wing]

Cause I use music speed changer

I put the cover I mixed under vocals reduction and it sounds more realistic

It probably removed some frequencies

[analog obsidian]

Yeah i guessed it
They did use eq to reduce the artifacts

[brittle wing]

This isn't equalizer... it's some kind of filter.

[analog obsidian]

Riduttore means reduction?

[brittle wing]

Vocal remover

[analog obsidian]

Oh

[brittle wing]

It leaves out the bvs

[analog obsidian]

Damn those output have tons of editing

[brittle wing]

But I adjusted them both in a way I can hear not lowpass and highpass frequencies and both lead and backing

[brittle wing]

I use this just to test

How good the model is

[crude flame]

Anyone know how i can change the optimizer rvc uses?

[glacial pollen]

Yea, it's pretty easy

hold on

this path:
rvc root\infer\modules\train\train.py

optim_g = torch.optim.AdamW(
    net_g.parameters(),
    hps.train.learning_rate,
    betas=hps.train.betas,
    eps=hps.train.eps,
    #weight_decay=1e-5,
)
optim_d = torch.optim.AdamW(
    net_d.parameters(),
    hps.train.learning_rate,
    betas=hps.train.betas,
    eps=hps.train.eps,
    #weight_decay=1e-5,
)

this section

depending on how you import the optimizer

or if you use torch's optim package

[crude flame]

So i would just swap out the AdamW part for whatever optimizer? or can i only do that for some optimizers and with others i have to do more?

[glacial pollen]

say you get torch-optimizer package:

pip install torch-optimizer

in the train.py you could do this:
add the statement somewhere in the beginning of the script;

import torch_optimizer as optim

then

ex.: optimizer = optim.RAdam(model.parameters(), lr=0.001)

but in this case it'd be:

optim_g = optim.RAdam( ....

where "... " is whatever the script has

you'd just swap what's necessary in optim g and d

Generally any optim you'll find ( if standalone, on github ) should have init or usage note

Yet you get the base idea, you just wanna make sure you:

1. Import the package containing the optimier / optimizer itself properly ( at the beginning of the script )
2. handle it properly at optim_g = xxxxxx( and optim_d = xxxxxx(
   based on the usage example from a given repo or whatever

Would be easier to show if I had perms for posting pictures uhhh

@red kayak 'd mind to grant me the perm?

[red kayak]

go on

[glacial pollen]

yea, would you mind giving me perms for posting pictures?

[rare goblet]

Ayo? @glacial pollen level 21 !!!

[glacial pollen]

oh

[red kayak]

i did

[glacial pollen]

Arigatou!

[red kayak]

:3

[glacial pollen]

@crude flame
There, that's the import for scenario you use "torch_optimizer" package

[crude flame]

So i would make that import and change the optim like so

[glacial pollen]

and initializing it

To explain it easier for you, this:

import torch_optimizer as optim

imports torch_optimizer

a package having optimizers

registers it as " optim "

so then you handle it as optim.xxxxxx

where xxxx is what you're referencing from the package

in this case, you'd aim for RAdam ( an eample ofc )

so it becomes:

optim.RAdam

"RAdam from package optim "

Hope it's clear now!

[crude flame]

I understand

thanks!

[glacial pollen]

Neattt, just remember

not all of em handle it the same

[crude flame]

yeah

double check

[glacial pollen]

sometimes you gotta import an already optimizer script

for instance ranger21.py

standalone

then it could be:

from xxxx import xxxx

so, from package x1 import the x2

x2 could be an init part of the optimizer's code

so the optim's script knows you're referencing it.
But yea, again, it should be noted in optim's repo or something ~
Good luck!

[simple ore]

You can easilly swap AdamW with another, most are just drop-in files you can put into training folder and then just import

but the results are not that different / beter

[glacial pollen]

In fact, about a year ago or so I've experimented shit ton with that as I was learning rvc
and tbf, it's not worth it

[simple ore]

in my tests AdaBelief has better gradients, but not much difference anywhere

it is just a simple change

[glacial pollen]

radam, ranger, diffgrad and all of that. It's either unstable, has exploding gradients too often or just requires manual finetuning

not only that, adaptive lr is screwed no matter what ( but that coulda been bug in my code )

there are issues as well with AMP

so, fp32 is the only way or standard fp16 and that's not always ideal

[simple ore]

i've made accelerate script, so got rid of all those AMP loops

[glacial pollen]

ohhh, that's neat actually

but ye, playing with optims, considering sensitive hparams of pretrains isn't worth it really

maybe sdg

[simple ore]

warmup scheduler may be an interesting choice to try

[glacial pollen]

tried that with radam and lookahead tho it's too time consuming imo to find an optimal setting

probs more worth it for higher scale projects

[simple ore]

better spend time to figure out why hifigan does weirdness during intial epochs

[tough cape]

the app does not run when ever i try to it says Microsoft defender prevented an unrecognized app from starting

it's v.1.5.3.18a

is it the latest one? am i doing smth wrong?

[glacial pollen]

You mean the vanished gradients?

or the funky initialization?

gradients cause of AMP / fp16, doesn't occur in fp32 mode which is actually the best choice for hq models yet, yea, time consuming af
as for initialization, definitely warmup and learning rate matters

rvc devs didn't expect that broad recognition and the hparams are.. well let's just say barely " viable " for so many diverse sets we all have

[simple ore]

the horrible noise

[glacial pollen]

Got any tf event examples?

[simple ore]

if you train without pretrain

[tough cape]

mb wrong chat

[glacial pollen]

ah

why'd you go that way tho

[simple ore]

https://drive.google.com/file/d/1SmzHLol-ShTvcd-aK5kRP_Yp0q5gILaO/view?usp=sharing

[glacial pollen]

oh, well

[simple ore]

well, if I want to train with a clean set

[glacial pollen]

how many epochs did you do

without transfer learning it could take, well, shit ton of work

[simple ore]

that's just one epoch, but the noise lasts for a long time, 100+ still noticeable

[glacial pollen]

that's the thing

pretrains were trained for at least 10 to 30k epochs

iirc 20k or so

that's why transfer learning is required, to not go through that hassle

[simple ore]

no, og pretrain was trained on VCTK 630 epochs or so

[glacial pollen]

630? where did you get that from

[simple ore]

from file d/g model

iteration 636

[glacial pollen]

Hmmmm, then perhaps Nadare referenced something else back then

[knotty moth]

I bet he hasn't wasted his money on it

[simple ore]

and I can tell it is VCTK because speaker dimension is 109

[glacial pollen]

yeee, it is vctk corpus

[simple ore]

if you try to infer audio using og weights with those speakers, whey come out hell of a noisy

[glacial pollen]

Sad there's no actual true hi fidelity sets laying around

but then, robustness to noise has to be a thing - rvc devs expected casual noisy sets use rather than professional / clean hifi usage
so that's that

[simple ore]

yes, have to have a noise, but not too much

[brittle wing]

Uh they do

[simple ore]

my model trained on top of og with a little noise itself is pretty crisp in the end

[glacial pollen]

Yea

[simple ore]

has to have some noise for the model to handle it in the source audio

[glacial pollen]

Wasn't the case for me, I exclude it personally in pauses so the model doesn't replicate it

annoys the f out of me in infers

tho, I guess, that's one way to mitigate collapses a lil

Which is imo rvc's worst pain

[knotty moth]

too bad if he can't get refund from them stealing RVC models

[simple ore]

lemme make a silence test

[brittle wing]

What is refund English ain't my first language

[brittle wing]

No one is stealing you just upload the models and use them

But you pay for it

@low shard how did you find my TikTok?

