✨│ai-help

[low shard]

Weren't u kicked for telling poopmaster to touch that zap thing for mosquitos and shock himself?

[karmic oxide]

no i was never kicked

LOL

[low shard]

Prob u left urself then

[proven hill]

AUAHUAHAHUAHUHAUHAHJA

[low shard]

It happened after that

[karmic oxide]

if you are going to make baseless claims actually fact check everything

[low shard]

Poopmaster told me himself other stuff argued with you about the zap thing

[karmic oxide]

yes people argued about something, that everyone said not to do, including me in the messages

[proven hill]

wake up
argue with arcelia
zap myself

mfw

[karmic oxide]

anyways as i was saying, the only thing the fork does better is for amd and intel arc. its not even better for nvidia, its the same at the best case scenario

[low shard]

This is a help channel, there's no point going to argue here

[karmic oxide]

as you bring up stuff on ur own!

[low shard]

I already know this argument is useless since y'all were the same trying to forcefully tell @pastel oak to put the original wkada above the fork in the guide list

[karmic oxide]

man i dont think you should be a sr mod, you are letting ur emotions kick in! must be weird having a mod team with literal children on it!

no i never did

[proven hill]

what

[odd shale]

what the

[karmic oxide]

i literally told him his issues in his guide, told him how to fix it, i havent reached out to them in like 8 months

once again, if you are going to make claims, please fact check them

[analog obsidian]

[proven hill]

me rn

[karmic oxide]

if anyone has reached out to them on my behalf, it still isnt me, since i genuinely dont care about it

i even offered the entire original guide over at one point

but go off king

[low shard]

Anyways

No point in arguing here as I said

This is an help channel

People are here to get help not to see some arguing

[proven hill]

on a rampage for telling him theres stuff to fix?

[karmic oxide]

i went off of misinformation in the guide, which i had them fix

!!!!!

thats not doing the claim u said

of "put the guide above!!!"

[odd shale]

Oh no

[low shard]

.

[low shard]

.

[low shard]

.

[pastel oak]

Nah the fixes were fine i mentioned that too
Was after that but aint that deep anymore rlly

[low shard]

Case closed in this channel

[odd shale]

Hello Shad my dear mod/helper

🐢

[pastel oak]

This convo is going past each other

[proven hill]

in what channel can we argue then?

[pastel oak]

Lmaoo

[pastel oak]

🥬

[odd shale]

🐢: bites the lettuce

[distant musk]

[proven hill]

i mean, i wanna watch

im there for the memes

[pastel oak]

[distant musk]

https://tenor.com/view/abowww-amauril-gif-24427565

https://tenor.com/view/yessir-4t-4tomistalt-4tomist-4tmeme-gif-14459048726828889418

[brittle wing]

i trained a model on RefineGAN and 441k on the main applio branch and its not showing up on my inferencing model list

is there a switch i have to toggle in order to get it to inference in applio mode?

it's probably because the final model pth didn't save

i literally lost a model...

[prisma kettle]

Can anyone help me out with UVR? I tried doing the vocal splitter and just got a peaking buzzing sound... then I tried another model and there's weird beeping sounds in my audio

Like distorted beeping

[pastel oak]

Post a screenshot of your UVR interface before hitting the convert button in here

where you can see which model you used etc.

[prisma kettle]

1 minute I'm testing to see if it's a problem with my machine or overall

Shiiit I gotta eat dinner

[thorn abyss]

i have a dataset that's at 44.1k, but theres not a sample rate of 44k on applio, should i train with 48k or 40k?

[pastel oak]

Well im about to head off so ill just drop a temp fix in here based on assumptions

• Assuming vocal splitter is splitting vocal from background music, I hope youre using a good model like bs roformer or mdx23 which are recommendable. If you havent messed with settings the default ones should work just fine

Else, tho uvr is great and runs on my pc i prefer to use https://mvsep.com/en with bs-roformer viperx the newest one listed, think 2024.08 version. Create an account which is for free, but lets you skip 95% of queues and get your splitting ready in usually 3 minutes

[pastel oak]

This is a split opinion thing. To keep it short and summarize the core argument:

People that say use 40k: because upsamling to 48k means the ai has to generatively fill frequencies your dataset doesnt have. Can sound worse

People that say use 48k: just upsample because you got the range of 40k to 44.1k which is important. Upsampling doesnt damage it too much

Maybe someone else thats analysed these things can give a definite argument for one of them but thats what I heard from most

[thorn abyss]

i can use wtv i think the best right?

[pastel oak]

Pretty much

Id try both if you got the time

[thorn abyss]

i prolly stick to 40k even that i'm loosing some frequencies

the datasets like 30 mins

that gon take a while

but imma try both

thank you

[prisma kettle]

Thank you, I'm back now and will try these things. I was doing the "Vocal Split Mode" found in UVR's Additional Settings, which I think is meant for splitting lead/backing vocals. Using the UVR BVE 4B SN 44100-1 model. Kinda weird that it is hidden in the settings. Is that mode janky?

[prisma kettle]

Yeah something's up with UVR on my other machine, it was bugging there but fine on laptop

[languid oracle]

What realtime voice changer should i use for my amd gpu

[odd shale]

You can use deiteris' w-okada fork.

https://rentry.co/ForkVoiceChangerGuide

Scroll down and download the AMD version.

[shell sonnet]

Hey everyone, unsure where to ask so apologies if this is the wrong place

I’m a little behind when it comes to RVC, and it seems like there’s been a lot of developments. I’ve been just using mangio RVC for as long as I can remember

But, now I’m seeing so many options, like weights.gg, Applio, Voicecraft, W Okada, F5, even seeing Sovits reappear again

Where should I go to relearn my stuff?

Or, what should I be using?

[elder wren]

Looking to make ai song covers (mostly for a bit in a video)

After reading the guide I have a few questions.

1. Which audio isolation program is reccomended.
2. If I download the original RVC will it be able to do the song cover or do I have to download an extension for that feature
   3.specs for local training ai can't find it

[simple ore]

depends on what you want to do

[elder wren]

Crepe vs harvest which is the better option for songs

I got it to work!!!! But the vocals sound rough maybe I need to isolate it better any suggestions

[shell sonnet]

im looking to just clone voices. I am not as interested in TTS

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

Ive been looking at this guide, but im still a little overwhelmed by all these choices

It also just doesnt seem to cover exactly what I should use on a case-by-case basis. Moreso a comprehensive beginner’s guide on what things like “epochs” are

[knotty moth]

to break it down, what is your use case, inference or training?

[knotty moth]

still using harvest, not rmvpe? I suspect you might be using the old tiger18n's rvc gui

[shell sonnet]

I believe i need both

[elder wren]

Oh I saw rmvpe but my gpu is a 7900xtx so I thought it could handle the other options

[knotty moth]

harvest only runs in cpu mode, and rmvpe is the sota

[thorn abyss]

can i use tts with a rvc model?

[hallow thistle]

Applio has TTS option.

[thorn abyss]

i never figured out how to use it with a custom model

[simple ore]

Applio TTS is a demonstation. It is text -> Microsoft Edge screen reader API (online) -> tts output file -> rvc voice change -> output audio

[thorn abyss]

yeye i get it

but i'm getting an error

it says "Value: 0 is not in the list of choices:" and then a bunch of audiofiles name

[simple ore]

show a screenshot of the error

[thorn abyss]

aight

im using the hugginface space btw

im currently installing applio cause maybe theres a problem with hugginface

idk

[young halo]

Ok covers colabs definitely hate to see me coming

[young halo]

What Is that space for? I didn't know they had one

[thorn abyss]

inference

https://huggingface.co/spaces/IAHispano/Applio

[young halo]

I'll see if this one works for me 😩

[thorn abyss]

this one works pretty much all the time

but it has like a limit

or a cuota

[young halo]

That one didn't work either 😭

[thorn abyss]

bruhh

how

what you trying to do?

run inference

or train

[young halo]

I guess I'll never now how my model sounds...

