✨│ai-help

[frosty wolf]

..

you have got to be kidding me

what is it now...

[graceful obsidian]

what the

[frosty wolf]

did it fail to nothing or something?

[graceful obsidian]

that is something I haven't encountered nor seen anywhere yet

the heck

ah, or it could be the python

actually

[frosty wolf]

3.12 -> 3.10??

[graceful obsidian]

you know how to add python to path?

3.10.6 ye

[frosty wolf]

im not familiar with python at all sorry

[graceful obsidian]

https://www.python.org/ftp/python/3.10.6/python-3.10.6-amd64.exe

then just download this anddddd

[frosty wolf]

?

do i check this

[graceful obsidian]

ye

if that won't work, there's one guide that will help

[frosty wolf]

should i uninstall my previous version first?

[graceful obsidian]

ideally ye

[frosty wolf]

ok im uninstalling 3.12.3

3.10.6 is now installed

should i try again now?

[graceful obsidian]

verify:

py
and
pip

in cmd

[frosty wolf]

oh ok

[graceful obsidian]

to see if all works fine

[frosty wolf]

cha ching for pip

and py too

[graceful obsidian]

neat, now, retry with reclist

[frosty wolf]

okay

[floral token]

@graceful obsidian Me again! I'm seeing half-precision floating-point: False and no supported Nvidia GPU found. I did a fresh install using only your fork

[graceful obsidian]

aha, ye

would be the driver then

[floral token]

odd. it was detected using Mainline RVC.

[graceful obsidian]

I think it had different detection mechanics

so far I feel like rvc in general had 3-4 revisions

[floral token]

I do have a driver update but I'm not severely outdated. I'll see what's up

[graceful obsidian]

nono, I meant the specific cuda

pytorch

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

[frosty wolf]

shit

[floral token]

thanks, I'll poke around @graceful obsidian

[frosty wolf]

oh wait do i just need to run that command it gave me to update pip?

orrrr nah

[graceful obsidian]

that one's for charm

as for our case.. hmm.. can you open up the c++ thingy

and see if there's " C++/CLI support for v142 build tools (14.24) "

[frosty wolf]

anway dyk what went wrong

[graceful obsidian]

or something similar

[frosty wolf]

uh i tried

[rare goblet]

Ayo? @frosty wolf level 6 !!!

[frosty wolf]

just brought me here, the wrong screen

[graceful obsidian]

go into available

[frosty wolf]

doesnt seem right

[graceful obsidian]

ikr, that's odd

alr so, go into the last one

build tools, go for modify

and if that doesn't then maybe " more "

[frosty wolf]

aha that worked

[graceful obsidian]

neat

[frosty wolf]

what am i looking for

[graceful obsidian]

now, do you see / can find " windows 8.1 / 10 SDK "

or something of that sort

[frosty wolf]

wait what about this

[graceful obsidian]

individual comp. section

[graceful obsidian]

also, can you show the language packs too?

[frosty wolf]

those are just literal languages

like english spanish etc

[graceful obsidian]

oh, like that alr

now, see if you can find SDK

" windows 8.1 / 10 SDK "

[frosty wolf]

whats the " / " for

[floral token]

at what directory should i be running this command?

[graceful obsidian]

search for sdk itself

a

[graceful obsidian]

doesn't matter really, anywhere from cmd

[graceful obsidian]

hmmm.. well, there goes available " solutions "

[frosty wolf]

so its a lost cause? my pc is so dumb i swear

what did i do wrong

[tawny lily]

Hi all. Can someone explain what the feature retrieval rate option during inference means? And should the pitch option always be 0?

[graceful obsidian]

try to restart the pc and try then

if that won't work then uhhh, yea, not sure what we could do

[frosty wolf]

last resort goes crazy

[floral token]

This is what I get; seems like all is up to date

[rare goblet]

Ayo? @floral token level 4 !!!

[frosty wolf]

alright be right back after restarting my pc

[graceful obsidian]

alr

[graceful obsidian]

hmmm... that's weird

so if torch's up to date

bruh

let's try the hardcode way then

[frosty wolf]

alright im back

so run the thing again?

pip install -r requirements-win-for-realtime_vc_gui.txt?

[graceful obsidian]

@floral token

replace your config.py with this one ^
rvc dir / configs / will be here

[graceful obsidian]

yes

[frosty wolf]

and as it turns out

we may have wasted our time

cant have shite in detroit

[floral token]

done, still not recognizing GPU unless you had another step @graceful obsidian

[graceful obsidian]

last thing I can think of is this:
https://aka.ms/vs/17/release/vc_redist.x64.exe

[graceful obsidian]

huh

[frosty wolf]

installed, trying again

[graceful obsidian]

try " nvidia-smi " @floral token

in cmd

[frosty wolf]

hmmm no such file?

heres bigger version, also it didnt work AGAIN

[graceful obsidian]

well shit, I'm sorry but I've totally no idea what is going on

[frosty wolf]

aw damn thats a shame

[graceful obsidian]

hmmm. here it's recognized so the heck's up with rvc

[frosty wolf]

i wonder what on earth is causing this

[graceful obsidian]

yea.. that's a good question

[frosty wolf]

perhaps we can try this?

[graceful obsidian]

heck, try one more time
go into v2 configs, in 40k and 48k replace fp16 false to true

do the same for "inuse" v2 configs as ^

[frosty wolf]

??

[graceful obsidian]

that's the same thing, it'll have reclist as well

the exact same one

[frosty wolf]

god dammit

[floral token]

I don't have a 40k in Configs > v2 fwiw

[frosty wolf]

just my luck that i get the one error nobody knows how to solve

[graceful obsidian]

lemme think of another way

[floral token]

also my fp16_run is already set to true now

[graceful obsidian]

alright, ye, this one should work

change the

false to true

[frosty wolf]

neither voice changers work 😭

[graceful obsidian]

thanks to this, no matter wut, you'll be getting true

without having it overwritten by rvc

[graceful obsidian]

what was the issue with w-okada again?

[floral token]

where is this that line?

[graceful obsidian]

line 130th

config.py

in configs dir
or just ctrl+f and search for "fp32"

[frosty wolf]

its VERY complicated

i have no idea how i would go about explaining it

but i can try

[graceful obsidian]

try with your own words first

we ain't moving anywhere until we figured out if it's a definitive pass

lol

[floral token]

@graceful obsidian still no change

[graceful obsidian]

ignore

that's just a print statement

[floral token]

still showing that in the UI

[graceful obsidian]

hmmmmm

[frosty wolf]

alright, so, when i boot it up, my headphones glitchout and any audio that goes through them gets really distorted. only fix is unplugging and replugging them. another thing, i cant hear myself when i click start, but i think that i just need to use passthru. and the voice changer doesnt change my voice, it doesnt convert. AND also upon first download, the start_http.bat file wont open unless i go into the properties of it and unblock it. and sometimes an error pops up upon startup that causes me to have to restart it.