[low shard]

huh? u shared the tiktok link

[brittle wing]

Uh it's okay

I only shared the clip

Not my account

Oh I understand I shared it out of my saved posts

[simple ore]

so pure silence gets filled in

but to hear it I have to crank my headphones way up

[glacial pollen]

not only that

see the dc offset

or generally that range

[simple ore]

but it did cancel the noise from the clip

[glacial pollen]

I mean yea, tho I personally still exclude it whenever I can
since I mostly work with audio I process myself and that includes spectral denoising, don't want my models handling or replicating these

Just my workflow

[simple ore]

and here's someone else's rip off of 11Labs model results

just a horrible hiss at 100hz

but they somehow managed again to cancel the original noise

but, I guess, since they did not have pure silence included in the training they did poorly handling that

[glacial pollen]

there's 2 ways out of it, they have're using detection threshold or either do some pre-run on noise cancelling

[simple ore]

buuuuut

If I use 0 index

no noise

and the noise from the clip got canceled

so ideally you may want a model trained with noise, but the index generated from clean samples

the noise is captured as a 'feature' of the voice

[glacial pollen]

wouldn't be ideal

Denoising alters the spectrum even so slightly, even psycho acoustics change a bit

I mean, it could work but I'd much rather have 100% of spectrum matching

[low shard]

when u share a video on tiktok, it shows who shared it

[brittle wing]

Uh yes

[low shard]

it even sked me if i wanna both watch and follow u, or just watch the video 😭

idk if it also tells u that i opened the link

[carmine fox]

Whats the best microphone thats just plug n play into my PC for the best and most realistic sounding voices?

[glacial pollen]

A microphone won't help you get "realistic voices"

It is the model, microphone is actually least important as long it's not an actual potato

Unless you're talking about training models

and not using them ( and that is, in w-okada or such - for realtime )

[carmine fox]

Im not getting any decent result with anything no matter what I do what settings i tweak, compared to listening to others on youtube for example

So a mic doesnt matter at all?

[glacial pollen]

First off, I'd need a demo of the output, then I can tell you what's the deal
another thing is, it's more about settings than a mic

even a poor laptop mic as long the model's decent and settings are properly picked, will sound just fine

Fidelity is a matter of the model, stability is mic + settings, but mainly settings

Aside, no model is the same. If a youtuber had a model "A" sound nice and clean doesn't mean a model "B" will do the same

[fleet fiber]

Here

[red kayak]

Sure I'll check them out later

[lavish viper]

Meow
Hello
I want to ask for help
after re-installing the machine, it does not start after re-downloading
Multkor something had to be patched, to which I received a link here
but it is no longer there

• I want to make an ai model, but I don't know what it takes
  In theory, I downloaded the one that also contains it

[near wyvern]

I exported this to onyx but seem to be having issues with it sounding like AI sometimes when speaking. I also have an issue where sometimes in discord if I am speaking and someone else speaks it will pick up their audio and replay it back so everyone can hear in the discord while being voice changes

idk what alot of the best settings are

[glacial pollen]

My friend, you shouldn't use Mangio rvc

it is outdated, buggy and is such a mess code-wise

Go for Applio or original RVC

edit:
Go for applio, I think I should actually update my repo first before recommending it, it's a lil outdated meh.
https://github.com/IAHispano/Applio

[lavish viper]

and is 4 GB of vram enough for this or do you need google collab to make the model?

[glacial pollen]

@lavish viper I'd say it's bare minimum to have it functioning ( as in, being able to somewhat train )

but sadly don't expect to go wild with some parameters like batch_size and / or huge datasets + batch_size

that's just what it is

Alternatively, I am pretty sure there are some other options out there ( not involving your machine), just search on this discord.
But if I was you? I'd still choose local training, even if limited or a lil slow

[lavish viper]

Does this only work with an Nvidia video card or the other one?

Ez hogy hogy magyar felületü?
Van itt aki tud magyarul? 😅

[glacial pollen]

nvidia

unless they worked out rocm, but I wouldn't know that ~ Not my field, researching it

[crude flame]

zluda works on applio so amd works

[brittle wing]

-svc

[azure marsh]

https://cdn.discordapp.com/attachments/1159822429137412177/1194008385113301003/svc.gif?ex=65c13ef6&is=65aec9f6&hm=ce1f57bc316793f641c80f9b359ba9db7d1312af6918b1aed9c8468c74035bac&

[brittle wing]

-svc

[azure marsh]

https://cdn.discordapp.com/attachments/1159822429137412177/1194008385113301003/svc.gif?ex=65c13ef6&is=65aec9f6&hm=ce1f57bc316793f641c80f9b359ba9db7d1312af6918b1aed9c8468c74035bac&

[brittle wing]

-svc

[azure marsh]

https://cdn.discordapp.com/attachments/1159822429137412177/1194008385113301003/svc.gif?ex=65c13ef6&is=65aec9f6&hm=ce1f57bc316793f641c80f9b359ba9db7d1312af6918b1aed9c8468c74035bac&

[cloud finch]

Applio error when i convert?

[simple ore]

to train a model you need something better than a 10-dollar mic

[simple ore]

too big of a file, use split audio under advanced settings

[simple ore]

with nvidia card that has 4GB at most you can do inference, and even then you may need to disable some features like CUDNN

[lavish viper]

What should be written here? CMD keeps throwing errors, I have 2 video cards, do I need its number or the number of the gpu cores? or something else

[simple ore]

4 GPUS? 🙂

only one of your GPU supports CUDA, by default it should be 0

[lavish viper]

meanwhile, I asked the co-pilot, he said to enter 0 through
the problem is that I don't know English, I'm also writing this with google translator😅

[simple ore]

and you should see its name under advanced options

You can not run training with 4GB though.

it is not enough

[rare goblet]

Ayo? @lavish viper level 3 !!!

[lavish viper]

ups

[simple ore]

miracle

maybe with a small model

[lavish viper]

I hope it's okay, the fan in the poor laptop turned on a lot, but there is such an amd card in the desktop PC, would this work?

[red kayak]

well theres compression going on here hence why the audio may sound a little worse

[simple ore]

you can use Applio with AMD GPU

see install guide for Zluda

[fleet fiber]

that's all what i wanted to prove so in the end, the results are different, not the same and sounds better on Jammable

[red kayak]

u are exoporting as m4a

not wav

which is the original format rvc outputs

[red kayak]

therefore you cant prove your point that way if quality is compromised by compression. does that make sense?

[lavish viper]

How long does this take approx?

[simple ore]

your run is going at 77s/epoch

[rare goblet]

Ayo? @simple ore level 20 !!!

[simple ore]

it is not that large set, like 20 minutes?

100-150 epochs should be enough to get something good

or 50 for something okay

[lavish viper]

He is currently writing this
if i want can i stop it?

[simple ore]

it saves every 10 epochs, so you can stop it after next if you want

you can also go to inference tab, click Refresh and pick the latest saved one to try it

also... did you actually select v1 as the architecture?

[lavish viper]

no
v2
version 1
😅

[brittle wing]

Well it doesn't allow WAV files unless you have nitro also audio format has nothing to do with how it affects the sound often

This person sent me the m4a jammable output and yet I only converted the Applio and the Ilaria RVC zero thing into m4a

[red kayak]

1. you can upload them on your google drive + you can export as lvl 8 flac
2. it does cuz lossy compression algorithms at such low bit rates have a very audible difference

[brittle wing]

I personally export in wav

Idk everyone has their preference speaking of audio format

[red kayak]

audio is much more nuanced and exporting at a very different bit rate can make it harder to evaluate

[brittle wing]

Also just because something is WAV doesn't mean it's always good quality.

[brittle wing]

I know I'm not dumb

[red kayak]

but u fail to understand how compression can affect the audio quality.

[brittle wing]

I don't

I know it all

[red kayak]

okay then if you wanted a fair comparison, you 2 shouldve provided wav files or flac files

[brittle wing]

You just don't wanna admit the jammable sample sounds the best

[brittle wing]

The person sent me an m4a file

And I exported in WAV and converted to m4a

[red kayak]

thats the issue, both m4a files are of different bit rates what is it that u cannot understand

obviously the one with the higher bit rate will sound better

its math

[brittle wing]

Isn't m4a always 128 kbps

Or 320

[red kayak]

and then you tell me you know it all

[brittle wing]

That's not the case here

M4a can be 320 kbps I know dude

[red kayak]

but its not limited to just 128 and 320

[brittle wing]

Then message them for a WAV raw output

[red kayak]

i did

[brittle wing]

Not me

Anddd okay

[red kayak]

you just came here and stirred up the convo, hence why im replying

[brittle wing]

Because this person is correct

[red kayak]

hopefully this helps u understand how audio works

[brittle wing]

I know already WAV is the best format why are you treating me as if I'm some type of dumb person like the hell?