[young halo]

Inference

[thorn abyss]

you tried ilaria rvc?

hold on

did you included the .index file in the zip file

try uploading your model to hugginface

in a zip file with your .pth and your .index files

[young halo]

Yeah, i did that already

[thorn abyss]

do you want me to try it out for you

send me a sample that you want me to run with your model

[young halo]

Please and thank you 😭
https://huggingface.co/ivoryprincemilo/WILLOW_COPINGMECHANISM_era/resolve/main/willowcopver1.zip?download=true

[thorn abyss]

it's all good

[young halo]

It's not fully trained yet, i wanna do around 400 and 500 epochs, but i wanna know how it sounds so far with 200

[thorn abyss]

you should check tensorboard while training

so you can tell when it's starting to overtrain

[young halo]

The graph showed somewhere around 1k, I'm not sure exactly how it works... Overtraining is the lowest point on the graph, isn't it?

[thorn abyss]

yep

you looking at that on loss total right?

and the smoothing set at 0.987

this .5 index rate and default pitch

[young halo]

Damn...

I'll probably have to remake my dataset

[thorn abyss]

its just a lil noisy

[young halo]

Yeah, but it wasn't so easy to clean up since the album i made from is full of distortion and reverb effects...

[thorn abyss]

yeah, makes sense

what you using to denoise your datasets?

[young halo]

USMVP or something like that

[thorn abyss]

mvsep?

[young halo]

YEAH, THAT, i didn't remember the name 😭

[thorn abyss]

what model are you using?

cause theres a lot of them

[young halo]

I used UVR before, but it didn't have all the models i needed

[young halo]

For denoise and stuff?

[thorn abyss]

mainly to isolate your vocals

what model are you using?

or what's your method to extract vocals

[young halo]

Let me check

I tried these

[thorn abyss]

okay

hold on

imma dm you

[young halo]

Ok!

[gilded flicker]

Hey

why does my program keep saying trial?

[thorn abyss]

which program?

[glad zealot]

Uninstall it and go back to the download page of vac, download the lite version instead of the trial version

The lite should be at the bottom

[brazen spear]

Ok i will restart thx man

[knotty moth]

https://software.muzychenko.net/freeware/vac470lite.zip
you should use this direct link instead of the website itself that may mislead you to downloading the trial one

[topaz zinc]

it's good! much better

[thin edge]

which pretrain is the best for Mainline RVC ?

[simple ore]

default one is fine

unless you need singing and stuff

[thin edge]

okay

and

where can i find "training/weights" from "For syncing graphs you need to train however many epochs you have set you save frequency then go into the file manager and find your model which should be in training/weights"

[simple ore]

mainline rvc is dumb so synching graphs is done this stupid way

you can just set the value to 50 steps and be done with it

or just spitball it to number of samples/batch size

in mainline rvc the model is in /assets/weights

[gilded flicker]

ty ser

[gilded flicker]

ty ser

Anyone here have connections to big gaming youtubes etc?

[tame mica]

what

tried the colab jst for main branch lol

screenshot got cut off before oops

[analog obsidian]

Klm uses old refinegan generator, it doesn’t work in the latest update

Use this https://discord.com/channels/1159260121998827560/1327283027776245810

[tame mica]

oh

got it

guess i gotta downsample

sadge

ill try tmrw ig

https://cdn.discordapp.com/emojis/1335956961434533932.gif?size=48&animated=true&name=tetodance~1

[brittle wing]

is ai cover broken again?

@thorn abyss would you mind making my cover too?

covergen has a problem rn

[polar tendon]

Yeah i think it's completely broken

Not working either for me

Also does the original RVC WebUI work with current python versions?

[crude flame]

nope

you just have either spaces in the file name or the file that youre trying to infer with isnt there

[polar tendon]

The WebUI on colab broke

[brittle wing]

everyone is having this issue

[crude flame]

you sure? it doesnt look like it did from that ss

[brittle wing]

can you use the google collab webUI than

pls

[polar tendon]

That's NoUI

[brittle wing]

and try to output a song

[polar tendon]

WebUI's broken on Colab.

[brittle wing]

I know I agree with you

just that if Razar said it isnt than maybe he can go to WebUI and try to output us a song

[polar tendon]

I think he didn't see it's NoUI.

[brittle wing]

so do we use NoUI rn since WebUI broken?

[polar tendon]

that you were running

[polar tendon]

If NoUI even works then i guess

doesn't seem like it tho

[brittle wing]

so nothing works? the only ai converts that do work rn is weight but theirs arent so good

[polar tendon]

And i do not know how to fix it

[polar tendon]

I suppose you have to extract the vocals, so download the song and use MVSep with SCNet while also having premium usage turned off, then extract the back vocals, run the main vocals through RVC and combine them.

[brittle wing]

everything u said is new to me, what

before I go up and learn all that are they all working rn?

[polar tendon]

Yes.

MVSep is working.

[brittle wing]

is the process longer and more complex than webUi?

[polar tendon]

And as of RVC, use Huggingface

[brittle wing]

also is it on par atleast with WebUI?
webui was the best tbh for ai cover converting for me (the output always sounds the best)

[polar tendon]

Unfortunately that's so, but i suppose if there's not a better solution (and you're planning to use several songs), then i guess.

[crude flame]

Dont use SCNet use bs or mel roformer since thet are better

[crude flame]

you can do it locally depending on your pc, you can use huggingface spaces and yeah weights

[brittle wing]

I use weight just now, but it sounded terrible

only very few songs ever come out satisfied in weights for me

[polar tendon]

The huggingface AICoverGen works, but you can only use a few songs and have a few goes at it before you run out of GPU quota or however it's called.

[brittle wing]

can you do it for me? I dont have alot of space left on my C disk and my GPU cant even load a TTS model for me

its just 1 song

[polar tendon]

What's the AI model link?

I'll DM the result

[brittle wing]

ill send you my hugging face model and the song?

[polar tendon]

Sure

[brittle wing]

here? or dm

[polar tendon]

Your choice really.

Both works

[brittle wing]

i dont mind here if its safe

I got nothing to hide

https://huggingface.co/idontgoddamn/TendouAlice/resolve/main/TendouAlice.zip?download=true

https://youtu.be/MDc1mjrIsPM?si=Pj8LSmkvJ2dH3xWS

@polar tendon here

[crude flame]

it works

i just tested it

[brittle wing]

what

how

[crude flame]

you did something wrong

[brittle wing]

what I do

[crude flame]

idk

[brittle wing]

wait ill do it again and than if it happens again ill send the ss of the model used and song generated

[polar tendon]

If it errors out, send a screenshot with the parameters used.

[brittle wing]

👌

[stiff goblet]

What is the recommended batch size for 12 min dataset audio ? (medium quality, 16Khz cutoff)

[polar tendon]

OH WAIT I KNOW WHAT WAS WRONG

[crude flame]

@viscid moss There is a bug with the colab version of your Ai cover gen where if you use split audio it gives an error but when you dont split the audio it works fine

[crude flame]

depends on the dataset

[brittle wing]

what was wrong?

[stiff goblet]

Thx

[brittle wing]

[polar tendon]

colab's ip was getting blocked from youtube based on the screenshot

[crude flame]

what colab are yall using?

[polar tendon]

HTTP Error 403: Forbidden

[polar tendon]

I looked at the screenshot again

[brittle wing]

youtube angy?

[crude flame]

https://colab.research.google.com/github/Eddycrack864/RVC-AI-Cover-Maker-UI/blob/main/assets/RVCAICoverMakerUI.ipynb try this one

[brittle wing]

yea it happened again

💢

[viscid moss]

Right

[brittle wing]

this is new to me what, How do I use this

[viscid moss]

Download the audo for ur own

Leme send u a guide

[brittle wing]

you dont like copy paste hugging face model address or put youtube link

[viscid moss]

.

[viscid moss]

https://docs.google.com/document/d/e/2PACX-1vThk7Qo7yCWNVbxOmahl2R8_Jgi6TFuMBUIi-PWre_HIN0lFTq-dr37Rh5iJlGgYb_vFapXMHt2W8Kp/pub

[polar tendon]

It's a WebUI apparently

[viscid moss]

The problem it's just the audio download

If u put a path to the audio file it will work

[brittle wing]

I dont know what to be doing here..

It doesnt ask for a link of the video I want my ai to convert to

or asking for my ai model address

[brittle wing]

I tried other songs same thing

or is it just all youtube no longer works now?

youtube went that hard?