[graceful obsidian]

you have any custom audio drivers?

or perhaps realtek rather than standard microsoft?

[frosty wolf]

im using realtek

[graceful obsidian]

that could be an issue
tried going with default?

[floral token]

the inuse configs were still showing as false. trying again

[frosty wolf]

no..? idk how

also wouldnt that play through my speakers or something?

idk

[graceful obsidian]

I'd first try with standard win driver for audio, without realtek

[frosty wolf]

im all for it, but how

[graceful obsidian]

so
win+r and type devmgmt

[frosty wolf]

okay

[graceful obsidian]

Under audio devices select realtek, then right click and uninstall

[frosty wolf]

this orrrrr

[graceful obsidian]

yes

actually, show what you have in there

[frosty wolf]

okayt

[graceful obsidian]

this section

audio controllers / devices or whatever it is

in eng win

[floral token]

the distortion sounds like a sample rate issue to me @frosty wolf

[graceful obsidian]

and in speaker

you have wut

it goes through win audio and reroutes ish
or detects headphones right away

[frosty wolf]

nah it distorts EVERYTHING not just the vc

like youtube and stuff

[graceful obsidian]

oh, sorta buffer underruns

what's your specs?

cpu, ram

some i5 or i7? ryzen 3/5?
ram above 8 gigs or 8 ? below?

[floral token]

configs in the inuse folder were not changed to true so did that, still no luck @graceful obsidian

[graceful obsidian]

oh, then that's not the issue

[graceful obsidian]

well fuck

something's up with the gpu or driver

maybe try to DDU the driver

and reinstall it

could be nvidia screwed something up

[frosty wolf]

i could use obs to record what happens

if that would help in any way

[graceful obsidian]

can you first show me 2 things

[frosty wolf]

yessir

what?

[graceful obsidian]

in control panel > audio > playback

aaaaand, your w-okada's ui

[frosty wolf]

sorry headphones youre about to get fucked by okada

[graceful obsidian]

unplug em

[floral token]

that is the driver located where?

[graceful obsidian]

and vol down all

[graceful obsidian]

DDU is an utility for driver removal but ye, have you done it before?

[frosty wolf]

only if audio is playing will it make a wretched sound

[graceful obsidian]

then don't play anything just yet save ur ears

[graceful obsidian]

https://www.youtube.com/watch?v=xn8z39tiEL0

[floral token]

nope I should be able to figure it out. which driver am I looking to uninstall/reinstall?

[graceful obsidian]

better follow it, ddu can screw u up if done incorrectly

you wanna get rid of all nvidia remainings you have

[frosty wolf]

looks like its reinstalling

[graceful obsidian]

and install fresh driver for ur rtx

[frosty wolf]

its taking a bit to open

[graceful obsidian]

mmmmm

[frosty wolf]

kinda stuck here when i opened start.http

[graceful obsidian]

try to press enter few times

while having console clicked ( have it active, click on it with mouse )

[frosty wolf]

nothing

ACK

my ears

[graceful obsidian]

maybe try to open it my way

a

[frosty wolf]

i tried playing audio to see if it still cucked my headphones

...it did

oh its open now

[graceful obsidian]

can you ss the ui?

mainly the bottom section

where there's " audio " and such

[graceful obsidian]

can you uncheck the monitor one?

[frosty wolf]

yeah

[graceful obsidian]

do that

[graceful obsidian]

next, show me what's under output

the options you have

[graceful obsidian]

set the default, first one

[frosty wolf]

done

[graceful obsidian]

now, take off headphones

[frosty wolf]

okay

[graceful obsidian]

🗿 now, see if it does the tricky sneaky things

with audio

[frosty wolf]

well, i unplugged them and replugged them so its not

lol should i restart?

[rare goblet]

Ayo? @frosty wolf level 7 !!!

[graceful obsidian]

hmmmm... restart w-okada then
but this time

" MMVCServerSIO.exe "

use that

[frosty wolf]

okay

[graceful obsidian]

just skippin' the bat part

[frosty wolf]

i just now had to go into properties and unblock it

[graceful obsidian]

heck

[frosty wolf]

i think my pc is glitched and when i click it its supposed to pop up with something like "this is sketchy would you like to run anyway?"

but it doesnt

so i gotta manually do it

[graceful obsidian]

a ye

[frosty wolf]

anyway its open what now

[graceful obsidian]

did it do the sound stuff?

[frosty wolf]

test audio

itis

i heard it from my headphones

which are OFF

[graceful obsidian]

hmmmmm... bro

[frosty wolf]

loud asf

[graceful obsidian]

to recap, you use 18a

cuda version

of w-okada

[frosty wolf]

yes

[graceful obsidian]

thennn, maybe try dml one

[frosty wolf]

i swear i tried that but i will try again

[graceful obsidian]

oh

you did?
this one?
https://huggingface.co/wok000/vcclient000/resolve/main/MMVCServerSIO_win_onnxdirectML-cuda_v.1.5.3.15.zip?download=true

[frosty wolf]

noope

[graceful obsidian]

that's the 15 one

unless you tested 18a

perhaps just rollback to the 15, ye

[frosty wolf]

i tested 18a

[graceful obsidian]

if 18 ones cause problems

try dml cuda 15, ye

[frosty wolf]

idk how this would help, but its worth a shot

[graceful obsidian]

might be some bugs in newer ver occuring on your hardware but ye, give it a shit

if still no luck then ig we're out of options

at least until someone figured it out

[frosty wolf]

we have been at this for so long holy cow

persistance

extracting

[graceful obsidian]

yeee, about that, think I'll be going in a bit, gotta be careful to not screw up my sleeping schedule again

but let's see with ver 15 first

[frosty wolf]

thats understandable

in the end if it doesnt work its fine i guess i'll just be kinda sad

[graceful obsidian]

yea.. I get that part

if you'll be really darn up for rvc, we could try with new win installation ( new partition so your main windows is safe ) tomorrow maybe

that is, if all failed and you still wanna have rvc / voice changer capable device

[frosty wolf]

alright

tomorrows friday ‼️

[graceful obsidian]

for me it's already 5 hours in

lol

[frosty wolf]

oh damn

its only 9:36 pm thursday for me

[graceful obsidian]

yuh, timezone
Poland here so, utc +1

[frosty wolf]

cst

usa

[graceful obsidian]

o

hmmm, how's the archive?