[red kayak]

im explaining to you why the jammble output sounds better than the rvc one. the jammble output is higher in bit rate hence why it sounds better what do u not understand

[cloud finch]

Does split audio take longer? mines taking forever now that i did split audio

[brittle wing]

Bitrate has nothing to do with sound
I can send you a muffled sound in a WAV format and the same clear one in 320 kbps mp3 one and the mp3 one will be better

Yo just don't wanna admit it's better come on

[red kayak]

you dont know anything about audio or/and your ears are just bad.

[simple ore]

sample rate conversions take time, so it may help to have a file prepared with a correct sample rate

[brittle wing]

Are you serious my ears are bad bro I have literally very sensitive ears

[simple ore]

but then it also downsamples to 16k as well

[red kayak]

bitrate is the quality of the audio

you are using the same settings to infer audio

so it doesnt matter

[cloud finch]

So im assuming split audio is a sample rate conversion? Cause when i dont use split audio it takes like 20 seconds ( on smaller audios)

[brittle wing]

I'm not hearing impaired or disabled bro did you just call me deaf

[lavish viper]

It turned out quite well, only the mouse wheel remained inside and I was too lazy to listen to the sound for the whole 20 minutes, but I will fix that too later

[simple ore]

splitting does not add any extra time, not much anyway

[lavish viper]

Thank you very much for your help

[simple ore]

having original sample rate matching your new model saves ~20 sec for 45m mp3

[cloud finch]

so whats going on here?

[simple ore]

you're upscaling for no good reason

dont

[cloud finch]

OH

i forgot i enabled that

Is upscaling bad?

[brittle wing]

Bro really you just don't wanna admit

[rare goblet]

Ayo? @brittle wing level 9 !!!

[brittle wing]

No

[simple ore]

I dont think it makes any improvemen as the audio to infer still loads at 16k

[cloud finch]

interesting

[cloud finch]

👁️

[simple ore]

source audio is only used to extract features and f0

[red kayak]

https://chatgpt.com/share/670c4496-caf8-8002-81c4-4f5580cdf270
just read what the chat bot has to say

[simple ore]

and features are captured from 16k files during trainig

[red kayak]

cuz u have no clue whats going on

[brittle wing]

This person just prefers m4a format

[red kayak]

i cannot compare a more compressed file

[brittle wing]

Okay message.THEM

[cloud finch]

[brittle wing]

I always use WAV or good quality audio

Idk

[red kayak]

jammable also cleans up the original vocals, the dude infering the audio through ilaria rvc just doesnt know how to clean the input audio themselves

[brittle wing]

I'm the one who inferred through Ilaria.

They only sent the jammable thing

But the preset I use to do the cover literally has a noise gate in

[brittle wing]

How do I clean it then

It was literally made with Mel-roformer karaoke which is noise free
Unlike MDX and I've architectures that generate noise

[finite prawn]

what's the 'fo curve' and how is it used?

[simple ore]

f0 is a fundamental frequency, a pitch of audio signal

a curve is a recorded number of values for a specific audio

RVC has an option to replace f0 curve during inference from a set from a file, but I've yet to see someone successfully doing it

in theory you can use it to lower a pitch in a song in specific parts to prevent the voice from cracking

[brittle wing]

@red kayak

I literally went through a hearing test bro my ears are fine

[glacial pollen]

ey, that's something I haven't thought of lol
Alternatively, one could just make own dramatic pitch slopes

Perhaps the easiest way would be to just run the vocal through f0 detection then have it edited manually

[simple ore]

anyway, I dont think the RVC code works at all

[glacial pollen]

Can't tell either, Pretty sure I saw it in the pipeline

[simple ore]

that pitch replacement code is top tier garbage

[glacial pollen]

just don't remember if it was actually implemented or was a dummy placeholder

tbh, rvc'd benefit a gui refinement with some node-based pitch editor

similar approach to synthv / vocaloid

[simple ore]

it does something, but without any infomation what needs to be in the file best I could do is making it 'correct' a padding and half the audio

[glacial pollen]

I am 80% sure it's meant to be in a format similar to what f0 outputs

[simple ore]

not really, different values

[glacial pollen]

iirc, the pipeline accepts the f0s in place of actual f0/feature extractor's

so that's just my guess based on that

[simple ore]

i've dumped f0 as a file before this code ``` if inp_f0 is not None:
delta_t = np.round(
(inp_f0[:, 0].max() - inp_f0[:, 0].min()) * tf0 + 1
).astype("int16")
replace_f0 = np.interp(
list(range(delta_t)), inp_f0[:, 0] * 100, inp_f0[:, 1]
)
shape = f0[self.x_pad * tf0 : self.x_pad * tf0 + len(replace_f0)].shape[0]
f0[self.x_pad * tf0 : self.x_pad * tf0 + len(replace_f0)] = replace_f0[
:shape
]

then I tried to adjust it and feed it as a text file

[glacial pollen]

and how's results? works or nah?

tho

I think it requires numpy arrays

rather than values in text

can be wrong tho

But then, I am not a specialist in it at all so can't help with dimension requirement and such

[simple ore]

no, it reads a text and then converts to np array

                    lines = f.read().strip("\n").split("\n")
                inp_f0 = []
                for line in lines:
                    inp_f0.append([float(i) for i in line.split(",")])
                inp_f0 = np.array(inp_f0, dtype="float32")

two columns, I think

[glacial pollen]

oh, then that's even less of problems

[pastel roost]

-svc

[rare goblet]

Ayo? @pastel roost level 1 !!!

[azure marsh]

I'd recommend RVC

[pastel roost]

-svc

[azure marsh]

https://cdn.discordapp.com/attachments/1159822429137412177/1194008385113301003/svc.gif?ex=65c13ef6&is=65aec9f6&hm=ce1f57bc316793f641c80f9b359ba9db7d1312af6918b1aed9c8468c74035bac&

[pastel roost]

NOOO

-rvc

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.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

[near wyvern]

anyone know a fix? Sample Rate is not supported by device:
Checking Available Sample Rate:
[checkSamplingRate] Error opening InputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening OutputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening InputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening OutputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening InputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening OutputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening OutputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening InputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening OutputStream: Invalid sample rate [PaErrorCode -9997]
[checkSamplingRate] Error opening InputStream: Invalid sample rate [PaErrorCode -9997]
Available Sample Rate:
[Input]: [48000]
[Output]: [192000]
continue...

[cloud finch]

are there any effects i can put on this to make it sound more realistic?

[young halo]

Help, why do i get this error when it's clearly a zipfile?

[crude bolt]

read the error message, it already says what is wrong:

Check which sample rate is supported by your device , in specific your selected audio device. The one you set as input and output device.

there are many ways to check this; search with google "how to check your available sample rates for your audio device".

Once you know this adjust your settings accordingly.

[crude bolt]

define first, or try to explain us why it does not sound realistic enough for you.

your ears work diffrently than mine. I can give alot reasons but its for you to train to listen to this voice and know what is wrong first.

have you even tried your path that you input

it shows me a webpage.

with a download link

you need to give the url to that download link , not to the webpage.

i assume you clicked "copy download link" and pasted it.

Instead you need to rightclick the word "download", copy link address and paste in : https://huggingface.co/ivoryprincemilo/Faouzia/resolve/main/faouziatest4.zip?download=true

@young halo

[low shard]

what are u spusing sned link, and also of the model u are using

[knotty moth]

ofc that "model" link returns as an html, try replace /blob with /resolve

[regal sentinel]

Hello , where cacn I download the lastest version for NVDIA user?

[flat zealot]

this is like my 2nd time training a model and the 1st time i didnt know what i was doing

but how do u know when ur overtraining using tensorboard

theres no doc for it either

wait nvm i found it

[low shard]

latest version of what?

realtime voice changing for calls?

[rare goblet]

Ayo? @brittle wing level 5 !!!