[viscid moss]

YT

[brittle wing]

[polar tendon]

Colab is getting blocked from YouTube, nothing gets downloaded

[viscid moss]

⬆️

[brittle wing]

so no more youtube links?

[viscid moss]

nope

Lemme edit that to:

"Place the path to audio file"

[brittle wing]

how do I put an audio file

[viscid moss]

Gimme 5 mins

[viscid moss]

https://docs.google.com/document/d/e/2PACX-1vThk7Qo7yCWNVbxOmahl2R8_Jgi6TFuMBUIi-PWre_HIN0lFTq-dr37Rh5iJlGgYb_vFapXMHt2W8Kp/pub

[polar tendon]

File Browser

[simple ore]

that's the split audio code I've not seen in a very long time

[brittle wing]

step 3 says place a youtube link

[viscid moss]

Well, I just take the ownership and haven't tried all that stuff. I'll check that

[viscid moss]

Go to Optional :p

Guide also needs to be updated

-colab

Damn, still down

no way

[polar tendon]

The bot?

[brittle wing]

doing it rn

gotta file upload the mp3

[viscid moss]

ye Automaze

[polar tendon]

Ah

Also does the original RVC WebUI work with current python versions?

[viscid moss]

nah, I think only works with 3.9 and 3.10

[polar tendon]

Any specific pip versions i should look for?

[brittle wing]

@viscid moss work

[polar tendon]

And 3.8 as in the README

[crude flame]

💀

[polar tendon]

Now you know what to do when YouTube doesn't work.

[brittle wing]

why youtube do this

[viscid moss]

I told u. It's only that YT sh!t

[polar tendon]

Because you aren't really supposed to just download the videos, YouTube doesn't like that.

[brittle wing]

[viscid moss]

real

[brittle wing]

what

[viscid moss]

Not completely sure tbh

[viscid moss]

WTF is that error lmao

[brittle wing]

what went work

the ai cover was success

THATS WHAT IM SAYING

[viscid moss]

lmao

ignore that

[brittle wing]

ignore how

bro it wont let me get my thing

[viscid moss]

Just press OK

[brittle wing]

ok

still didint get it bro

[viscid moss]

lmao Colab being Colab

[brittle wing]

its right there why wont it let me bruh!

[viscid moss]

try to copy that file into ur Drive to save the cover

[brittle wing]

[viscid moss]

rename the file

[brittle wing]

to?

[viscid moss]

Cover.wav :p

[brittle wing]

how do I do this

i renamed it to Cover.wav

copy path

how do I paste it to my drive

[viscid moss]

To copy to ur drive use this command
Crate a new cell of code and replace:
!cp cover_path drive path

to get the cover path do right click on it and press copy path

same for Drive path

[brittle wing]

bro what

all I understood was to copy path the output

everything else I dont get
i never learned to use google drive

[viscid moss]

ur Drive path is basically the path that u place on SONG_INPUT

just delete the audio name

it will be something like this

/content/drive/MyDrive/

I'm editing that colab notebook rn

[brittle wing]

wtf did I do-

its all gone-

[viscid moss]

hahaha check ur Drive

[brittle wing]

empty

[viscid moss]

Damn, at this point just restart the colab

I'll also try to make that cover

[viscid moss]

I'm also Neuro-sama fan

:p

[brittle wing]

I press upload

and than it left

[viscid moss]

lol

maybe VC? I can't talk but u can follow my steps

[brittle wing]

ok

[supple moss]

Thanks guys I was able to use spikes voice to sing my way :3

can I get your guys opinion? and how do i upload a mp3 here?

[naive crow]

can i get some help rq

[oblique viper]

are there any datasets I can use to try out applio?

I want to figure out how it works in the first place before I commit to making a dataset

[lone nest]

Is there a filesize limit on Applio colab?

[lone nest]

Grab some from https://www.sounds-resource.com/ ig

[low shard]

what's the issue

[oblique viper]

hey do I have to stop training my model myself or will it stop?

if I have to stop it manually then how long should I leave it for?

I'm talking about this part

and if you can explain what generate index means that'd help

[stiff goblet]

Is contentvec is better than the original hubert base?

[simple ore]

same

[odd shale]

Both are the same.

[pallid patio]

There are colabs working?

[odd shale]

Of course, Applio is one that works till this day.

Tho what's your GPU?

[oblique viper]

what settings should I set for training a model in Applio VRC

[oblique viper]

hey can you help me with putting good settings in the Applio VRC thing?

or if there's any guide for the settings

[odd shale]

All i'll suggest on applio is playing around with pitch and use clean audio to infer with models.

https://docs.aihub.gg/

There you also got a guide.

[unique rock]

can someone help me?

[oblique viper]

why is the Applio VRC training the model using my CPU and not my GPU?

the GPU should be better suited for it so I'm confused

lol

[charred heron]

Can anyone help with Refinegan I got errors during the "Save Every Epoch
" stage I would need to dm since I don't have perms 😭

[simple ore]

what errors?

[simple ore]

go to training tab and see advanced settings

[unborn jolt]

how do i convert models to onnx?

[oblique viper]

It says "Unfortunately, there is no compatible GPU available to support your training.", my GPU is 6700 XT

[simple ore]

https://docs.applio.org/applio/getting-started/installation#amd-gpu-support-windows

[charred heron]

Traceback (most recent call last) but more specifics are in dm

[oblique viper]

I already installed that

[simple ore]

then you did not follow the instructions

[latent kettle]

How do I access kaggle or colab notebooks

[low shard]

you don’t need to for the latest wokada deiteris fork

if you’re talking about realtime voice changer, that’s wokada not rvc, say ur pc gpu and the guide link u followed in #🔍│help-w-okada

[low shard]

which google colab notebook are u using

[wintry apex]

is there any way to fix the ping when running the deiteris okada alongside a game? I'm trying to run it and marvel rivals at the same time but my ping skyrockets in the voicechanger and makes it to where it just doesnt function at all. As soon as i close the game ping goes down to 0 and it works perfectly

[low shard]

That's because you're running 2 very intensive tasks at the same time
Also, this is the wrong channel
Show ur pc GPU and a screenshot of ur wokada while in game in #🔍│help-w-okada

[wintry apex]

oh ok i wasnt sure which one thank you

[latent kettle]

How do I access kaggle or colab notebooks

-colab

-kaggle

Why these commands are not working

[low shard]

the bot is down

[unique rock]

Rvc disconnected

[thin edge]

damn

[low shard]

rvc disconnected is not supported anymore

check #📰│dev-updates

it had a last update which should fix things, but it will NEVER get supported or updated again

Because it runs on mangio rvc, which is way slower and abandoned since 2023

what’s ur pc gpu

[thin edge]

bro should i wait until E300 ? (i write Total E 300)

Epoch: 57 [2025-02-06 11:24:29] | (0:00:10.485343)

sus af

[latent kettle]

Any alternative?

[low shard]

The bot is up again

[karmic olive]

<:aismug:1159365471368400948> WHO PINGED GRRR!!!!

• Client's average ping: 48ms
• Time passed since last ready: 2 hours 20 minutes 13.6 seconds

[low shard]

You can use the commands in #🤖│bots

[thin edge]

How much difference is there between 300 epochs and 500-800 epochs?

[low shard]

more epochs don't mean more quality

I think we already talked in #🔍│help-w-okada when u asked me which epoch to choose

U need to read the tensorboard to know that

https://docs.aihub.gg/rvc/resources/training/#tensorboard

[thin edge]

I've seen it, choose the one closest to OT, right? I just want to know what the factors are for model quality.

[low shard]

no, you need to first of all set it up like in the guide, then, choose the lowest loss/g/total value

the epochs and step won't mean more means more quality or smt, the quality all depends on the dataset lenght, quality and the way you read the tensorboard

[thin edge]

aight

[oblique viper]

at how many steps/epochs does a model start to sound good?

ik ideally it's right before the overtraining point

but if you don't reach the overtraining point at how much it should sound fine

[odd shale]

Actually there's no accurate answer for that.

All will depend on your dataset, training settings and the voice itself.

[tropic goblet]

https://discord.com/channels/1159260121998827560/1329255505872551936 anyone know why this doesnt work on w okada :/ ?