[frosty wolf]

heres the process

its open, and well

[graceful obsidian]

lemme guess

[frosty wolf]

heres that error i was talking about

[graceful obsidian]

win11 be like

gosh, glad I am a fan of 10

[frosty wolf]

good news, audio isnt cucked

[graceful obsidian]

haha!

there we go

[frosty wolf]

...i havent set my audio stuff yet

[graceful obsidian]

oh, 🗿

[frosty wolf]

lets hope it stays un-cucked

FUC

my ears

and i reloaded and got the error again

lost cause

[graceful obsidian]

hmmm, only logical explanations to this I might have is:
A) realtek colisions or something up with audio driver / win11 firewalls, permissions or whatever
B) win11 itself

[frosty wolf]

well this fuckin sucks

[graceful obsidian]

if I was you

[frosty wolf]

i cant use these peak vc's

[graceful obsidian]

personally

I'd get win10, have another partition for that

get neat clean python 3.10.6

[frosty wolf]

am i even able to do that on a win11 pc

[graceful obsidian]

driver for nvidia
and w-okada

[frosty wolf]

what would happen to my files

[violet heron]

What’s your GPU?

[frosty wolf]

i got a lot of stuff on here

[graceful obsidian]

you can divide any of your hdd's free space into another partition

[floral token]

just went through DDU, clean reinstalled drivers for nvidia. still not recognizing my GPU @graceful obsidian lel

[frosty wolf]

nvidia gtx 1650

[graceful obsidian]

which will be reserved for another windows, so win 10

but not sure how win11 handles dual boot

[violet heron]

good enough

[graceful obsidian]

would need to research it

[graceful obsidian]

hm

[frosty wolf]

maaaaaan i really wanted this to work

[graceful obsidian]

seems like we have 2 sussy as hell cases here

• unrecognized rtx 3k series gpu

• wheel / visual basic issues
• audio under/overflows / earr*p3s

[violet heron]

Have you tried the other realtime RVC?

[frosty wolf]

yes

lost cause

[graceful obsidian]

ye we did, issues with visual 14

[frosty wolf]

will not install correctly

[graceful obsidian]

no luck whatsoever

build tools, nothing

[violet heron]

sad I hope you get it working somehow

[floral token]

I'm running the web ui bat right? @graceful obsidian

[frosty wolf]

its fine its only a voice changer

[graceful obsidian]

yes

[frosty wolf]

its not like i NEED it...

but i do really want it

[graceful obsidian]

ngl, having AI voices / covers + games is imo top notch fun nowadays

[frosty wolf]

oh well, it is what it is

[graceful obsidian]

so

not surprised

well, I'll read up on win11 dual boot and potentially, if you'll want, we could tinker with win 10 partition

some time if you want

[frosty wolf]

idk how that would work

so i would like switch to win10?

if so how would that affect my current files? i got a lot of files on this rinky dink pc of mine

[graceful obsidian]

those stay unaffected

switching would occur at the boot

where you pick win11 or 10, if it's still working like that

• barely know shit bout w11 tbf
  as I hate that thing lol

But ye, that maybe sometime
Gonna go now. Sorry I couldn't resolve your issues bigshot n charm

my brain's out of ideas 😩 ✊

and energy

[frosty wolf]

its alright, im very grateful that you at least triedd

thanks

[graceful obsidian]

glad I could try
so ye, cya aroundddd

night ~

[frosty wolf]

gn bro

[frosty valley]

hey, i have and issue. why does when i try to voice change okada in discord calls it works perfect well but when i try to on mic on games like roblox, valorant and others. It just lags? is there any way i can fix this?

[odd valve]

maybe running the game takes up pc resources and doesnt leave much for okada

check ur cpu and gpu usage when running the games and okada at the same t ime

[frosty wolf]

im gonna try the google collab version...

its gonna be shit but its all i got

[thorn abyss]

aight, no one answered lmao

[rare goblet]

Ayo? @thorn abyss level 2 !!!

[thorn abyss]

i tried both

didn't work

nah it's okay

thanks 4 answering

dw

[frosty valley]

Ive checked it but when i open task manager the usage was only up to 50%-70 with okada and roblox running the same time

[hexed oriole]

how can i test voice ai here plz?

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

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

🇪🇸 Spain

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

🇫🇷 France

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

🇩🇪 Germany

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

🇯🇵 Japan

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

🇰🇷 Korea

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

🇵🇱 Poland

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

🇷🇺 Russia

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

🇮🇹 Italy

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

[strong surge]

/guides realtime

[sour gyro]

Hi! I have this error when i try to load the voice model to Colab. How can i fix it?

JSONDecodeError Traceback (most recent call last)
<ipython-input-6-1ef848b1bd32> in <cell line: 31>()
31 if os.path.exists(config_path):
32 # File exists, proceed with creation of creds and client
---> 33 creds = Credentials.from_service_account_file(config_path, scopes=scope)
34 client = gspread.authorize(creds)
35 else:

Colab i use: https://colab.research.google.com/drive/1Gj6UTf2gicndUW_tVheVhTXIIYpFTYc7?usp=sharing#scrollTo=OVQoLQJXS7WX&uniqifier=1
Voice model: https://drive.google.com/file/d/1PzoeBllu0gk3GVyOj_Qc-jqr0adRbqrw/view?usp=drive_link

[low shard]

outdated colab

use only the updated ones in our docs https://docs.aihub.wtf

[brittle wing]

Whenever I run go-realtime-gui I get this error:

C:\Users\Toby\Desktop\RVC\RVC1006Nvidia>runtime\python.exe gui_v1.py
Traceback (most recent call last):
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\gui_v1.py", line 59, in <module>
import torch
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch_init_.py", line 122, in <module>
raise err
OSError: [WinError 1114] A dynamic link library (DLL) initialization routine failed. Error loading "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.

I made two path environmental variables for cuda bin and libnvvp as well

[brittle wing]

yo nick the
🇪🇸 Spain
https://docs.aihub.wtf/v/spain/ doesnt work

[leaden yacht]

So I js used vocalremover.org to get the vocals from a song, but the problem is that the background singing is also included which really messes up the voice. Is there anything like a tool or smth I can use to extract the vocals without background singing?

[nimble pawn]

can you help me set it up later? im also an amd user so a better rvc experience would be nice

[forest sluice]

https://mvsep.com

[leaden yacht]

thx

[forest sluice]

Well, we don't have Spainish supported

[leaden yacht]

I found the official guide first, so I'm gonna try my luck with UVR before I use that tho

but still thanks

[pastel oak]

@frosty wolf ping me again when youre online, we can do the prepackaged version

[tawdry shell]

is this the chat where I can ask for audio/editing help?