[sudden lintel]

can someone link me with tutorial to add the voices i downloaded

[astral perch]

What happend to the RVC interface of kaggle

There is no longer a V2 and V1 option

[simple ore]

why would you want v1 anyway?

[brittle wing]

-colab

[azure marsh]

☁️ Google Colabs

• 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, -overtraining, -realtime, -rvc, -help

[brittle wing]

-hf

[azure marsh]

<:huggingface:1179800228946268270> Hugginface Spaces

• UVR5 UI, by Eddy and Ilaria Huggingface Spaces
• Ilaria RVC Zero, by thestingerx Huggingface Spaces
• RVC⚡ZERO, by r3gm Huggingface Spaces
• Applio, by IA Hispano Huggingface Spaces
• 🆕 FaceFusion UI, by Nick088 Huggingface Spaces

More commands: -audio, - colabs, -kaggle, -local, -overtraining, -realtime, -rvc, -help

[low shard]

v2 is better

[astral perch]

There is no option anymore

[rare goblet]

Ayo? @astral perch level 12 !!!

[astral perch]

No I mean they're both not there

[low shard]

to use them realtime for voice changer in calls? or for pre-recorded audios?

And what's ur pc gpu?

[low shard]

which kaggle are u using

[astral perch]

How can you see?

https://www.kaggle.com/code/hinabl/mainline

This one

[low shard]

in the training tab?

[astral perch]

Yes, now i can oly change the cpu cores but not the rvc version

[brittle wing]

Excuse me are you a native English speaker

Is there a native English speaker to help me further with the transcription?
Like literally AI sites and services don't pick words correctly like a human ear would

[low shard]

nope i'm italian

[brittle wing]

Mamma mia

Davvero

Sorprendente

[brittle wing]

Almeno sai parlare inglese bene mi puoi aiutare

[low shard]

sure i could try

[brittle wing]

Ah grazie ti scrivo ❤️

[errant flame]

Can someone explain why it's not picking up my mic when using an RVC model but using a breatrice model it works fine?

[low shard]

I'm not a wokada helper but i seen others talk about this:
if only the beatrice model working, its a failed install, try to delete pretrain folder and model_dir and let it reinstall those

[final lichen]

how do I run them?

[rare goblet]

Ayo? @final lichen level 1 !!!

[green citrus]

-local

[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

Credits to Faze Masta and Antasma for compiling these links.

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

[knotty moth]

what? were you and even AIs struggling on some rare english words that non-native english may not understand?

[brittle wing]

AI can't compare tk human hearing also Ilaria RVC zero doesn't decrease volume of converted vocals, I was just with a wrong impression
Yes I was kinda struggling

Most of the lyrics were guessed correctly mostly complete at the starting I succeeded alone

[knotty moth]

im abit curious, what accent is it?

[brittle wing]

He's black

[knotty moth]

prob some regional american?

[brittle wing]

I haven't seen his face he never showed it in his videos he's a spiritual YouTuber no longer active and his song is absolutely rare, I asked for it and got it sent in 2020

[knotty moth]

uh seems like his own style

[brittle wing]

?

[low shard]

Run what?

[final lichen]

Rec models

[rare goblet]

Ayo? @final lichen level 2 !!!

[final lichen]

Rvc*

[low shard]

so, inference (using models) on pre-recorded audios, or realtime for calls?

and whats ur pc gpu?

[final lichen]

3080

[final lichen]

Realtime

[low shard]

alr, ur pc is good enough

-rt

[azure marsh]

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

[low shard]

1st link

[final lichen]

That's the thing, I can't use a van cable

Like idk how to set it u

Up

[low shard]

its wokads, uses RVC models, specilized for realtime, and its the fork (modified version) which has better performance

[low shard]

its explained in the guide u have to read

[final lichen]

Okay

Thank you

[low shard]

yw

[final lichen]

Have a nice day/night

[low shard]

you too

[crude bolt]

morning

just reading up a bit 😉

[errant flame]

how do i make it reinstall it?

[crude bolt]

#🔍│help-w-okada message

[errant flame]

thank you

[rare goblet]

Ayo? @errant flame level 1 !!!

[hot lagoon]

-kaggle

[azure marsh]

📘 Kaggle Notebooks

• Applio Notebook, by Vidal Kaggle
• Applio Notebook, by Shirou Kaggle
• Music Source Separation, by Shirou Kaggle
• UVR5 NO UI, by Eddy Kaggle
• RVC Mainline, by Hina Kaggle
• Original W-Okada's Voice Changer, Kaggle
• Modified W-Okada's Voice Changer, Kaggle
• 🆕 UVR5 UI, by Eddy, ArisDev & Nick088 Kaggle
• 🆕 RVC AI Cover Maker UI, by Shirou & ArisDev Kaggle
• 📖 How to use RVC Mainline Kaggle by Cauthess

Note: Kaggle limits GPU usage to 30 hours per week.

[lean surge]

yo

wondering how I can install rvc?

https://github.com/fumiama/Retrieval-based-Voice-Conversion-WebUI
is this supposed to be the one I should download

[crude bolt]

read a bit up - alot questions come down to same thing 😛

[lean surge]

ok thanks I'll read a bit up

[crude bolt]

rvc is more the realtime voice change in a broad way - the w-okada is the software we often recommend if you have a capable computer to run it on.

i wouldnt recommend fumiama. it is not this easy to get it up and running when you are new to rvc and have no developer background.

[feral marlin]

Is crepe mode still available on rvc?

[simple ore]

yes, but why? you want to sit in front of a computer for hours while it crawls thru your file?

[low shard]

wha't us pc gpu btw?

[sly niche]

hey guys any1 else had an issue where the client isnt opening anymore? i see the window pop up for a split second then it dissapears

[pastel oak]

next time specify what program ur using but im assuming wokada
https://cdn.discordapp.com/attachments/1159290161683767298/1159607580184821760/issue.mp4?ex=670f8b77&is=670e39f7&hm=870fc51bd7550ed1580c28e43073670d36c84c331926bf40d9cef06ac8ea33cb&

[dry vine]

hello, i want to avoid the disconnecting for inactivity, i am following the workaround but there doesnt seem to be an *other tool * option in the console

[rare goblet]

Ayo? @dry vine level 1 !!!

[low shard]

hey, which browser are u using?

[dry vine]

chrome

[low shard]

i'm italian, but don't you see something like this when u do 3 dots (might need to scroll a bit down) ?

[dry vine]

is this the 3 dots after pressing control+ shift+j?

[low shard]

when i wrote this, i meant u can choose between one of those ways

i might need to specify it better

[dry vine]

oh ok

so i opened console, now where do i paste the code?

[low shard]

does this sound more clear ?

[low shard]

when u are in the console (dont mind the errors, im doing it on a random site just to show u), write first allow pasting and click Enter. After that, paste the code found in the google doc

[brittle wing]

Ilaria RVC is very robotic and it generates noise in the output even through the sample is completely noiseless I like fr it generates noise
The models I use are the best.Its not from the model, @red kayak literally said idk how to Denoise my samples but no my samples aren't filtered with MDX and BSRoformer models that generate noise, I used MelBand Roformer here

[low shard]

for some reason it didnt send the image lol

[dry vine]

syntax error

[low shard]

could you show me a screenshot?

cus i just went on a random colab (applio) and the script didn't spit me out any errors

[dry vine]

in my opinion you can try putting a solid black dot before the * Click the ... * part so other people know that it is another option, and not the next step

[low shard]

huh pretty weird, could u try the other way pls?

Click the gear on the top right corner.Go to Experiments. In Filter type "pasting", and untick Show warning about Self-XSS when pasting code.

[dry vine]

it is not there

do i try just pasting the code?