[oblique viper]

why is there no run-install.bat

I keep downloading the zip from the github and huggingface

both have no run-install.bat

that's annoying because I want it on my USB and there's no run-install.bat

I reinstalled the zip file from both hugging face and github

installed a brand new zip file and there's no install.bat file again

I'll try installing the install file straight from the github page

[simple ore]

maybe it is a hint you dont need run-install, hmm?

maybe it is a hint to just use run-applio.bat?

mystery

what would be the purpose of 6GB 'compiled' version vs 5MB .zip file with a source I wonder

guess we'll never know

[knotty moth]

you're using the precompiled release one, try clone the main repo which is the actual latest one

[oblique viper]

oh

[oblique viper]

https://tenor.com/view/lol-gif-23285124

nvm

[oblique viper]

thank you

[limber delta]

why is my processor loaded at 80% and the 3080ti at 1-5%?

[simple ore]

what app?

[limber delta]

MMVCServerSIO

win_onnxgpu-cuda_v.1.5.3.18a

[pastel oak]

Further questions ask in #🔍│help-w-okada

But delete your voice changer and download NVIDIA from this
https://rentry.co/Forkvoicechangerguide

Has optimized hardware usage

[brittle wing]

somebody know how do i fix the error "unhandlerejection" and TypeError: Cannot read properties of null (reading 'type')

[low shard]

tell ur pc gpu, what are u doing, what program are u using

[brittle wing]

nah already fixed but ty

[blazing solar]

-colab

[karmic olive]

📒 Google Colab Notebooks

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

ℹ️ Note

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

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

[zinc crow]

Buonasera @low shard, @proven hill. Non so se vi ricordate di me... Comunque volevo chiedervi come posso fare a creare una cosa del genere https://www.youtube.com/watch?v=1fWqJD0uXlQ. Attendo Vostre, a presto!

[low shard]

Mi sa che l'unico modo è lo stesso cantare come abbiamo detto prima

[zinc crow]

Non c'è un modo di allenare il modello?

Io ho sentito che si poteva allenare il modello AI per poi farlo cantare (o parlare) come il real

[low shard]

I modelli RVC sono Speech To Speech, se lo alleni e gli dai te che canti in input si funziona

[zinc crow]

Applio è buono?

[brittle wing]

Sì certo

[low shard]

Si

[distant turtle]

-colab

[karmic olive]

📒 Google Colab Notebooks

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

ℹ️ Note

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

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

[proper oxide]

does anyone know if I can test a RVC backup audio before it's done fully
im at like 50 epochs but im going for 300 with a dataset of around and hour and a half
i dont use discord a lot so tag me if you have any answers

[brittle wing]

anyone got a download for mac?

[tame mica]

yea

each epoch is fully finished so you can use it

except if it has a d/g in the front

xD

[low shard]

Explain what you need and what you want to do

[brittle wing]

Which Applio version works with KLM 5 Refinegan?

[tame mica]

main branch

should be latest

[brittle wing]

Applio 3.2.8?

[simple ore]

main branch from github

[brittle wing]

This one?

https://github.com/IAHispano/Applio/releases

3.2.8?

[simple ore]

it has not been released, so the "main" branch

[brittle wing]

Sorry, how do I find main branch?

[simple ore]

it is when you get when you download the code from the main repository

https://github.com/IAHispano/Applio/archive/refs/heads/main.zip

[brittle wing]

Thank you!

[austere drift]

can someone train a model for me, my system is not enough pls

[low shard]

you can use cloud

Train (make) RVC Models on cloud:

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

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio by Vidal (UI, no guide as of right now)
  • 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):
  • Applio (UI)
  • Mainline (UI, No guide as of right now)

3. Be sure to know about the tensorboard

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

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

You can use either:

• Weights.gg: Easiest Possible Ever Automatic
• Ilaria RVC Zero: Fastest free on cloud
• Applio (ui)

Else, u can do requests in #1159289738314919936 or #1191429836321849435

[proven hill]

5 dolla

[proper oxide]

well based off this run, its now 180 epochs, it hasnt loaded the pth file that isnt d,p

[tame mica]

yes you can use that one

[proper oxide]

but which file am i using if i cant see the main pth

[tame mica]

what

[proper oxide]

once I load a backup, do I just export it to save as a zip

[odd flower]

Why harvest is better than any other?

[tame mica]

are you trying to save it? then yeah

put the pth and index file in a folder then zip it

[simple ore]

who told you that?

[hallow thistle]

Rmvpe is way even better than that.

[sinful radish]

Can anyone help me ?
I have a 1:16m long audio from a Voiceactor in a game and ive been trying to figure out what the optimal training method would be.
i watched a tutorial from a guy that used a 2:40 ish long audio and got almost perfect results. Yet when i do that i get my own voice just cringy higher pitched as a result.
i try to give everything i entered into Mangio RVC
audio datasample length; 1:16m
Transpose : male 2 female -> 12
algorithm: rmvpe
training Epochs: total of 200 i guess

do i have to train longer or do i need more audio samples ?

[low shard]

YouTube tuts are old

Mangio RVC is abandoned since like 2 years

Mangio Is a Fork of RVC, a modified version which is now slow compared to other forks or the original mainline

There isn't a right amount of epochs, you have to train using high quality and long dataset and use the tensorboard

What's your PC GPU

[sinful radish]

gtx 1660 Ti

[sinful radish]

so what exactly do i have to use instead ?

[low shard]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

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 daily, not granted, of GPU

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

[sinful radish]

oh thx alot

[upbeat kindle]

@ashen dirge

[fast scarab]

Yo everyone, does anyone know how long it takes to train a voice on Apollo with 1000 epochs? Thanks in advance!

[thin tulip]

does AMD support this voice mod?

[odd shale]

Do you mean w-okada?

Because we don't give support to programs like voice.ai

[odd shale]

Actually training to 1000 Epochs on Applio would be overkill/and can lead to overtraining.

It's better to watch the tensorboard and stop training when the g/loss total graph goes up after a while.

[thin tulip]

im trying to download the VB audio cable, and before i do i wanna make sure its capatable with my GPU

[odd shale]

First of all, what's your gpu?

[thin tulip]

AMD Radeon RX 7700 XT

[knotty moth]

go to #🔍│help-w-okada and download the voice changer from the pinned guide there

[odd shale]

Hehe that's exactly what i was gonna say

[thin tulip]

thx

[scarlet wedge]

how to upload .pth file in weights folder?

there's not even a button to upload my model

in mainline

[knotty moth]

I have no more clue but using kaggle in mobile is a pita

[odd flower]

I trained like 30+ model for my own voice on Rmvpe but with the same dataset in the third try I got a perfect model of my voice using the same dataset, batchsize and on 280 epoch

[proper oxide]

which pth

[rotund oracle]

does anybody know any free ai websites that can generate unique and genuine expressive voices that sound real

[odd shale]

#1159290752195633273

[toxic tide]

hey guys, new to ai stuff, is it possible to create an ai song with replaced lyrics while keeping original voice?

[neat talon]

People, what is the most optimal number of eras for rwmp? Everywhere they say different things

[neat lintel]

also new to whole ai song/cover creating stuff, main focus is on creating rdr2 songs/covers, tried arthur morgan model on one song and it sounds off, distorted, unclear, loads of noise etc, i was using harvest as a method, does anyone have any insights that would help me navigate trhough RVC and how to use it properly, would be much appreciated

[tough dirge]

Heyyy i downloaded some models from #1175430844685484042 but none seem to be working in tortoise tts ? I got screenshots but i cant send em aparently

But it says this

error
Unsupported audio format provided: .pth```
inside my voice folder I got a pth and an index

the only folder that works is the default "random" which DOES have .pth inside so im confused why the ones i downloaded are just not supported

[simple ore]

1. tortoise tts does not support rvc voice models

2. tortoise tts is a slow garbage tts

[tough dirge]

any alternatives you got in mind ? 🥺
Im using it bc its the only thing i found that allows me to use in python scripts i make

i wanna change voice, text and output to different directories from the comfort of a python file

[low shard]

RVC is STS, not TTS natively

[tough dirge]

oh im silly, i forgot to mention i have audiofiles created based on the text, i just need to run it through rvc so its not a generic microsoft voice, but instead the voice of the model 😄

[neat lintel]

Guys what would be the ideal number to set crep hop to, i'm new so idk, also any other settings i should change so it works properly?