[pastel oak]

Suppose so

[tawdry shell]

yeah cuz I've been messing with the Alastor AI model, and I've been trying to remove his radio filter if that makes sense

Dunno how and so far I've tried messing with it a lot already

closest I got just made the radio effect softer hahaha

[pastel oak]

You can try to use UVR local or go to mvsep.com and try one of the models that fits the description best

BS-Roformer could work depending on what radio signal it is.. else Bandit Plus, maybe MelReformer

If those fail, you could with Izotope RX 10 with Spectal Denoise if theres a spot in the audio that only contains the signal to learn the pattern of the frequencies and remove them completely.. else the repair tool to highlight the frequency and remove it

I dont know what kind of signal it is so I am throwing a few suggestions in there for the mvsep models

Izotope rx10 costs big money tho

[tawdry shell]

BS-Roformer???

[pastel oak]

Thats a model on mvsep

The ones i listed there

[tawdry shell]

ah alright

[brittle wing]

Whenever I run go-realtime-gui I get this error:

C:\Users\Toby\Desktop\RVC\RVC1006Nvidia>runtime\python.exe gui_v1.py
Traceback (most recent call last):
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\gui_v1.py", line 59, in <module>
import torch
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch_init_.py", line 122, in <module>
raise err
OSError: [WinError 1114] A dynamic link library (DLL) initialization routine failed. Error loading "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.

[rare goblet]

Ayo? @brittle wing level 1 !!!

[tawdry shell]

I had the same error too before, downloading the python 64 bit library fixed it tho

I was using the 32 bit when the error popped up

[brittle wing]

How do I determine my python bit

[tawdry shell]

mvsep has a queue I think? uh oh

[tawdry shell]

it's from the download page

how did you download the library and where???

[pastel oak]

if you make an account you pretty much skip most of them and its for free, like max 5 queues usually

[tawdry shell]

ohhhh okie

[rare goblet]

Ayo? @tawdry shell level 3 !!!

[brittle wing]

AMD website

python-3.12.3-amd64.exe

[pastel oak]

Never seen this error, before ill let you know in a few minutes

what gpu do you have?

[brittle wing]

I already have 64bit

RTX 1070

[pastel oak]

Well thats the problem

1. the rvc1006 folder you downloaded comes with the right python version anyway

2. python 3.12 doesnt work most of the time, only 3.10

[brittle wing]

So just uninstall 3.12 and install 3.10?

[tawdry shell]

yep

[brittle wing]

Will I need to reinstall cuda and all that?

[pastel oak]

Yea

[pastel oak]

No

[brittle wing]

Any specific version of 3.10 or just 3.10?

[pastel oak]

any of the 3.10

Id maybe try uninstalling then run go realtime, if it doesnt work then manually install 3.10

On my first time i didnt install python myself so could work right away

[brittle wing]

C:\Users\Toby\Desktop\RVC\RVC1006Nvidia>runtime\python.exe gui_v1.py
Traceback (most recent call last):
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\gui_v1.py", line 59, in <module>
import torch
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch_init_.py", line 122, in <module>
raise err
OSError: [WinError 1114] A dynamic link library (DLL) initialization routine failed. Error loading "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.

C:\Users\Toby\Desktop\RVC\RVC1006Nvidia>pause
Press any key to continue . . .

Without python

C:\Users\Toby\Desktop\RVC\RVC1006Nvidia>runtime\python.exe gui_v1.py
Traceback (most recent call last):
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\gui_v1.py", line 59, in <module>
import torch
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch_init_.py", line 122, in <module>
raise err
OSError: [WinError 1114] A dynamic link library (DLL) initialization routine failed. Error loading "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.

With python 3.10

[brittle wing]

Python 3.10 didn't fix it

Could it be a driver issue?

[pastel oak]

Open the folder it says failed to load up and check if the file even exists

if you have antivirus i think it might have deleted the file. or extracting didnt work

runtime / lib / site-packages / torch / lib

check for cublas64_11.dll

[sand iris]

microsoft c++ redistribuable or something would be needed

if you have cuda installed on host it should be able to load though

OSError: [WinError 1114] A dynamic link library (DLL) initialization routine failed. Error loading "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.

[brittle wing]

This?

[rare goblet]

Ayo? @brittle wing level 2 !!!

[sand iris]

Unsure which version

Honestly, install CUDA from nvidia, it would get everything needed

"Cuda toolkit driver"

[brittle wing]

Did that

[brittle wing]

Did that already

[sand iris]

Ah fuck

[brittle wing]

Repairing this did nothing

[tawdry shell]

mvsep had a stroke trying to remove the radio filter LOL

guess I'm gonna have to mess around with audacity

[proud elbow]

radio filter, u mean vocoder? idt there's one that converts it into dry vocals

[tawdry shell]

yep, it's why rn I'm just trying to manipulate the audio in audacity

[brittle wing]

Tried reinstalling the drivers, did not fix the issue

Whenever I run go-realtime-gui I get this error:

C:\Users\Toby\Desktop\RVC\RVC1006Nvidia>runtime\python.exe gui_v1.py
Traceback (most recent call last):
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\gui_v1.py", line 59, in <module>
import torch
File "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch_init_.py", line 122, in <module>
raise err
OSError: [WinError 1114] A dynamic link library (DLL) initialization routine failed. Error loading "C:\Users\Toby\Desktop\RVC\RVC1006Nvidia\runtime\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.

[graceful obsidian]

yuh, no issues with that

[leaden yacht]

when training a model, you don't need to have wifi right?

[rare goblet]

Ayo? @leaden yacht level 1 !!!

[leaden yacht]

For both tensorboard and the training itself

[queen kraken]

For local training, no, you don't need wifi nor any other internet connection

[rare goblet]

Ayo? @queen kraken level 5 !!!

[leaden yacht]

alr thx

[leaden yacht]

how do I figure out the sample rate?

I have about 9 mins of talking I wanna use but idk what the sample rate should be

and should I enable pitch guidance?

[pastel oak]

download spek you can see it on there easily

[pastel oak]

ye

[leaden yacht]

this?

[pastel oak]

ye

[leaden yacht]

btw, is an RTX 4060 (Laptop) good for training?

[leaden yacht]

and I got this

[pastel oak]

do 32k

[pastel oak]

should be fine

[leaden yacht]

howd u get that

js curious so I can do it myself later on

[pastel oak]

the frequencies reach up to around 17k. you multiply that by two = 34k

theres only 32 40 48, and upscaling to 40 would be bad cause it has to "guess" some lacking frquencies, so you go down

[leaden yacht]

with the frequencies reaching up to 17k, do u mean the left side?

[pastel oak]

Yes

its like around 17khz

[leaden yacht]

ah good I asked, because I almost put 48 as I thought the right side needed to be multiplied (its around 24-25 there I think)

So I need to watch the left side, and multiply that and then js the closest?