[low shard]

try that and tell me

might be they updated chrome

[quasi dagger]

You copied the word "allow pasting" instead of the code

[quasi dagger]

Also you forgot the set interval part which triggers it repeatedly

[low shard]

that was bc chrome makes so u get warnings that doesnt let u copy paste code in the dev tab

atleast when i wrote that guide

unless they removed it?

bc i genuinely can't see it in the settings anymore

[quasi dagger]

I'm on phone rn

[low shard]

it was in the other page

google docs be like

[low shard]

idk but i cant see any news about it being removed, but at the same time cant find it

[simple ore]

dont use UI colab, your account gonna get banned

or at least dont use it on a main account

[quasi dagger]

Might have updated

[low shard]

i do it since a year on like 4 accs lol

they barely ever ban tbh, mostly disconnect if u get caught

i remember hina getting banned, and unbanned the next day

[low shard]

ye will check and update the guide

[analog obsidian]

iirc you're allowed to run webuis if you pay for google colab pro (i am aware hina's colab bypassed this)

[low shard]

btw like all of the people on cloud here broke google colab and kaggle terms basically

[low shard]

yes, only on the pro tier of colab
on kaggle not at all

and yes most colabs here are encrypted since a year

[quasi dagger]

I think the workarounds also worked on non UIs since the RVC disconnected had a notice to use workarounds (never tried though)

[brittle wing]

how do yoou make it sound real and authentic

[fresh cairn]

what if im on arch linux??

[pastel oak]

theres a linux section on the guide

[low shard]

there's a linux part

[fresh cairn]

only for the download

and the merge

not running

i dont know if i need to use a bottles or something

[rare goblet]

Ayo? @fresh cairn level 5 !!!

[fresh cairn]

it doesnt run via terminal or double click

[rich fiber]

my voice changer fades and sometimes works but other times it stutters and sounds super choppy, anyone know why?

i'm free to hop on a vc if anyone is free

[pastel oak]

whats your gpu

[rich fiber]

geforce rtx 3070

[pastel oak]

Whats the full name of the file you downloaded? Where it says mmvcserversio_onnxgpucudablabal some numbers?

or alpha/beta?

[rich fiber]

i'm using the base voices

[pastel oak]

thats not what i asked

whatever, ill assume u have the right version

[rich fiber]

ohh sorry

[pastel oak]

whats your f0 det chunk and extra

[rare goblet]

Ayo? @rich fiber level 1 !!!

[pastel oak]

screenshot

[rich fiber]

ahh i fixed the issue

sorry, thanks for the help anyways

[pastel oak]

It looks ok, you can up extra to 16384 and reduce chunk for less delay

[rich fiber]

thank you

[dry vine]

is it normal for applio on cloud to have very slow download speed?

[rare goblet]

Ayo? @dry vine level 2 !!!

[brittle wing]

-colab

[azure marsh]

☁️ Google Colabs

• 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, -overtraining, -realtime, -rvc, -help

[brittle wing]

-hf

[azure marsh]

<:huggingface:1179800228946268270> Hugginface Spaces

• UVR5 UI, by Eddy and Ilaria Huggingface Spaces
• Ilaria RVC Zero, by thestingerx Huggingface Spaces
• RVC⚡ZERO, by r3gm Huggingface Spaces
• Applio, by IA Hispano Huggingface Spaces
• 🆕 FaceFusion UI, by Nick088 Huggingface Spaces

More commands: -audio, - colabs, -kaggle, -local, -overtraining, -realtime, -rvc, -help

[brittle wing]

-rvc

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.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

[brittle wing]

-kaggle

[azure marsh]

📘 Kaggle Notebooks

• Applio Notebook, by Vidal Kaggle
• Applio Notebook, by Shirou Kaggle
• Music Source Separation, by Shirou Kaggle
• UVR5 NO UI, by Eddy Kaggle
• RVC Mainline, by Hina Kaggle
• Original W-Okada's Voice Changer, Kaggle
• Modified W-Okada's Voice Changer, Kaggle
• 🆕 UVR5 UI, by Eddy, ArisDev & Nick088 Kaggle
• 🆕 RVC AI Cover Maker UI, by Shirou & ArisDev Kaggle
• 📖 How to use RVC Mainline Kaggle by Cauthess

Note: Kaggle limits GPU usage to 30 hours per week.

[fresh cairn]

i cant figure out how to run it though

[odd copper]

For some reason I get a 6 second delay on any voice modeules that arent the prebuilt anime girl

[vast harbor]

sonic

[pastel oak]

If you have an amd gpu, you need to hit the "Export to onnx" button when you upload one yourself. And then upload that new file. Thats the only difference between the prebuilt voices

[low shard]

other than the guide i dunno how to help you, maybe @pastel oak

[pastel oak]

I mean the MMVCServerSIO is supposed to run right after you extracted everything. If not i cant help u im not linux

You can ping Emojikage, the developer

[brazen willow]

@low shard

[low shard]

yea dw about the links, u need to worry only about the status

and that means its downloaded, just go to the Inference tab, and do refresh models

[brazen willow]

ohhhh

[timid falcon]

why does kaggle sometimes replace logos with text???

[low shard]

May be a wifi skill issue

[knotty moth]

https://tenor.com/view/none-no-image-error-red-x-cat-gif-11953747

[timid falcon]

i have good internet

when i click on the console it just randomly does that

[timid falcon]

no idea if it's my fault but kaggle seems broken

training index takes forever and probably doesn't work, tensorboard doesn't do anything except show the very first value and nothing else

[shy ember]

is there any tutorial for RVC

im pretty new to all this and im currently downloading the colab just not sure how it works

[lament igloo]

Hi,
yesterday I trained a new Model with 1000 Epochs, but I forgot to check the "Save a small final model to the 'weights' folder at each save point" option. While training I let RVC save every 5th epoch the model and optimizer state. Tensorboard now shows me, that step 15.800 out of 19.800 could be the best. Is there a possibility to create a small final model from those big G_15800.pth and D_15800.pth files later (now) without needing to do the whole training again? ? Please, help 😄

[simple ore]

yes it is possible

[lament igloo]

ehm, could you please tell me how? 😄

[simple ore]

I used a python script

[rare goblet]

Ayo? @simple ore level 22 !!!

[lament igloo]

can you give me a link to that script, please?

[rare goblet]

Ayo? @lament igloo level 1 !!!

[lament igloo]

@Arclame: Yes, sir 😄

Python Level 1 as well 😄

[simple ore]

well, I use Applio, so script is for Applio

[lament igloo]

no problem. I think it works the same way

[simple ore]

yes, just use libraries from RVC, they are more or less the same

technically it can be done another way

take existing pth from your logs, take g file

[lament igloo]

Ok?

[simple ore]

load both, copy weights from g file into the model.pth, save with a new name

[knotty moth]

as long you don't lose the preprocessed and feature files

[simple ore]

I have not tried that

[lament igloo]

I tried to load the G file (and the D file) as my model. Didn't work

[knotty moth]

put them in the logs folder alongside the preprocessed files

[lament igloo]

@Allisa I haven't lost any of those files. All are there (due to my every 5 Epochs save, it my Folder in log dir has 245 GB 🙂 )

They ARE in the logs folder.

All of them

[knotty moth]

that's why you should enable the "save only latest ckpt"

[lament igloo]

All G- and D- *.pth files

[lament igloo]

Yes, but this is my Question: I forgot that and want to know if I can create this small model file now (from the D- and G- *.pth files in my Log Dir)

[knotty moth]

nah the weight files can only be created through the training session

[lament igloo]

@Noobies says "NO"

[rare goblet]

Ayo? @lament igloo level 2 !!!

[lament igloo]

What Level 2? In not knowing anything? 😄

[knotty moth]

again these pth files are different, the latter is the weight file and it can only be created through the training session

(tbh I have no idea how a pair of G and D files may contain a weight file) nvm it doesn't

https://tenor.com/view/huh-cat-gif-26460616

[lament igloo]

I think noobies script is not what I'm looking for. Although I'm a loser in Python it seems to me not to be a script to make the final small model out of the epoch saves (G- and D-Files in logs dir). Maybe I'm wrong, but I can't see, where I had to put in my data in that script to manage this.

[knotty moth]

if not sure start over training with "save only latest ckpt" enabled in order not to waste a hundred gigs, and don't forget to analyze tensorboard graph and pick some best weight files

[lament igloo]

yes but I already trained 1000 epochs. And I now want to grab the model from epoch 795 out of it. Because in Tensorboard its the best one. But I forgot to check the "Save a small final model to the 'weights' folder at each save point" option and now I dont have these small models. Do you understand my dilemma? But I have all the Data from the Training, as I sai: All the D- and G-Files in logs dir (saved at every 5th epoch) Isnt the small model created out of those files?