[low shard]

I think https://pypi.org/project/rvc/#description or https://github.com/blaisewf/rvc-cli could help

[low shard]

What are u doing and using and what's ur PC GPU?

[tough dirge]

ty ill take a look !

[neat lintel]

Ryzen 7 4800 h gtx 1650, i'm trying to do some ai covers/songs

[grand knoll]

Hi, I have trouble training a model, I have 3h good quality in French, I tried epoch from 100-1000, the results are robotic and no similar to dataset

[low shard]

Alright, you should use rmvpe as the pitch extraction since it's robust and most recommend one

Also, which program are you using?

[neat lintel]

I'm using RVC

[low shard]

You could also play with the pitch, lower means feminine, higher means more masculine

[low shard]

Yeah but there's different versions and forks, I'm trying to check out if u got a good or old one

That's why I'm asking, it's to be sure u don't got an old one from a yt tut

[neat lintel]

Oh i see, i think i have the latest version

[low shard]

May you link me the download link you used or tutorial?

[neat lintel]

Sure

here https://www.youtube.com/watch?v=oOBjntI2xK0

[low shard]

Damn that's really old

RVC GUI is a fork (modified version) that has been abandoned since 2023

YouTube tuts are old, don't follow them

[neat lintel]

yeah i'm on the latest one from 2023, like i said i'm new idk what i'm doing lol

[low shard]

Dw, uninstall that one you got, I will give you updated ones in a sec

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

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 daily, not granted, of GPU

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

For any clarification, ask me

[grand knoll]

did u suceed make rvc in other language as english?

[low shard]

Be sure that those 3 hours are good quality, quality matters a bit more, also be sure to be using the tensorboard

The original Pretrain has been trained on English only, maybe there's some french Pretrain but I don't remember if there was one

[neat lintel]

thanks a lot, will check them out

so every one of these should give me much better results than this old RVC that i used right?

[grand knoll]

the audio has been cleaned with UVR, tensorbord for loss? i use mathplot

[vestal cloud]

How do i get mel denoiser on uvr

[low shard]

These ones are more updated and otpimized than the one you got

[neat lintel]

Okay thanks, much appreciated!

[simple ore]

you need to start with explaining which RVC you're using

[grand knoll]

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

[simple ore]

f0 method?

[knotty moth]

you might be using the voice of multiple speakers

[grand knoll]

['infer/modules/train/extract/extract_f0_rmvpe.py', '2', '1', '0', 'D:\RVC\RVC\RVC1006Nvidia/logs/Lou', 'True']
todo-f0-1904
f0ing,now-0,all-1904,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_1.wav
['infer/modules/train/extract/extract_f0_rmvpe.py', '2', '0', '0', 'D:\RVC\RVC\RVC1006Nvidia/logs/Lou', 'True']
todo-f0-1905
f0ing,now-0,all-1905,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_0.wav
f0ing,now-381,all-1905,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_1724.wav
f0ing,now-380,all-1904,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_1722.wav
f0ing,now-762,all-1905,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_25.wav
f0ing,now-760,all-1904,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_2497.wav
f0ing,now-1143,all-1905,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_3224.wav
f0ing,now-1140,all-1904,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_322.wav
f0ing,now-1524,all-1905,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_3964.wav
f0ing,now-1520,all-1904,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_3957.wav
f0ing,now-1900,all-1904,-D:\RVC\RVC\RVC1006Nvidia/logs/Lou/1_16k_wavs/0_992.wav
['infer/modules/train/extract_feature_print.py', 'cuda:0', '1', '0', '0', 'D:\RVC\RVC\RVC1006Nvidia/logs/Lou', 'v2']
D:\RVC\RVC\RVC1006Nvidia/logs/Lou
load model(s) from assets/hubert/hubert_base.pt
move model to cuda
all-feature-3809
now-3809,all-0,0_0.wav,(149, 768)
now-3809,all-380,0_1359.wav,(149, 768)
now-3809,all-760,0_1721.wav,(149, 768)
now-3809,all-1140,0_2112.wav,(149, 768)
now-3809,all-1520,0_2496.wav,(149, 768)
now-3809,all-1900,0_2855.wav,(149, 768)
now-3809,all-2280,0_3219.wav,(149, 768)
now-3809,all-2660,0_358.wav,(149, 768)
now-3809,all-3040,0_3956.wav,(149, 768)
now-3809,all-3420,0_632.wav,(149, 768)
now-3809,all-3800,0_991.wav,(149, 768)
all-feature-done

[simple ore]

I just asked for the method you used, not the log dump

rmvpe is okay

so see the tensorboard output if you had it enabled

[neat lintel]

just tried illaria rvc zero with rmvpe and it somehow ended up sounding much worse than the RVC GUI from 2023 that i tried

[marble forum]

sorry please i have been away for a while. Please how can i download a model from #🔍│find-models to use in real time gui?????

[modern surge]

-rvc

[karmic olive]

📚 Documentation

AI HUB Docs

https://docs.ai-hub.wtf

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

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

[modern surge]

-colab

[karmic olive]

📒 Google Colab Notebooks

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

ℹ️ Note

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

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

[modern surge]

-local

[karmic olive]

Not available yet

[austere kelp]

what voice changer should I use for this

[hallow thistle]

For W-Okada the "realtime voice changer", go to #🔍│help-w-okada

[austere kelp]

where would i download it

[hallow thistle]

Because people keep mistaken "RVC" for realtime voice changer, I always do this everytime.

Again.

!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.

[lethal kindle]

yo i need some suggestions

I wanna troll my friends so does anyone reccomend a female voice model

[lethal kindle]

Heck yeah

[hallow thistle]

No.

[lethal kindle]

aw

gotta find my own

[polar tendon]

Update for Colab AICoverGen and everything else that could download YouTube videos on colab: YouTube blocked all Colab IPs (pretty much) and now it needs a token from a legitimate account to download videos

[glad zealot]

yup, using pupetteer could bypass it or some specific user agent but not quite sure

[toxic oyster]

Hey- so... can anyone help me with a small thing? I'm new to Voice.AI, and I've been trying to upload an avatar for a voice I downloaded. But it keeps saying "invalid image dimensions," the hell is that supposed to mean?

[knotty moth]

try download it manually

[abstract ivy]

Help

How to use rvc?

[pastel oak]

Are you trying to create models, make an ai cover, or use realtime for calls

[pastel oak]

We do not offer help for voice.ai, we offer local free open source options here

I can send you a link if youd like, would need to know your GPU first if its possible

[toxic oyster]

Oh- mb. Sure, tho. It would be nice

[abstract ivy]

I just want to make a sample because my device is still low

[pastel oak]

So an ai cover (converting an audio file into another with a model). I dont know what device is "low" means, I assume you mean you dont have a good pc?

The easiest to do online is via Ilaria RVC:
https://huggingface.co/spaces/TheStinger/Ilaria_RVC

An explanation on how to upload a model on there and converting samples ("inference") can be found on the guide
https://docs.google.com/document/d/1YbXcLFPaGjhOdG5NFkK3QrucCEpHZBwFUxkeMO8aB18/edit?tab=t.0

[pastel oak]

Still need to know your gpu name

[marble forum]

Anyone?????

[pastel oak]

Youre in the wrong help channel, for the future use #🔍│help-w-okada for realtime

Click on "View Model", it goes to weights.gg where you can download it from. Might need to be signed up

Or use #1175430844685484042

[marble forum]

Okay thanks

How do I go about creating a song with a cloned voice?

[pastel oak]

You can record the song yourself with your normal voice, and then use rvc inferencing to turn that into a cover with the voice model

There are apps like suno ai that create a whole song for you with beat and all, but unsure if you can select a specific voice model to be used, havent used it

[marble forum]

Is there a text to speech one? Like one I can write the lyrics instead and it just sings the lyrics with the cloned voice?

[hallow thistle]

You can't expect a TTS model to sing a whole song. TTS can only speak like a normal speech.

[marble forum]

not TTS bro that is why i asked if "there is anyone" because i am not talking about anyone specifically

[lost pewter]

can u help me my gpu is rx 5700 xt

[pastel oak]

For voice changer?

[lost pewter]

yes for rvc

just give me link if u can

and my gpu support it?