[rare goblet]

Ayo? @leaden yacht level 2 !!!

[pastel oak]

nah dB is decibel, aka. how loud it is

[leaden yacht]

yeah js realised that this wasnt the brightest plan 💀

[pastel oak]

not the "closest", because even if you have 19k = 38k, you should use 32k

[leaden yacht]

So I always need to go down?

[pastel oak]

you might be able to get away with 38k dataset on 40k training but itd be "better" to go 32k because 38-40k are missing

[leaden yacht]

and if higher than either 32, 40 or 48 choose the highest I can

[pastel oak]

Yea

so if you have 42k you choose 40k

[leaden yacht]

and if I have 47k still 40k

[pastel oak]

yea

[leaden yacht]

alright

thx

imma see if I can train this thing now

[pastel oak]

theres a general thumb rule on cpu cores but i cant really remember ngl

[leaden yacht]

what should I put for the singer id?

[pastel oak]

i use 7 on my 15 cores

[leaden yacht]

js still 0?

[pastel oak]

ye

[leaden yacht]

I'm guessing I need to/am recommended to use rmvpe?

[pastel oak]

Yes

[midnight fern]

How tf do i make a cover with ios

[leaden yacht]

I think the only option you have is google collab

[midnight fern]

Link?

[leaden yacht]

https://colab.research.google.com/github/iahispano/applio/blob/master/assets/Applio.ipynb
This is the applio collab

Don't have any other links but if u want I can look for them in the official guide. Also, read the guide at https://docs.aihub.wtf/rvc/cloud/applio-colab/

it explains what you need to do to make a cover or train a model

how to prepare the audio etc

are these settings good to go?

[pastel oak]

do you have tensorboard?

[leaden yacht]

mhm

js opened it

[pastel oak]

how much vram does the 4060 have

8?

[leaden yacht]

yup

8gb vram

[pastel oak]

do save frequency 10 or 15
do max total epochs because youll be monitoring tensorboard anyway
batch size 8 should be fine

[queen kraken]

if spek indicat 17k which is suitable for 32k, what if I train it for 40k? what would be the result?

[leaden yacht]

Why 10-15 actually? isn't it smarter to pick smth like 5 or does it really not differ that much?

[midnight fern]

I mean to make a ai cover with a existing model

[pastel oak]

Im not a pro at this so my answer might not be 100% accurate

The AI has to guess every frequency in the 17-20k range, which will create inconsistencies or additional noise etc.

[leaden yacht]

and once it starts to overtrain, and I get the steps and lets say it was at epoch 55, and it saves every 10 epochs, do I js pick the 50?

[pastel oak]

Well you can do 5 for maximum accuracy but its probably not needed

[leaden yacht]

thats there too

[rare goblet]

Ayo? @leaden yacht level 3 !!!

[midnight fern]

Oh

[leaden yacht]

u can both train and make covers with that collab

just go through the guide where it says inference

[pastel oak]

the easiest is using Ilaria RVC its straight forward and simple

[pastel oak]

Yea, BEFORE the OT point

[midnight fern]

can u give me a link

for ios

[queen kraken]

thanks Shad. ill try and vice versa

[pastel oak]

uh idk if it matters if youre on ios or not

[leaden yacht]

so the closest save before?

[pastel oak]

ya

[leaden yacht]

same as the sample rate js go down then

alright

[pastel oak]

https://docs.aihub.wtf/rvc/cloud/ilaria-rvc/

[leaden yacht]

Alr, ill do 10 and if anything happens I can still redo with 10 later

total training epochs to max right?

this good?

[pastel oak]

yes

wait

batch size 8

[last lava]

hi everyone

[leaden yacht]

what does that do?

[leaden yacht]

sup

[last lava]

is there any update on rvc voice model traning? i have issue on some models when i used them for text to speech.. they dont sound like how i hear it in this channel model samples.

i want the rvc to follow the words, sentences, and punctuations accurately like elevenlabs. can you guys help me how i can do that please?

[last lava]

i tried all my best from picking a sample from elevenlabs to make a model 😭

[leaden yacht]

shad, what do I set smoothing to?

I set it to 0.987 for now as I saw in the guide but not sure

[pastel oak]

Batch Size, how many clips it grabs, and will learn the similarities between the clips

if youre unsure about how clean your dataset is, a higher number is good, if its a really good one, you can go lower

I just always use 8, it works well as a middleground if you wanna call it like that. Thats probably not the proper terminology or fact but thats how I explain it to myself

[leaden yacht]

and is this good?

[pastel oak]

its only been 200 steps so cant say much about that xd

[last lava]

you guys are genius i can see that

[leaden yacht]

^

[queen kraken]

have you tried gpt-sovits? as fat as I know, it's best for TTS

far*

[last lava]

so its better than rvc? is it free ? can i run it on googlecolab?

[leaden yacht]

does this mean that step 460 is epoch 22?

[queen kraken]

I think so. I haven't tried it for myself as it doesn't support my native language.

https://colab.research.google.com/github/RVC-Boss/GPT-SoVITS/blob/main/colab_webui.ipynb

[pastel oak]

In your weights folder you can see the exsct epoch to step of every model variant

[last lava]

aight bro thanks for your supprot

[leaden yacht]

ah thx

So js to be clear, once it starts overtraining I need to figure out the steps at the lowest point and then go down to the save with the closest amount of steps I can find in my weights folder?

[leaden yacht]

is it normal for it to go up again a little and then down?

[pastel oak]

yea

[pastel oak]

some inconsistencies are normal yes

the cleaner your dataset the less theyll be going up and down

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

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

🇪🇸 Spain

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

🇫🇷 France

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

🇩🇪 Germany

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

🇯🇵 Japan

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

🇰🇷 Korea

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

🇵🇱 Poland

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

🇷🇺 Russia

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

🇮🇹 Italy

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

[proper shale]

*depends on ur batch too, if you have it too low, training will be very very unstable

[leaden yacht]

that explains a lot

no idea if this is good but still lol

[hardy oriole]

Where can I get the ai cover website link?

[proper shale]

yeah seems to be going fine ig

[leaden yacht]

yeah uh

[proper shale]

though it didnt have a big dip

[leaden yacht]

about that

[proper shale]

are ya using ov2 ?

[leaden yacht]

Once I uncheck and recheck this

[proper shale]

https://docs.aihub.wtf/ has all the ones u wanna use

[leaden yacht]

It kind of changed

A lot

[proper shale]

oh nvm that's good

[pastel oak]

you changed the scope of the y axis, dw

[leaden yacht]

ohh

[pastel oak]

its going good

[leaden yacht]

I js did that lol

[proper shale]

yeah i was being dum

loll

[pastel oak]

Hes using default pretrain not ov2 btw

[leaden yacht]

whats ov2?