[knotty moth]

(1000 epochs) / 5 * (1.2 GB of a pair of G and D files) = 240 GB
u sure u need that huge size? instead of:
(1000 epochs) / 5 * (54 MB of a weight pth file) = 10.8 GB
plus you can still analyze the tensorboard graph after the training session

[lament igloo]

Dont you understand what I am saying? It was a mistake! I forgot to activate this option

[simple ore]

@lament igloo you know, the easiest option perhaps if you put D/G files from your model into custom pretrain

[lament igloo]

and now i dont want to do the training again

[simple ore]

and train 1 epoch of the same data set with those and save model pth

[analog obsidian]

i can convert your g file to small weight model

upload it to huggingface

(only G not d)

but anyways mainline can convert G to small weight in the ckpt processing tab

iirc is at the bottom of everything so scroll down

there should be a setting with a description that says this convert G to small weight or something like that, i dont remember lol

[lament igloo]

@analog obsidian Thanks for your help. What is mainline? A toll? If so, I can do it on my own

[analog obsidian]

the original rvc

https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/releases/tag/2.2.231006

[knotty moth]

mainline = 主线

[analog obsidian]

if you're using that version, find the ckpt processing tab (i don't remember its exact name) and scroll down until you find something related to convert G to small weight

[knotty moth]

yea the bottom part after the model merge function

[lament igloo]

I'm using the original RVC. Yeah, great. I've found it 😄 ... so its not impossible (@Alisa 😉 ) ... Ok, everything there seems to be clear. Does anybody know what "Model information to be placed:" means?

[quasi dagger]

Those G and D are checkpoints to continue training from there (when resuming training)

[lament igloo]

Great. It worked. Thks to all of you for your help and have a nice day 😉

[jovial sierra]

Someone can help me with the colab version?

[honest anvil]

hey I got a problem... whenever I upload a voice to the voice changer (I made sure that I choose the right file .pth) but after uploading I still cant find the voice added to the voices list

[brittle wing]

if i've got a short audio, do i put more or less epochs

[odd shale]

Which is the length of your audio/dataset?

[low shard]

you need to use the tensorboard, there isn't a right amount of epochs https://docs.ai-hub.wtf/rvc/resources/epochs-tensorboard/

[simple ore]

well, you can get a good result in 10 epochs

or less

[low shard]

that sounds undertrained

[simple ore]

give me a test sample, i'll run my 10e model on it

[low shard]

are u talking about the Lion thing u did in #🔊│ai-development

[simple ore]

Lion did not do a good job

regular adamw

with default pretrain and 45m sample set

[brittle wing]

yeah like a 20 second audio

[brittle wing]

yeah but should it be more like 500 or more like 50

[low shard]

ig u could try the ai hub anthem

[odd shale]

Yeah.. I would suggest you to get at least 5 or 10 mins of dataset.

[low shard]

#✨│announcements message

[odd shale]

Because 20 seconds won't be enough and will may lead to bad results.

In my opinion, at least.

[brittle wing]

20 seconds is all i can get

there's literally no more official audio

[brittle wing]

is there a cloud way like a collab or smth

[low shard]

of using the tensorboard? or training in general

[brittle wing]

tensorboard

[low shard]

yea depends on what u using to train

[brittle wing]

rvc disconnected

v2

[low shard]

u should get the tensorboard after u start training

so wouldn't need to install it

btw kaggle way better for training, in terms of more gpu time

[odd shale]

You got 2 options, either you discard making the model of that voice or still try doing it.

I wouldn't even dare using 20 seconds of audio.

[odd shale]

I love Kaggle a lot haha.

I got 2 accounts there

[brittle wing]

well it's all i can do
I've heard a lot of ai models with like 30 second datasets and i am aware of the quality some of the voices may turn out to be

[brittle wing]

how do i do that

[odd shale]

I think you can read the kaggle docs.

-kaggle

[azure marsh]

📘 Kaggle Notebooks

• Applio Notebook, by Vidal Kaggle
• Applio Notebook, by Shirou Kaggle
• Music Source Separation, by Shirou Kaggle
• UVR5 NO UI, by Eddy Kaggle
• RVC Mainline, by Hina Kaggle
• Original W-Okada's Voice Changer, Kaggle
• Modified W-Okada's Voice Changer, Kaggle
• 🆕 UVR5 UI, by Eddy, ArisDev & Nick088 Kaggle
• 🆕 RVC AI Cover Maker UI, by Shirou & ArisDev Kaggle
• 📖 How to use RVC Mainline Kaggle by Cauthess

Note: Kaggle limits GPU usage to 30 hours per week.

[lost lodge]

Hello! Can you tell me how to create a voice model here?

[odd shale]

Read the cauthess one.

[odd shale]

You can start by reading the docs buddy.

[low shard]

what's ur pc gpu?

[odd shale]

Yeah, what's your GPU bud?

[lost lodge]

where do i check my gpu

[low shard]

it just shows up while u train, read the rvcdisconnected guide https://docs.ai-hub.wtf/rvc/cloud/rvc-disconnected or either use Kaggle which has more gpu time

[low shard]

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

Before that, perchance are you on mac?

[lost lodge]

no im on windows 10

[low shard]

good, then do what i said above

[lost lodge]

i5-6500T

[low shard]

that's the cpu

go to the gpu tab

[lost lodge]

HD Graphics 530

[low shard]

u got integrated graphics, which means its bad and u cant do it locally (on ur pc)

[brittle wing]

Can you explain what gpu time is?

[low shard]

U can train on cloud tho

[lost lodge]

oh :<

[low shard]

As you dont got a good PC, its better you use cloud (remote good pc) for training an RVC Voice Model:

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
  • RVCDISCONNECTED (no ui)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio by Vidal (UI)
  • Applio by Shirou (UI, no guide as of right now)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • Mainline (UI, No guide as of right now)
  • Applio (UI, No guide as of right now)

[lost lodge]

how do i do it?

[low shard]

Kaggle gives more gpu time, its suggested

[lost lodge]

hmm

ima choose Mainline i guess

[low shard]

basically, gpus are expensive, so they cant give u for free unlimited time everyday with the gpu, so ur free account is limited to tot time of gpu to use, after that u cant use the gpu

For example on google colab u got only around 4 hours daily of gpu, but u could get even less bc its not granteed as its free

While kaggle gives 30 hours weekly of free gpu granted, but its a bit harder and requires phone number

[lost lodge]

on the kaggles

[low shard]

ye

[lost lodge]

i dont need to download anything right?

[low shard]

yup u dont, its on a remote good pc, not urs

[lost lodge]

Oh ok

thank you, ima try to voice clone and make a song cover

[low shard]

yw

[low shard]

for making ai covers, i suggest u to use ilaria rvc zero btw

[brave garnet]

Ilaria RVC: CLICK HERE 🤗

Guide on how to use it: CLICK HERE 📝

Don't forget to thank Ilaria if you find it useful! 💖

[lost lodge]

oh ok

is kaggle just for voice cloning?

and i can use Ilaria after??

[low shard]

ig u could also do inference (use models) on kaggle but not sure about that tbh as its not in the guide

[lost lodge]

Okay

[lost lodge]

i dont understand this part

what do i do to go to the notebook

my screen rn

nvm i got it all on :D

[rare goblet]

Ayo? @lost lodge level 5 !!!

[lost lodge]

wheres headless mode

:0

[lost lodge]

sorry for ping D:

i need help

[lavish viper]

Meow
Hello
I would also like to ask for help
It's already quite good, but the sound is a bit robotic
I don't know how I could improve this, I took out as much noise as possible and also cut out the noise below 150Hz

the other is that it can only be a min 10 max 30 minute audio or can there be up to 3-4 of these in the folder from which I teach?

[lost lodge]

i already ran the first cell

whats wrong?

help me

[serene hull]

how to separate vocal, can u recommend me some rvc

[lost lodge]

what does these 2 usually take

[lost lodge]

this one is taking forever-

wait its an installation

[lost lodge]

???