[pastel oak]

https://rentry.co/forkvoicechangerguide

Download AMD, virtual cable, read audio setup discord&games section

Your gpu will work fine

Any other questions ask in #🔍│help-w-okada

[languid tulip]

I did everything according to the instructions.(https://docs.google.com/document/d/1XuxQYiqEhYrdYeCZRRLrmV_ciMKo0bV-jTCGHu_-5Cc/edit?tab=t.0) But in the Training Model field it says this(AssertionError: You need to download a pretrain! Please run the "Download Pretrained Model" cell before continuing.)
How can I fix it?

[odd shale]

Emm.. I think that guide is outdated, and RVC Disconnected is not maintained anymore.

What's your GPU?

You can either try installing RVC Mainline or Applio locally if you got a decent GPU.

[languid tulip]

gtx 1060 3gb, good?

[shy blade]

quick question: what is the overall best quality base model to finetune from in terms of similarity to human speech?

[odd shale]

I'm not sure...

But i think you'll be at least able to install either Mainline or Applio to use models locally.

https://docs.aihub.gg/rvc/local/mainline/

[signal crater]

which rvc i should use for training female voice? I have RTX 4060, i can run in local,

should i go for mainline or applio?

[odd shale]

We mostly recommend you to stick to OG pretrain on RVC.

[shy blade]

that's the card I use, I found it can run small models that don't take a lot of vram, but there is quite a bit of delay if you don't want artifacts

[odd shale]

Oh, you were talking about realtime/w-okada?

[crude flame]

Applio

[shy blade]

more just general processing speed but yes

[odd shale]

In that case you can install Deiteris' w-okada fork

Here you got a guide

[shy blade]

I just assumed that's what the asker was requesting information on

[odd shale]

https://rentry.co/ForkVoiceChangerGuide

[odd shale]

Ah maybe i got confused.

[shy blade]

@languid tulip, did you mean realtime inference, or training (or something else)?

[odd shale]

He was talking about RVC training, not realtime

[shy blade]

ah, apologies

[odd shale]

Dw, it's fine.

[shy blade]

I failed to read above

[odd shale]

But for your case as of you mentioned delay, i guess you can use deiteris' fork for realtime.

[shy blade]

I'll check it out

what's the difference between these?

[shy blade]

but yeah, for training, the 1060 is worthless

[odd shale]

Certain versions are for a specified GPU.

[shy blade]

_[:21]

[odd shale]

Yeah, i feel like you're right.

[odd shale]

Oh yeah my bad, as GTX is nvidia, then scroll down on the guide and download the nvidia version

[signal crater]

guys where can i find a sample dataset of any female model?
first time am using applio so just wanna upload a training dataset and see how it goes

[crude flame]

there are no sample datasets, but ig you could use a voice from ears

[signal crater]

Wym?

[crude flame]

there are no sample datasets specifically for testing applio but you can grab a public dataset like facebook's ears and then grab one of the voices from there and test applio

[knotty moth]

go find the source and learn how to process & clean it

[brittle wing]

I’m having issues with KLM 5 model, not working with Replay Application?

[simple ore]

RefineGAN is not supported in other application yet

[brittle wing]

Any that you know of, that do?

[simple ore]

Applio, main branch, not officially released yet

[brittle wing]

Thank you

[languid tulip]

Why does nothing happen when I run the “TensorVENV.bat” file?

[valid spruce]

Help

Any collabs to train besides RVC Disconnected?

He's broken again

[low shard]

#📰│dev-updates rvc disconnected will never come back again, it got it's end of life last update

We don't support it anymore

[low shard]

Train (make) RVC Models on cloud:

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

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio by Vidal (UI, no guide as of right now)
  • 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):
  • Applio (UI)
  • Mainline (UI, No guide as of right now)

3. Be sure to know about the tensorboard

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

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

You can use either:

• Weights.gg: Easiest Possible Ever Automatic
• Ilaria RVC Zero: Fastest free on cloud
• Applio (ui)

Did u check ur PC GPU first?

[brittle wing]

Hey, I have a model already done, what's the easiest way to use them on TTS?

[brittle wing]

Using this Applio, how can I load a model there?

[valid spruce]

Thank u so much💙 🫂

[vestal cloud]

how do i stop training

[hoary spear]

how do i fix this error? 'NoneType' object has no attribute 'setdefault'

I'm using a YT link tho

[viscid moss]

CoverGen Colab?

[hoary spear]

I'm using this https://colab.research.google.com/github/hinabl/AICoverGen-Colab/blob/main/Hina_Mod_AICoverGen_colab.ipynb#scrollTo=NEglTq6Ya9d0

[viscid moss]

ye same

Colab blocks YT downloads, u need to upload the audio manually

[hoary spear]

oh

that's fine ig

[vestal cloud]

how do i stop training with rvc?

[simple ore]

ctrl-c in terminal window

[lusty barn]

Can someone help me with setting up audio

[low shard]

Yw

[odd shale]

Setting up audio on.. Realtime?

In that case head over to the #🔍│help-w-okada channel

[lusty barn]

I fixed it

[odd shale]

Oh, nice.

[low shard]

There are different Text To Speech (TTS) AIs:

GPT So Vits: RVC isn't as good as GPT So Vits for tts, but gpt so vits (few shot tts, which means needs just a lil training for models) can't use rvc models (and viceversa), and its only limited to: english, chinese & japanese, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/

Freemium 11labs: An easy way to do TTS is https://elevenlabs.io/, you can't use RVC model on this but its a mostly premium easy way for good quality TTS

FishSpeech: FishSpeech is a 0 shot (no explicit training needed) TTS, if you got a good pc you can use it locally else use their site

With RVC Models:

RVC is natively for Speech To Speech, but forks such as ilaria rvc mainline & applio have built in tts (using Microsoft Edge TTS to make a generated tts audio, which i suggest you to choose a tts model that is the same gender and language of the rvc model you wanna use, and then convert it with rvc)

If you wanna do tts locally with RVC Voice Models (if you got a good pc):

• You can get Applio in our docs
• While Ilaria RVC Mainline here (no guide as of right now)

If you don't got a good pc you can do tts with RVC Voice Models on cloud:

• Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide

• Use Applio UI Colab (with google colab T4 free daily limit gpu)

• if you don't wanna use edge tts, you could try another tts ai from our tts index and use the output as an input in rvc

[grave frigate]

Hey! I just wanted a little help if possible: I was testing a Japanese RVC I trained myself (200 steps, around 30 minutes) but I was trying to make it pronounce things in other languages like Portuguese. Sometimes it pronounces things very wrong and I don't know what I should do to make it sound right. I'm using Applio 3.2.6. Any suggestions please?

[low shard]

That's because it's not trained on other languages

The most you can do is turn Lower the index rate making the accent lower

[grave frigate]

I trained it just so I could hear this voice saying things clearly in other languages 🥲 So it's kinda impossible?

[grave frigate]

I'll try it to see if I can get any better results.. TuT

[low shard]

Was it mostly trained on japanese and a lil on Portuguese?

[grave frigate]

Just Japanese. It's a Japanese voice actress I like the voice and I wished to hear her speak things in my own language

[low shard]

Imagine you studied only the Roman empire, you wouldn't be able to perform well if I asked you about another historical time

Same thing goes with AI, if you trained it only on japanese, you can't expect it to go well with other languages

[grave frigate]

I see y.y Sometimes she can speak phonemes she wasn't trained for and other times she pronounces the same phonemes completely differently. I thought I could push it to speak clearly

but thank you very much for helping

[simple ore]

it is not about the language

it is about prononcuation and having specific sounds present in the dataset

[low shard]

Pronunciation can be different based on the accent and language

[simple ore]

yeah, but if the language has no L sound, then trying to infer L sound would result in weirdness

[brittle wing]

Do I have to manually change the audio frequency of each sample or RVC resamples my audio when I choose it from the drop down menu through "preprocessing"?

[knotty moth]

you mean 200 epochs?
200 steps = ~5 epochs lol

[simple ore]

it automatically convertes source audios to target sample rate

[brittle wing]

SO I DON'T HAVE TO CONVERT ALL 40 SAMPLES TO 32 HZ?

Thanks!!!

[brittle wing]

Is dat true?