[proper shale]

hovering around low 30s, seems to be a decent dataset

[proper shale]

a pretrain

[leaden yacht]

ah

[proper shale]

it's good for short datasets but can cause issues on most datasets

because you need denoising

[leaden yacht]

is this good?

in the end its going up

[proper shale]

but if u have a good 5 min denoised dataset it's a good thing

[proper shale]

yeah dw about it

[pastel oak]

You can relax and let it train for a while, dont be scared hahaha

[leaden yacht]

alright

[proper shale]

g/total is just average

[leaden yacht]

this is my first time actually training lol

but I'm pretty paranoid about this thing as my previous laptop did not like to train this stuff

I had to change the code to use my rtx 1650 to even train the index, and then tensorboard didnt even work

Basically an error every 5 mins

[rare goblet]

Ayo? @leaden yacht level 4 !!!

[pastel oak]

Ya wouldnt recommend training with a GTX

[leaden yacht]

I said rtx

[pastel oak]

rtx masterclass

theres no rtx 1650

or is there

[proper shale]

gotta thank nvidia for the terrible naming scheme

[leaden yacht]

oh nvm

[proper shale]

but no, it's a GTX

[leaden yacht]

ur right yeah

[pastel oak]

all models starting 2XXX are rtx

[proper shale]

i was also confused abt it 5 years ago dw

shit was mad confusing

[leaden yacht]

Don't even get why they use RTX and GTX

[leaden yacht]

It was lol

[proper shale]

gtx and rtx mean something but

i forgor

[leaden yacht]

up until now I thought I had an rtx 💀

[leaden yacht]

at least the 4060 I have in my laptop now isnt so damn problematic

[pastel oak]

yea the 4060 is the best budget laptop you can get

[frosty wolf]

well i got school

[pastel oak]

budget maybe wrong word but you get it

[leaden yacht]

I mean I got it for a lot cheaper than I thought I would

Ryzen 7 7735HS with Radeon graphics, rtx 4060 8gb vram, 16gb ddr5

for €1000

which is a pretty good price

btw as of now, the smoothing is set to 0.987 as I saw in the guide

is that good or do I need to change it?

[fluid topaz]

nice one, i think i got the same thing for $1200, except its 32gb ram

[leaden yacht]

nice

im planning to upgrade it to 32gb

for about 100 eur

What brand did u get? I got an ASUS TUF Gaming laptop

[fluid topaz]

no need to change, its good setting

[leaden yacht]

Ah great

[fluid topaz]

lenovo legion 5 15

[leaden yacht]

nicee

does this seem good?

[spark pelican]

Guys, why base of models on this server is so small? I check it earlier and it have way more models than how

[fluid topaz]

what graph is that?

[spark pelican]

And why all models uploaded no longer than a month?

[fluid topaz]

main thing you should watch are loss/d/total and loss/g/total

[graceful obsidian]

btw, about this

If the audio is 17k, so, that's 34khz

you could try 40khz

Just resample it to 40khz first

[leaden yacht]

wait both?

[leaden yacht]

this one is g/total

[graceful obsidian]

What you'll " gain " is just that extra 2khz handling

Similar as if you train 40k model on 37-38khz set vs 32k model on that set.
You just retain some higher harmonic and sibilants handling

[graceful obsidian]

as for this, check the pinned msgs

I've explained there the tensorboard's metrics
overtraining spotting n such

#✨│ai-help message
section nr 2

[leaden yacht]

alright thx

how long would it take js to train the model? Is there an estimate or js completely random? I used a dataset of about 9m and 17s

[graceful obsidian]

as for batch_size
( this is purely figurative )
if the set is 5-8 mins, try anything from 6 to 12
if it's above 8 or 10 mins, try from 12 to 16 ( I recommend trying out 16 or 14

[graceful obsidian]

case-dependent

I'd say, 9 mins could be anywhere from 30 ish to 120 epochs maybe, sometimes 180

but again, just figurative

If you have time and patience, you can do batch_size range finding first

also make sure to sync the steps ( log interval in case you don't have

[leaden yacht]

its at epoch 234

lol

[graceful obsidian]

otherwise tensorboard's values are inaccurate

[leaden yacht]

wdym

[graceful obsidian]

it's the last section

in the pinned msg

what rvc / fork you use? @leaden yacht

let's start with that, cause some forks incorporate " auto syncing " iirc, which isn't accurate

[leaden yacht]

mangio

[graceful obsidian]

dang

[leaden yacht]

v23.7.0

?

[graceful obsidian]

yea, it's wonky in few ways

if I remember it right it has hardcoded auto syncing

tl;dr, if you want accurate training, you'd have to change the fork

but the decision's up to you

[leaden yacht]

any recommendations on the fork?

[graceful obsidian]

try mine

[leaden yacht]

sure

[graceful obsidian]

based on newest mainline + mangio stuff

and few gimmicks for easier training

https://github.com/codename0og/hifi_testing_new

pretty much

only thing you have to do aside getting it and moving the hubert and few things where they should go

[leaden yacht]

Do i js download the repos as zip?

[graceful obsidian]

yes, then just move the hubert.pt and few other things I'll guide you with

[rare goblet]

Ayo? @graceful obsidian level 22 !!!

[graceful obsidian]

aside that:

that's for pip -r

cause mine's using sox resampling for best quality vs standard resampy / librosa as in other forks

[leaden yacht]

Some ppl make it for only the local disk, did u do that too? Or will it also work if I put it on my external ssd?

[graceful obsidian]

should work without any issues

[leaden yacht]

great

[graceful obsidian]

main thing here is to use python 3.10.6 and my reclist ( if something did conflict - list with the repo

and ye, those sox thingies I sent

[leaden yacht]

downloaded and extracted

[graceful obsidian]

you wanna take these
from mangio
and move em to fork's dir as shown ^

[leaden yacht]

imma install the python packages rq

alr I installed the pip pkgs

[leaden yacht]

only those 2?

[graceful obsidian]

oh right, pretrained v1 and v2 too

they land in assets folder
rvc dir/assets/pretrained_v1 pretrained_v2 each containing whatever it contains

[leaden yacht]

wait where are the files in the mangio folder?

[graceful obsidian]

in mangio they're either in the root of the fork or in assets ( depending on how they handled the mainline changes

lemme check rq

[leaden yacht]

no assets folder

[graceful obsidian]

ye, they're just like so

in the root

[leaden yacht]

Is that all?

[graceful obsidian]

neat, now pretrained v1 and v2

[leaden yacht]

no pretrained v1 folder btw

[graceful obsidian]

just rename it

_v1 ye

[leaden yacht]

so I need to move those 2 to assets

[rare goblet]

Ayo? @leaden yacht level 5 !!!