[lavish viper]

In the meantime, I also tried the AMD 6600XT, but it takes a long time compared to the laptop
I don't think it works
The program sees the GPU

[lost lodge]

nvm im dumb

[simple ore]

1. did not install VC redist, 2) did not patch torch

1 is most likely

2 could be if you tried to patch torch with Applio already open

[lost lodge]

im confused on this part

??

[lavish viper]

In theory, based on the tutorial, everything is there
I restarted the machine

[lost lodge]

[lost lodge]

can anyone help me on this part?

[azure marsh]

Hey, YB! 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

[lost lodge]

im confused

!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.

[lavish viper]

I don't understand it, but I don't think it works

but it seems to me that there was a rumor on the internet that the game was canceled because Nvidia didn't like it

Or the AMD Driver is just rubbish as always

[crude flame]

Run env\python -m pip uninstall torch torchvision torchaudio in the applio folder then run env\python -m pip install torch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 --upgrade --index-url https://download.pytorch.org/whl/cu118

[lavish viper]

this happened 5 times already

but I'm going to sleep, I'll try again tomorrow

[lost lodge]

i only have 1 wav file of a youtuber talking for 1 minute and 3 seconds

does it still work

[lost lodge]

1 hour straight and no response :(

i guess ima do it tmr

[modern gorge]

heyy can someone tell me wheres the tutorial?

like in what channel @acoustic scarab

[brittle wing]

i have a bug that i have no friends

[tame mica]

relatable

[golden pagoda]

im brand new and am wondering which version of rvc i should download? do i just download the top one on the release page of the github?

[burnt zealot]

Discarding ID3 tags because more suitable tags were found.
C:\Users\A\Documents\Audacity/audio176.wav->Suc

what this mean?

[unborn river]

help me

[knotty moth]

pls explain before I let you fall off the cliff

[fresh cairn]

wouldnt that get me banned?

[pastel oak]

Not in this case

[sly valve]

how to stop cutting in and out in a game? There not really delay but the talking kinda breaks.

[lavish viper]

I had to restart the machine after that too, but I'll try again, maybe it didn't work

[rare goblet]

Ayo? @lavish viper level 4 !!!

[lost lodge]

Who can help me on this part

[brittle wing]

-colab

[azure marsh]

☁️ Google Colabs

• 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, -overtraining, -realtime, -rvc, -help

[low shard]

did you make your dataset first?

like the cleaned one vocals only in .wav or .flac

[lost lodge]

No, how do I make it

I only got a audio .wav file of a ytber talking in 1 min

[low shard]

to train a model u need a dataset

https://docs.ai-hub.wtf/rvc/resources/datasets/

[lost lodge]

10 minutes audio

[lost lodge]

Thank you

[low shard]

yw

[plush compass]

How do you use the deecho-dereverb settings?

[pastel oak]

Depends on your GPU. Maybe cant handle the chunk you selected

[sly valve]

ok thanks

[simple ore]

limit game fps, so the GPU has a time for processing of other things

[chrome shard]

why does applio always say ran out of input

[rare goblet]

Ayo? @chrome shard level 2 !!!

[simple ore]

gradio perhaps

it is top tier garbage framework

but it is easy to make stuff with

[chrome shard]

how do i fix

im new to this stuff

[simple ore]

ignore it

[chrome shard]

i cant it errors applio

[simple ore]

just copy audio to local drive and put the path

[chrome shard]

applio dosent load

[simple ore]

what are you trying to do?

[chrome shard]

make a juice ai

[simple ore]

your model is damaged

or you're trying to load .zip or something

[chrome shard]

how do i delete the model

nvm i done it thanks

[low shard]

gradio is good, hopefully the weird ui on 5 got fixed as it got officially released

[simple ore]

uploading files been a shitshow for over a year or so

i dunno if 5 changes anything

hopefully it does

[scenic marlin]

Hello my hi ia has a bug that it doesn't want to open, how can I solve this?

[vapid moon]

Hello, my inference time is 500, how do I fix it?

[odd shale]

What's your GPU?

[odd shale]

!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.

[brittle wing]

any1 wanna perhaps help me figure out how many epochs do i need for my model with tensorboard

[lost lodge]

i keep getting this error

it was normal yesterday

but now it gives me error

any clue?

[lost lodge]

??

can anyone help me fix this error

[gusty scroll]

Is anyone else here having problems with RVC not working on Kaggle?

[rare goblet]

Ayo? @gusty scroll level 3 !!!

[lost lodge]

im using kaggle rn

[gusty scroll]

and it works normally?

[lost lodge]

no i keep getting this error

[rare goblet]

Ayo? @lost lodge level 6 !!!

[gusty scroll]

I only get that error if I try to immediately run it multiple times, be sure to close the pages you opened

[lost lodge]

i only ran it once and it gave me the error

[gusty scroll]

try clearing everything, closing the rvc and imjoy pages you may have opened and run it again just to be sure

[lost lodge]

Ok

[gusty scroll]

by the way, it doesn't work on my part too, but I'm getting a different kind of error

[mossy linden]

Hello, im super new to this. Im trying to setup a way to use this voice model #1238265015086088224 message but i cant find a way that gets me the results like the ones OP posted as examples. can anyone point me to a good guide i can follow for results like those? At first i tried using "Mangio RVC Fork Locally" using this guide, but couldnt figure out how to actually create audio. Then I used this "https://huggingface.co/spaces/TheStinger/Ilaria_RVC" illaria RVC but it sounded way lower quality than I want. Just being directed to a good guide would be much appreciated please.

[snow hazel]

Is there a good website that does good with robotic voices?

[lost lodge]

it gave me error

different one tho

[lost lodge]

your method worked

but when i enter the web

it says error

[gusty scroll]

try copypasting here the error you get at the bottom of the notebook page

[gusty scroll]

and tell me if the last cell is still running (buffering)

[lost lodge]

Ok

this???

[gusty scroll]

ImportError: /kaggle/tmp/training/.venv/lib/python3.10/site-packages/torch/lib/../../nvidia/cusparse/lib/libcusparse.so.12: undefined symbol: __nvJitLinkComplete_12_4, version libnvJitLink.so.12

[lost lodge]

ImportError: /kaggle/tmp/training/.venv/lib/python3.10/site-packages/torch/lib/../../nvidia/cusparse/lib/libcusparse.so.12: undefined symbol: __nvJitLinkComplete_12_4, version libnvJitLink.so.12

[gusty scroll]

yep, seems to be the same one I'm getting. At least now we know it's a common problem

[lost lodge]

yeah

did u find a solution for it?

[gusty scroll]

I already notified the creator and they said they will check it in a bit. Hopefully it's nothing big

[lost lodge]

yeah

hope its an easy fix

[hexed wedge]

#✨│ai-help how can i find voice modelss ?

[mossy linden]

https://discord.com/channels/1159260121998827560/1175430844685484042 go to this channel and search

[rare goblet]

Ayo? @mossy linden level 1 !!!

[hexed wedge]

i try but 404

[hexed wedge]

you are not authorized to post messages on this channel

system said that : you are not authorized to post messages on this channel

[lost lodge]

how long did u tell him?

[gusty scroll]

they said it's strangely working fine for them, but they can only test it properly in 12 hours

[lost lodge]

oh ok

[crude bolt]

anyone here having rvc working on focusrite solo.

the problem is with the mic selected the input isnt picking up anything - however the mic is audible through discord.

[brittle wing]

@unkempt shoal

[rare goblet]

Ayo? @brittle wing level 1 !!!

[unkempt shoal]

??

[crude bolt]

@unkempt shoal did you have in the past problems with your focusrite solo using w-okada ?

and was there a thing to get it working ?

[brittle wing]

^

[unkempt shoal]

Yep, how I managed it to work?

Uhh

just throw the Focusrite in the garbage

[brittle wing]

?????

[unkempt shoal]

I'm just kidding

But

[brittle wing]

no way

[unkempt shoal]

I ended up reinstalling the drivers

And it worked

If that doesn't work, then rollback into a previous version

If that doesn't work, throw it out.