[proven hill]

pretrains wont do miracles, everything depends on the dataset

[analog obsidian]

no

[brittle wing]

Nice

[brittle wing]

Is he/she right?

[analog obsidian]

half-true
your dataset needs to have high pitch data and you also have to use a pretrain that has high pitch data as well

[brittle wing]

Og pretrain isn't the best, as it has noise.

[analog obsidian]

noise is not really a bad thing

[brittle wing]

Understandable

[proven hill]

og pretrain is the best, others will produce noise

[analog obsidian]

og pretrain is bad for singers bc it was trained using monotone data

rvc injects silence to the dataset to teach the model to not generate noise in silence parts

• also learns from the natural silence in your dataset

[brittle wing]

There was a message of someone converting the of pretrain into a .py file and inferenced w it and it had noise.

[proven hill]

why would you do that

[analog obsidian]

yea thats normal but when you finetune rvc learns what silence is and mitigates that problem

its whatever

[brittle wing]

Someone else did it

[analog obsidian]

doesnt affects models

[brittle wing]

Mitigates?

[analog obsidian]

removes it

[brittle wing]

That's nice

[analog obsidian]

og pretrain naturally adds noise in the whole result
however when you train a model, rvc automatically injects 2 silence files to teach the model to not add such noise

so the result is a model that is able to inference silence without adding noise

[brittle wing]

Uh that's nice

[analog obsidian]

the pretrain having noise also helps rvc understand what noise is

so for example if u didn't clean your dataset's noise, rvc will know to separate noise vs voice

it'll learn that the noise is a feature of the dataset

pretrains that are extremely clean (like klm) requires a dataset extremely clean as well since the pretrain has no idea what noise is

[brittle wing]

I use Melroformer denoise 2, is it okay

Aggressive

Cause I noticed it even removes noise from speech in talking videos

[analog obsidian]

yea its good, klm requires a very clean dataset
og pretrain works fine with noisy audio

[brittle wing]

Nice is titan still useful

[analog obsidian]

idk i haven't personally compared it vs the rest

[brittle wing]

Oki

[knotty moth]

it was trained in the old fp16 and it might seem undertrained

[placid holly]

Pulling Applio from github and hunggingface (3.2.8 fix) is different 🥲

now i know

[placid holly]

Question, i tried testing train with KLM 5 RefineGAN, but somehow it said "The parameters of the pretrain model such as the sample rate or architecture do not match the selected model."

what should i do?

[simple ore]

what code did you download from github?

the models posted 3 days ago are compatible with the latest main code

[torn tundra]

Batch Size: 8
Dataset Length: 30
Pretrain: DMR V2
Sample Rate: 32k
where should I put these settings please tell me?

[simple ore]

nowhere, it is a model's description

[unique ledge]

Applio Colab Training is just not working at all, tried Preprocess Dataset and the console just says "Preprocess completed in 0.00 seconds on 00:00:00 seconds of audio.
Backup Complete: 1 new, 0 updated, 0 deleted."

im assuming preprocessing the audio shouldnt be that quick... so something has to be off, anyone know the issue? i've been following the Aplio Guide for the colab as well.

[simple ore]

show the settings

[unique ledge]

cant share photos in here, but my Model Name is set to "me" sampling rate "48000" dataset path "/content/drive/MyDrive/datasets/me/me.flac" Dataset Creator UNCHECKED

[unique ledge]

?

[simple ore]

it ask for a folder, not for a file name

@unique ledge

[unique ledge]

I tried both ways, I did "/content/drive/MyDrive/datasets" as well as "/content/drive/MyDrive/datasets/" with a slash at the end...

both just process it in 0.01 seconds

[prisma creek]

how i get the rvc?

[low shard]

What's ur PC GPU? And what do u want to do

[prisma creek]

i have nvidia

[low shard]

Nvidia is a brand who made a LOT of GPUS, which exactly

[prisma creek]

i wanna make ai cover songg

[prisma creek]

laptop gpu i dont know exactyl

3050

[low shard]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

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 daily, not granted, of GPU

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

[prisma creek]

which one dı you prefer

which one s best ?

[low shard]

You can choose whatever, if u want to do it locally without limits then applio is the best, if u want the easiest on cloud then weights.gg is the best

[prisma creek]

okey

thank so much for help

big love

[low shard]

Yw

[unique ledge]

any help?

[simple ore]

does this have a slash in the end?

no it does not

[unique ledge]

again, I'm using Applio colab... i dont even see that cell you just showed anywhere

[unique ledge]

I'm using the Applio UI version, not the Applio NoUI, i wasnt aware that was even a thing.

still, whats the issue with my colab?

[simple ore]

same thing

provide a path to the folder with dataset

[unique ledge]

I'm... well aware... i tried multiple different methods, putting it in a folder, putting it out of the folder in the mydrive, tried slash, tried no slash, just finishes the preprocess in 0.01 seconds... which obviously means its doing nothing with the .wav audio file.

should it be a .wav or should it be a zip?

I just attempted "/content/drive/MyDrive/me.wav" and would you look at the result.

"Preprocess completed in 0.01 seconds on 00:00:00 seconds of audio.
Backup Complete: 1 new, 0 updated, 0 deleted.
Files are up to date."

[simple ore]

once agan, 00 seconds means you got nothing in

[unique ledge]

yes and that's why im confused, clearly there is something in, im providing it the path to the .wav file, the audio is there, so why isnt it finding it?

[devout flume]

Have you tried copying the full path? Usually paths start with the hard drive followed by the folders till you hit the file

Unless your not locally doing it in which case I'd have to look at it myself to find anything

[simple ore]

path. to. the. folder. with. the. files

is google drive mounted?

[fervent rover]

Alright than, I’m trying the Applio No Ui, wish me good luck🤞🏻

[unique ledge]

Yes it is, I hit the cell at the very bottom, which has 4 cells in it, however, i should note that I stopped running the 4th cell at the very bottom of them, the one that does the "Download all custom pretrains" because from the guide, it seemed as though I didnt need it for training?? unless im mistaken, should I run that?

[devout flume]

o save time, unfold it & cancel the custom pretrain download, if you aren't going to use them. <--- the guide does indeed say this

so it should be fine

[unique ledge]

yep, canceled it and didnt use it

ah, got it working now, just did a different version of the path name

[devout flume]

i didnt seem to have to deal with a path name

i went to applio then clicked on the training tab and checked the dataset creator, uploaded my vocals and proceeded to do the rest of the stuff and am currently training

[unique rock]

does rvc disconnected still work?

[devout flume]

what version of the path name did you use?

[simple ore]

with UI version you do need to use a dataset creator, it basically creates a folder and puts your files in it

[fervent rover]

I Need Help, where did the pth file of it go?

I literally do everything of the video that it says, Raw Onions

[modern zodiac]

Is rvc still considered the best out there for making models?

[eager siren]

ngrok mainline colab

[fervent rover]

I am using the mainline Colab now, wish me good luck as well👍🏻👍🏻

[neon hemlock]

SKAP i need help

[clear ferry]

Sorry for the ping but what's the difference between the generator loss and the fm loss? I'm kind of new to ML/torch but it looks like the fm loss is based on the output from each conv1d layer within the discriminator, while the generator loss is based on the final output of the discriminator? Does each have a different purpose?

EDIT: If anyone randomly finds this for whatever reason, it's a GAN technique called Feature Matching. It helps avoid mode collapse

[simple ore]

discriminators looks at the real sample thru an answer sheet, then it looks at the generated sample thru the same answer sheet, records values from each and FM is the discrepancy between them

generator loss is a result of the discriminator looking at the generated samples and providing a probability of it being fake

yes, feature maps are indeed a result of each layer of discriminator network and the discriminator or generator score is the final value at the end

[clear ferry]

I see, thanks

[fierce leaf]

Hey everyone! I've been trying to get RVC running locally on my machine and I finally got it to work... sort of

I realized my macbook isn't up to the task

So I was wondering if anyone here know's what kind of a machine I need to run this stuff

[placid holly]

i just download the zip file and then run install.bat

it just download everything automaticly

so i guess it the latest one from there

[simple ore]

yeah, that will do

[placid holly]

i did that last night and "The parameters of the pretrain model such as the sample rate or architecture do not match the selected model."

the problem still there

[simple ore]

oh.. hm

something not right

well, no issues on my side

[placid holly]

hmmmm

i will just run intsall.bat again, if it working

[fervent rover]

Hey Guys, It Works The Mainline Colab

[knotty moth]

haven't tried the 40k version of the pretrain but the 32k one should work for the corresponding target sample rate

[placid holly]

i will also try it again this night, hope it work

[near thorn]

can someone send the link of aihub guide?