[graceful obsidian]

mangio is just heavily outdated and bugged quite a lot here n there

yes and renaming pretrained to pretrained_v1

[leaden yacht]

done

copied them over

[graceful obsidian]

Alrighty, try to do the test run

[leaden yacht]

should I stop the mangio training btw?

[graceful obsidian]

yes

[leaden yacht]

guessing I need to do pip install dotenv or smth

[graceful obsidian]

use my list

[leaden yacht]

I did that

those 2 things

[graceful obsidian]

in the address bar type in cmd

[leaden yacht]

but that did not download dotenv clearly

[leaden yacht]

Dw

I know

I use python regularly

I already installed dotenv using pip install dotenv

[graceful obsidian]

good, just use my reclist from there as I am not sure whether mangio was using outdated libs or not

[leaden yacht]

and nvm

my python is being problematic again

[graceful obsidian]

using 3.10.6 ?

[leaden yacht]

dotenv?

[graceful obsidian]

py

[leaden yacht]

ohhh that

[graceful obsidian]

the reclist is finetuned for my fork and 3.10.6 specifically, as in, 100% working

[leaden yacht]

3.10.11

[graceful obsidian]

can't say much about 3.11 or 3.12

oh

well, I'd get fresh 3.10.6 and use my reclist to avoid any conflicts

[leaden yacht]

alr

should I remove the old python?

[graceful obsidian]

not sure why few people get wheel issues

yea, purge it completely

unless you use python for something else too

then just dump ur packages, to be safe

[leaden yacht]

I do yeah

[graceful obsidian]

then do

[leaden yacht]

I dont need to

[graceful obsidian]

pip list

a

well then

[leaden yacht]

I can probably js do py -3.10.6

think that works

[graceful obsidian]

if you can assure there won't be packages conflict

[leaden yacht]

and jeez nvm

brb

[graceful obsidian]

just do:

pip list

then

pip list > packages.txt

[leaden yacht]

nah I dont need that

[graceful obsidian]

just precaution

[leaden yacht]

any packages I need I can js do with the pip install -r cmd

[graceful obsidian]

had few people whining one time they lost their packages

so

[leaden yacht]

Most things I use js have a requirements.txt

[graceful obsidian]

gotya

[leaden yacht]

alr I installed py 3.10.6

imma install your packages now

[graceful obsidian]

alr

[leaden yacht]

done

Istg its always something

was there a specific version of dotenv or js the latest?

[graceful obsidian]

nope, just the list and 3.10.6

[leaden yacht]

well I did that and got this

ah I found the problem

I needed to run pip install python-dotenv instead of js dotenv

[graceful obsidian]

weird

funky how that wouldn't occur previously 🤔 well good, gonna fix that

[leaden yacht]

It did not install anything at all, as it said req already statisfied, but it did change SOMETHING

as now I have another error :D

[graceful obsidian]

what is it?

[leaden yacht]

it didnt have the module av

maybe add that to ur list of reqs

sox==1.4.1
soxr==0.3.7
python-dotenv
av

and yup

its running now

[graceful obsidian]

it is there

funky funky

anyway ye, works?

[leaden yacht]

uhh

what

why isnt there 32k on the train tab

;-;

[graceful obsidian]

why you want 32?

I thought you wanted to train 40

like discussed above above

[leaden yacht]

Its better to use 32k?

[graceful obsidian]

if your audio's 34~ ( 17khz )

I'd go for 40

[leaden yacht]

refer to this convo

[leaden yacht]

doesnt that make the model worse?

[graceful obsidian]

if you use 32k model on 34khz dataset

you lose ~2khz

[leaden yacht]

but how do I resample it tho?

[graceful obsidian]

but I suppose, I can readd 32k

gimme a sec

[leaden yacht]

and uh

I found a bug

[pastel oak]

Theres always subjective opinions on that matter, everyone says something else

[leaden yacht]

if u open the train tab, go to rvc v1, and then back to 2 js look at the sample thingies (32k, 48k, 40k)

js so u know

[graceful obsidian]

replace the script

oh, yes that's correct

[graceful obsidian]

just a matter of whether you care about sibilants or not

sometimes 1k or 2k might not be worth it but 3-5

[leaden yacht]

should I js do this?

[graceful obsidian]

indeed can

[graceful obsidian]

nono, train 32 first

if you won't like it, you can try 40, then we'll talk about resampling

[leaden yacht]

ah alr

[graceful obsidian]

just gimme a sec gotta confirm one thing in the ui

[leaden yacht]

its restarting the gui now

[graceful obsidian]

32k was the thing I tested the least

[leaden yacht]

replaced the infer web

[graceful obsidian]

ye, should be working

do the first processing step on 32k

[leaden yacht]

Traceback (most recent call last):
File "T:\hifi_testing_new-main\infer\modules\train\preprocess.py", line 131, in pipeline
audio = load_audio(path, self.sr_trgt)
File "T:\hifi_testing_new-main\infer\lib\my_utils.py", line 17, in load_audio
raise RuntimeError(f"Failed to load audio: {e}")
RuntimeError: Failed to load audio: [WinError 2] The system cannot find the file specified

[graceful obsidian]

what path you use?

[leaden yacht]

T:\greg

called it greg as the character is gregory so yeah

Starting preprocessing
T:\greg/1_1_1_greg_(No Reverb)(Instrumental)(No Noise).flac -> Traceback (most recent call last):
File "T:\hifi_testing_new-main\infer\lib\my_utils.py", line 10, in load_audio
ffmpeg.input(file, threads=0)
File "C:\Users\Marti\AppData\Local\Programs\Python\Python310\lib\site-packages\ffmpeg_run.py", line 313, in run
process = run_async(
File "C:\Users\Marti\AppData\Local\Programs\Python\Python310\lib\site-packages\ffmpeg_run.py", line 284, in run_async
return subprocess.Popen(
File "C:\Users\Marti\AppData\Local\Programs\Python\Python310\lib\subprocess.py", line 969, in init
self._execute_child(args, executable, preexec_fn, close_fds,
File "C:\Users\Marti\AppData\Local\Programs\Python\Python310\lib\subprocess.py", line 1438, in _execute_child
hp, ht, pid, tid = _winapi.CreateProcess(executable, args,
FileNotFoundError: [WinError 2] The system cannot find the file specified

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "T:\hifi_testing_new-main\infer\modules\train\preprocess.py", line 131, in pipeline
audio = load_audio(path, self.sr_trgt)
File "T:\hifi_testing_new-main\infer\lib\my_utils.py", line 17, in load_audio
raise RuntimeError(f"Failed to load audio: {e}")
RuntimeError: Failed to load audio: [WinError 2] The system cannot find the file specified

Preprocessing finished

thats the full error

[graceful obsidian]

can you try 40k now?

let's see if that's the 32k issue

or path itself

[leaden yacht]

same error

I used this path for mangio and it worked

[graceful obsidian]

works for me too

[leaden yacht]

do I have to put it in the dataset folder of the fork itself maybe

[graceful obsidian]

32k as well

[leaden yacht]

imma try

[graceful obsidian]

try, maybe I hardcoded the search path for saturn purposes and forgot

[leaden yacht]

same error

[graceful obsidian]

hmmm, what is within your greg folder?