[crude bolt]

yea the only thing so far - atlas did reeinstall just the driver (or ran the latest update) ... he hasnt restarted the pc (yet) - think he is doing that now

[unkempt shoal]

How.. did you all know I had a Focusrite, and that problem?

[crude bolt]

search on discord keyword "focusrite solo" - you mentioned : 23/01/2024 17:03 -Looks like my focusrite solo was the problem

so thats how we just pinged you 😛

[unkempt shoal]

ohh

[brittle wing]

yess

[crude bolt]

and we saw you online 🙂 - worth a try poking you

[modern surge]

-colab

[azure marsh]

☁️ Google Colabs

• 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, -overtraining, -realtime, -rvc, -help

[cerulean sage]

How do I do mangio crepe locally

How do I do mangio crepe locally

How do I do mangio crepe locally

How do I do mangio crepe locally

How do I do mangio crepe locally

How do I do mangio crepe locally

[pastel oak]

Whats the point in spamming the question, stop it

Download applio, it has mangio's crepe. Its just called crepe in there
https://docs.ai-hub.wtf/

[brittle wing]

where is "FIX\song_output\random_number"?

[hallow dragon]

hey guys i need help, my voice has a static when i'm using this voice changer can someone help?
is there any configuration to solve this?

[azure marsh]

Hey, Ochi! 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

[sage mango]

we can only train models having a nvidia gpu? doesn't work in amd/radeon?

[low shard]

You could technically on AMD with Zluda, even if ofc Nvidia GPUs are the best

What's ur AMD GPU?

[knotty moth]

check Applio guide to use under ZLUDA

[sage mango]

i have crossfire 2 Rx580 8gb

[rare goblet]

Ayo? @sage mango level 1 !!!

[low shard]

Not sure if that's good enough but u can try https://docs.applio.org/getting-started/installation#amd-gpu-support-windows

[sage mango]

I'll try, thanks for the help

[low shard]

Yw

[simple ore]

580 is not going to bring any joy

no fp16, need to use old torch version (2.2.1)

[glad zealot]

fixed now, you need to re copy the notebook or just replace the 2nd cell with the new 2nd cell

[lost lodge]

Thanks!

[brittle wing]

-hf

[azure marsh]

<:huggingface:1179800228946268270> Hugginface Spaces

• UVR5 UI, by Eddy and Ilaria Huggingface Spaces
• Ilaria RVC Zero, by thestingerx Huggingface Spaces
• RVC⚡ZERO, by r3gm Huggingface Spaces
• Applio, by IA Hispano Huggingface Spaces
• 🆕 FaceFusion UI, by Nick088 Huggingface Spaces

More commands: -audio, - colabs, -kaggle, -local, -overtraining, -realtime, -rvc, -help

[brittle wing]

-colab

[azure marsh]

☁️ Google Colabs

• 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, -overtraining, -realtime, -rvc, -help

[hollow rune]

-gradio

[rare goblet]

Ayo? @hollow rune level 1 !!!

[hallow dragon]

Can someone please help 🥲

[rare goblet]

Ayo? @hallow dragon level 1 !!!

[hasty warren]

Guys, I have a friend who has a card from AMD Radeon R9 200 series. What should he download from here ||https://huggingface.co/wok000/vcclient000/tree/main||

[rare goblet]

Ayo? @hasty warren level 1 !!!

[pastel oak]

Wrong help channel btw

[low shard]

hopefully people will use the right help channels one day

i see alot of people asking for wokada help in this channel

[pastel oak]

maybe we should rename wokada to mmvcserversio. that word can be found when you open the folder of wokada at least

or the simple "voice changer"

[low shard]

like if the avg user can read here, avg user justs calls it demo voice changer or smt

[low shard]

But technically RVC can be used as voice changer too, just not good for realtime, so that's technically not correct to say

[pastel oak]

you just unlocked a memory of people asking "my voice changer says demo can i get full version"

what have you done to me

[pastel oak]

yeah thats also true

[low shard]

😭

[austere osprey]

Error

[odd shale]

!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.

[knotty moth]

https://tenor.com/view/error-windows-glitch-gif-5012719

[low shard]

😭

[austere osprey]

ERROR VCCC

[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.

[low shard]

if u just say error no one can understand u

ELABORATE

[lost lodge]

what error

[pastel oak]

think hes 9 years old 😭

[low shard]

😭

[odd shale]

😭

[main junco]

i'm not

[rare goblet]

Ayo? @main junco level 1 !!!

[lost lodge]

i didnt understand wheres the new 2nd cell

[pastel oak]

no one was talking about u

[low shard]

😭

[lost lodge]

how do i get the new 2nd cell?

[lost lodge]

fr 😭

[dim pollen]

Hey all, I was wondering how can I make a model for a live speaking voice? I want to make one in spanish and make it sound as clean as possible (also laughing and stuff like that). I have no experience with getting a dataset or training them tho

[fresh cairn]

@wispy lodge was wondering if you could help me figure this one out? i can not figure it out after looking around online.

[lost lodge]

new error :(

how do i upgrade

[simple ore]

not an error, nothing to fix

[gilded frigate]

Having messed around with python before it may be better to just not upgrade it.
In my experience with some projects it kept yapping and complaining and erroring.

[lost lodge]

wait, its just a text?

im stupid then ;-;

[uncut folio]

hey im trying to finetune text to text AI and now i wanted to create my own text to speech or speech to speech AI and anybody knows how to do it becouse everytime i tried it failed and in the internet there are no good tutorials

[low shard]

So, you want a speech to speech, RVC, model right?

What's ur PC gpu

[low shard]

What's ur PC gpu

[dim pollen]

RTX 3070

[low shard]

As you got a good PC, you can use RVC locally, you can choose between:

• Applio: A fork of RVC with some extra features like Applio TTS, same quality tho
• Mainline: The original RVC

[dim pollen]

Is there a guide or something? Since idk where to start or do anything

[rare goblet]

Ayo? @dim pollen level 1 !!!

[dim pollen]

Oh nvm I saw, but how do I get a good dataset?

[simple ore]

it is something the developers need to take care of, as a user trying to upgrade gradio would be a never ending road of grief

[low shard]

The links are guides

[low shard]

You need to make it urself https://docs.ai-hub.wtf/rvc/resources/datasets/

[halcyon sable]

guys Hello everyone, I'd like to know how to produce the output file, as this is a model I got with good enough loss from the training. I got the G_5200.pth which is about 430MB and I like to get the smaller version colab_NoUI generates when you finish your training for example modelname_e200.pth that is about 60MB

[simple ore]

just select the checkbox to save the model every time you chose to save the D/G weights

[halcyon sable]

I can't retrain

I just have the file

[simple ore]

otherwise it saves only at the end

well, the model can be extracted from G file

[halcyon sable]

exactly, thanks, thats what I'm looking for

but I see in the commands they do cp weights/modelname_e200.pth drive/.../modelname.pth every time, how do they get these weigths then from the G file? what do I do to get it the same way they do from the G file

what precision do they use and parameters to extract that from the model?

[simple ore]

Retrieval-based-Voice-Conversion-WebUI\infer\lib\train\process_ckpt.py

extract_small_model function, I think

[halcyon sable]

thank you my man you were quick simple n useful. thanks 👍

[rare goblet]

Ayo? @halcyon sable level 3 !!!

[simple ore]

savee does the same if you pass it the config

it takes all the settings from hyperparmeters, the other function just have a hardcoded set, which should be the same.. maybe not, just check

[halcyon sable]

all good by now, but I have a question about this... what value is if_f0

[halcyon sable]

hahah bro it worked! it got saved in the weights folder. thank you for helping bro, I barely understood what I was looking for and I'm glad it worked.

did it this way...

thanks @simple ore !

[lost lodge]

oh

its been 4-6 days trying to make a RVC Model 🥲

[uncut folio]

Geforce RTX 3060

like text to speech and speech to speech and text to text like chatgpt but with my data

[lost lodge]

which developer do i ping to fix this glitch

[uncut folio]

im trying it on my schools server but i dont see progress with it

[rare goblet]

Ayo? @uncut folio level 1 !!!

[lost lodge]

(just so i wont ping the wrong one)

[simple ore]

it is not something that needs a fix