[hallow thistle]

-rvc

[karmic olive]

Suggestions for @near thorn

📚 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

[near thorn]

thnk u very much! 👍

[thin edge]

what should I see in TB g/total or mel

[dreamy wind]

how do I use the model ?

I cannot download them

[dreamy wind]

I tried running it, but the error no module named dotnev come out

@acoustic scarab

[acoustic scarab]

try this one
or this if you have amd/intel gpu

[dreamy wind]

Why does that error appear ?

[acoustic scarab]

its older and less capable than applio but i can assure you it will run

[acoustic scarab]

you need to send the entire error message for me to give any advice

[dreamy wind]

pip install python-dotenv

Can I just dm you

[acoustic scarab]

no

[dreamy wind]

It said I am missing these stuff

[acoustic scarab]

if it says try pip install python-dotenv then you can just do it and try again

[dreamy wind]

I tried that and now its telling me to install other stuff

[acoustic scarab]

gimme a minute

you're using windows?

[dreamy wind]

Yes

[acoustic scarab]

download this https://huggingface.co/IAHispano/Applio/resolve/main/Compiled/Windows/ApplioV3.2.8-bugfix.zip, unzip and run run-applio.bat

i think you got the non-compiled version, without the required packages

[dreamy wind]

Ok let me try

[dreamy wind]

Error not found

[dreamy wind]

Looks like my pc is bugging. So I restarted it now it works

[acoustic scarab]

👍

[dreamy wind]

What is Applio btw ?

[acoustic scarab]

fork of the thing i've sent you in the first message

[dreamy wind]

I mean like is it also rvc-realtime-voice ?

[acoustic scarab]

no, as far as i know it's not realtime

more like "make an ai cover"

[dreamy wind]

Can I not download rvc-realtime-voice ?

[thin edge]

tts ?

[acoustic scarab]

-okada

[dreamy wind]

because that is what I am trying to download from https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/tree/main

[acoustic scarab]

it's better to use this https://huggingface.co/wok000/vcclient000/blob/main/MMVCServerSIO_win_onnxgpu-cuda_v.1.5.3.18a.zip for realtime

[knotty moth]

the latest version is actually the main repo that is not officially released yet but has some features like the new model vocoder options: mrf-hifigan and refinegan over the default hifigan since the mainline version. you can clone it with the following way:

[thin edge]

Why does my tensor board only display changes every 200 steps?

[knotty moth]

it's recommended to use fork version of wokada voice changer, go to #🔍│help-w-okada and check out the pinned guide

[dire sparrow]

Does it work on rtx5080?

[dreamy wind]

Why not rvc real time ?

[acoustic scarab]

Most likely no, as far as I know 5080/5090 requires latest version of pytorch

[knotty moth]

oh man, another 50-series failure on their paper launch other than the stability issues
https://youtu.be/-KDA-h00VRc?t=645

[acoustic scarab]

https://tenor.com/view/mgs-metal-gear-solid-mgs2-metal-gear-solid-2-snake-gif-13225802252216398615

[knotty moth]

*refer to: #🔍│help-w-okada message

[warm oriole]

Hey ive downloaded a model with pth from weights.com and using it in appilio but it doesnt sound the same am i doing something wrong?

[thin edge]

How much is the google colab limit? They say 12 hours but I used collab only for 2 1/2 models (1/2 because of disconnect in the middle), I'm pretty sure it didn't take 12 hours.

[tame mica]

12 hours is the timeout iirc

[hallow thistle]

Who told you like that? Free tier in Google Colab limits to 4 hours daily usage if running with T4 GPU. With only CPU of course would give you that 12 hours when active.

If you leave the Colab session inactive for 90 minutes with no code running going on, it will automatically disconnect.

[low shard]

12 hours without gpu, the gpu is max 4 hours daily

but it’s random

[thin edge]

okay... that's bad ig

what happens if I use Local Mainline with GTX 1650?

[low shard]

Fine for inference, doable for training even if not much suggested

what are u looking for?

Train (make) RVC Models on cloud:

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

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio by Vidal (UI, no guide as of right now)
  • 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):
  • Applio (UI)
  • Mainline (UI, No guide as of right now)

3. Be sure to know about the tensorboard

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

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

You can use either:

• Weights.gg: Easiest Possible Ever Automatic
• Ilaria RVC Zero: Fastest free on cloud
• Applio (ui)

colab isn’t the only cloud

[thin edge]

place to make voice model ?

[low shard]

inference = use models
Training = make models

so it’s kinda doable but i wouldn’t suggest it that much

your choice

[thin edge]

BUT what if i use Mainline with GTX 1650 for training model How bad is it? Do I have to wait for hours or can I not do it at all, or will my pc explode?

[low shard]

It is doable, it’s possible, but it will take more time and might be limited on the batch size

[hallow thistle]

Yes, you'd have to wait for many hours if you train a voice model there with GeForce GTX 16 series GPU.

Your GPU won't explode if everything is set properly.

[thin edge]

Does Local Mainline not have “timeout”/“AFK” like colab? If not, I have no problem turning on the pc.

[hallow thistle]

Uh.

Locally RVC won't disconnect itself if running for a very long time.

[low shard]

It doesn’t have any timeout, it’s locally it runs on ur pc and it’s open source, colab has time out bc it’s using a remote good pc, which costs google money

btw there’s also kaggle

[knotty moth]

4 GB vram is not viable for minimum recommended batch size 4

perhaps batch size 2 but not really recommended since it could lead to more model instability

[hallow thistle]

If you close that web UI while it's training either by accidenr, RVC would still continue training in the background. Any attempt to launch web UI again after you did that, everything there will be reset and won't tell you if the model is finished or not. So it would be better to leave that web UI tab on.

[thin edge]

aight thankyou nick, seia, and touhou project person

[hallow thistle]

You're welcome.

[jolly veldt]

how can i make a voice with ai that i can use with a voice changer?

[warm oriole]

Anyone?

Hey ive downloaded a model with pth from weights.com and using it in appilio but it doesnt sound the same am i doing something wrong?

[low shard]

What's ur PC GPU and CPU?

[low shard]

Might depend on the settings

[solid bobcat]

Can I upload an mp3 recording to the program and get the output file processed by the neural network? Is there any other way to do this?

[unique rock]

Does anyone have any guides for Applio no UI?

how do you use this??

Empty

[tame mica]

make an ngrok account, find the authtoken then change the "gradio" to "ngrok" and put your authtoken in the "ngrok_token" column

[low shard]

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

[unique rock]

what is ng used for

Ngrok?

[low shard]

Ngrok is a tunnel to expose the local port from the google colab PC to the internet so you can use the Public url link

Gradio, which is a python package for making web UIs for AI, also has its own tunnel, the gradio tunnel

[unique rock]

I understand, thanks, another question, is it possible to train models of 50min of dataset? in either of the two colabs? ui or not ui, I did it in rvc disconnected for each training 100 epochs, will it be possible in these too?

[low shard]

Yes you can train 50 mins and how much epochs you want, but you shouldn't just train every single model to 100 epochs, you should use the tensorboard

Also rvc disconnected wont come back

[brittle wing]

https://discord.com/channels/1159260121998827560/1337292865407029288
Is it suitable for Applio No UI Collab?

[unique rock]

Lamentablemente y gracias, como cargo pre trains ? Como el klm5

[brittle wing]

Yes or no?

[simple ore]

if you use a main branch then yes

[brittle wing]

What was the way?

[unique rock]

Unfortunately and thank you, how do I charge pre trains? Like the klm5

[simple ore]

in the clone cell code remove the branch name

[brittle wing]

Not replace?

Only remove

[low shard]

you mean change?

you would need to use the main branch of applio

[simple ore]

remove that

[icy vessel]

how do i combine two voice models, i remember i saw something about it here but I don't remember where or how

[unique rock]

I should remove this then