[leaden yacht]

only 1 file

the .flac

about 9 mins

[graceful obsidian]

convert it to .wave

make it float 32 bit

ohhh, you do have ffmpeg, right?

[leaden yacht]

should I use audacity to make it 48k too?

[rare goblet]

Ayo? @leaden yacht level 6 !!!

[leaden yacht]

yeah

I think so

[graceful obsidian]

alr, so that's chedk
ye you can use audacity

keep the sample rate same as original

just change the ext to .wave and 32 bit float

[leaden yacht]

this?

what does the 32-bit float do?

[brittle wing]

hey whats the easiest way to make rcv ai cover with m1 mac

[graceful obsidian]

it's the bit depth

[rare goblet]

Ayo? @brittle wing level 1 !!!

[brittle wing]

make file higher fidelity

[graceful obsidian]

ist your og file's samplerate 48?

[leaden yacht]

Would it be bad to put it on 16?

[graceful obsidian]

rvc uses 32 float anyway

[leaden yacht]

js curious

[leaden yacht]

ah alr

[leaden yacht]

nah

[graceful obsidian]

we wanna avoid reudant conversions

[graceful obsidian]

then keep it 1:1 with og

but like, the file's samplerate

not the frequency response range

[brittle wing]

what ai model do i use for m1 mac

[leaden yacht]

these were the settings for my old one I think

[graceful obsidian]

wait, can you send me the file

gonna check if it's stereo or dupe mono

also match the samplerate if needed

[leaden yacht]

stereo

I left it as it is

its defi stereo

[graceful obsidian]

ye, send it in if you can ( dm
gonna pick the better channel

[brittle wing]

what ai model would you recommend i use for m1 mac

[leaden yacht]

in here?

[graceful obsidian]

can be dm

[graceful obsidian]

not sure, ain't apple specialist but

-rvc

[azure marsh]

Documentation

📚 All-In-One English documentation

https://docs.aihub.wtf/

❔ English FAQ

https://docs.aihub.wtf/faq

Credits: AI HUB FRANCE

✨ More guides

Full AI Voice Model Training Guide (Local)

Link: YouTube
Credits: Christopher Villanueva

Model training with Mainline RVC

Link: Rentry
credits: Raven (ravencutie21)

AICoverGen Colab Guide

Link: Google Docs
Credits: Eddy (Spanish Helper)

Create a model with RVC disconnected (colab)

Link: Google Docs
Credits: Angetyde

How To Make an AI Cover With Ilaria RVC

Link: Rentry
Credits: 👽 Julia (ailen2091)

[graceful obsidian]

but that one's probs outdated uhhh

well, check applio

[leaden yacht]

nope

cant in dms

[brittle wing]

okay thanks

[graceful obsidian]

https://github.com/IAHispano/Applio

[graceful obsidian]

why?

[leaden yacht]

probably cuz the server is boosted and allows bigger file uploads

I can send it here tho

[graceful obsidian]

oh yea, then you can here or gdrive then

[leaden yacht]

Took the voice lines from fnaf sb lol

[pastel oak]

pls delete the dataset asap after code downloaded it

[leaden yacht]

I will

did u get it code?

[graceful obsidian]

alr, gimme a sec

gonna prepare it

[leaden yacht]

on the top it also says 44100 Hz

so that confirms that

[graceful obsidian]

the file is 44.1

the actual audio data is not

[brittle wing]

wait can i literally just use the website and not run anything locally?

[leaden yacht]

hmmmmmmmmmm

[leaden yacht]

Ya think?

[graceful obsidian]

but that's because compression is involved

17.5khz + compression

was it originally 18khz ( 36khz ) from uvr?

or ripped directly from the game?

[leaden yacht]

I think from game

I got it from gdrive tho

[graceful obsidian]

now, look

given the response reaches even 17.5

it can kinda go 2 ways

32khz model ( 16khz ) or 40khz

but then, I suppose you won't lose much

those " lost " freqs you'll get by going 32 instead of 40 are mostly damaged anyway

anyways, uploading the file

[leaden yacht]

is that the fixed one?

[graceful obsidian]

ye

works on my end as well

[leaden yacht]

so I do need to use 32k?

[graceful obsidian]

that's from processed / 0_gt

[graceful obsidian]

you can

I'd first try 32 and only 40 if you didn't like 32 ( while keeping it all realistic ofc, can't have magic with low quality audio

[leaden yacht]

alr

but what was wrong with the thing?

why was it giving an error

is flac unsupported?

[graceful obsidian]

first check if it works

before we can make any conclusions

could as well be naming issues or something with ffmpeg

[leaden yacht]

Starting preprocessing
T:\greg/greg.wav -> Traceback (most recent call last):
File "T:\hifi_testing_new-main\infer\lib\my_utils.py", line 10, in load_audio
ffmpeg.input(file, threads=0)
File "C:\Users\Marti\AppData\Local\Programs\Python\Python310\lib\site-packages\ffmpeg_run.py", line 313, in run
process = run_async(
File "C:\Users\Marti\AppData\Local\Programs\Python\Python310\lib\site-packages\ffmpeg_run.py", line 284, in run_async
return subprocess.Popen(
File "C:\Users\Marti\AppData\Local\Programs\Python\Python310\lib\subprocess.py", line 969, in init
self._execute_child(args, executable, preexec_fn, close_fds,
File "C:\Users\Marti\AppData\Local\Programs\Python\Python310\lib\subprocess.py", line 1438, in _execute_child
hp, ht, pid, tid = _winapi.CreateProcess(executable, args,
FileNotFoundError: [WinError 2] The system cannot find the file specified

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "T:\hifi_testing_new-main\infer\modules\train\preprocess.py", line 131, in pipeline
audio = load_audio(path, self.sr_trgt)
File "T:\hifi_testing_new-main\infer\lib\my_utils.py", line 17, in load_audio
raise RuntimeError(f"Failed to load audio: {e}")
RuntimeError: Failed to load audio: [WinError 2] The system cannot find the file specified

Preprocessing finished