✨│ai-help

[stoic cloud]

I don't know what's going on buddy, I am using the same version as you are. I guess you just have to let it train

[astral perch]

Hmm weird

[round marsh]

for some reason tensorboard keeps resetting

is there any explanation for that?

[astral perch]

Btw @stoic cloud what is the best target sample rate to use

[round marsh]

48k but i did here it can cause buzzing

so i'd say 40k is safer

[astral perch]

What is the difference

[round marsh]

40k doesnt cause buzzing

[astral perch]

What do you mean by buzzing?

[round marsh]

probably like a buzz sounds in the bg of the model

[astral perch]

I see

I always just use 40K

I wish I could just stop the training Lol

Maybe it's a bug Idk

[stoic cloud]

40khz

[astral perch]

I saw people using a graph to see wich epoch has the best trained voice model. Doesn't RVC local have something like that or do you just have to compare

[round marsh]

@potent root how do i know what part of the graph is which epoch

[wispy burrow]

Tensorboard can be installed and run locally.
What version of RVC do you have?

[astral perch]

RVC-beta0717

[wispy burrow]

Ah, I still have that one...
Doesn't seem like an "installer" or a "launcher" came with it.

Installing Tensorboard locally for RVC
https://docs.google.com/document/d/1ihvqWnSyabvZ9jX6BsOxEY1FpplwjY3Jse3she_VO6I/edit

[potent root]

the graph's points are measured in steps

when you look in the pths that you have in the weights folder

they are also notated with steps

[round marsh]

yea i realized

i just finished training and stopped at 330 epochs

graph d spikes up outta nowhere and graph g kept going down

[potent root]

you're supposed to look into the g/total

not those graphs

g/total measures loss

[round marsh]

ohhhhhhh

crap

i was reading over the tut too

[round marsh]

im guessing when loss is going up, its overtrained?

[vivid rock]

Hi everyone, how do I get the vocals from a song if there are several voices? Because when I get the vocals from, for example, fifty fifty - cupid, the voices that sing along are distorted

[astral perch]

How does it work

[potent root]

exactly

[round marsh]

I'll keep that in mind. Thanks!

[astral perch]

Oh no I hate python lol

Nvm I'll just do it without the graph

[analog oxide]

hi, i think training ended what should i do now?

[wispy burrow]

Tensorboards reads certain files in the folder it's directed to.
The TbL bat file directs it to the logs folder if the bat file is placed in the RVC folder.

[wispy burrow]

Check the cmd. if it says something like "last checkpoint saved" or something along those lines it(s finished. Don't forget to train the index too if you haven't already.

[wispy burrow]

You don't have to do anything with Python yourself though...

[astral perch]

Btw I always have the setting Save small final model to weights on yes, and I always use those but are those optimized asking because it saying small

[astral perch]

I'll try tomorrow

[analog oxide]

i got theese two files index and pth where should i head now

[wispy burrow]

What did you use?
Was it done locally or wasit RVC Disconnected or something?

[analog oxide]

rvc disconnected

[wispy burrow]

If you want to use the model you created, it's AICoverGen-NoWebUI I think if you don't want to infer locally..

[analog oxide]

this?

[wispy burrow]

https://colab.research.google.com/github/ardha27/AICoverGen-NoUI-Colab/blob/main/CoverGen_No_UI.ipynb

[analog oxide]

i looks a bit more complicated

@wispy burrow i have to pack theese two things like file.zip - file - added.index, mod.pth

?

[wispy burrow]

I haven't used any of the Colabs so I don't really know.

[analog oxide]

oh

i need to figure it out

this colab feels not working

i got disconnected before starting doing anything

lets try again

[astral perch]

I always have the setting Save small final model to weights on yes, and I always use those but are those optimized asking because it saying small

[rare goblet]

@astral perch has reached level 4.

[brittle wing]

Since the model inference does not work on Applio, is it ok if I use RVC GUI? Does that support v2 models? Sorry if this is a dumb question.

[analog oxide]

model training stuff and aivoice thhing can run on that stuff ? CPU: 11th Gen Intel i5-11400F (12) @ 4.400GHz
GPU: NVIDIA GeForce GTX 1650
Memory: 4795MiB / 15863MiB

[glad zealot]

Seems to be

Training ye

[reef pier]

over a thousand seconds and i still have not gotten anything

[rare goblet]

@reef pier has reached level 2.

[reef pier]

what do i do yo

[reef pier]

oh please

[reef pier]

i'm using the rvc v2 huggingface space

[glad zealot]

Try reloading it?

[glad zealot]

Might be slow since more users are using it???

[reef pier]

but i'm using a duplicated space

[glad zealot]

Ahh ye not sure how hugging face works so can't really help with it

[reef pier]

well that sucks

well at least my audio file's here

that took literally 2000+ seconds

well the next conversion can't be that bad

[reef pier]

1300 seconds

not bad

[glad zealot]

That's like 30 mins?

Ahh no

20+

[ornate idol]

this is on the rvc v2 huggingface

it just isnt working for shit

[steep quartz]

getting this weird bug all of a sudden after like 10 seconds of MMVCServerSIO voice changer being open it slowly goes up to 30 seconds of delay and then stops

https://imgur.com/HrR66wm

i am on amd so might be an amd problem

[rare goblet]

@steep quartz has reached level 1.

[glad zealot]

#🔍│help-w-okada

[plucky locust]

Community Models

goku (Goku (English Dub))
luffy (Luffy (English Dub))
mickey (Mickey Mouse)
peter (Peter Griffin)
plankton (Plankton)
spongebob (Spongebob)
squidward (Squidward)
waluigi (Waluigi)
... and 200+ more

Check out all of the public community models at https://www.word.band/voice_models

You can use any model using the model id (bolded) in the /cover command. For example:
/cover model:goku youtube:https://www.youtube.com/watch?v=s9FCCiEs84k

To see your private models and register new ones, please start a DM with me!

scp containment breach control is not a registered model! Use /list to list all models, or use /register to register a new model.

This isn't an RVC2 model! Please make sure to use an RVC2 model from AI Hub. (Make sure the model is tagged with RVC v2)

Successfully registered model scp (scp containment breach control)!

Use the model now like this:
/cover model:scp youtube:https://www.youtube.com/watch?v=s9FCCiEs84k

[patent stirrup]

Can I use this colab?

[glad zealot]

yes

[sand iris]

no

WebUI

[glad zealot]

works atm so use it while it still works

[shy monolith]

applio giving this error when trying to run with cli (web interface is bugged)

[rare goblet]

@shy monolith has reached level 2.

[sand iris]

I'm not familiar with the applio CLI, but I suggest feeding it all the arguments it's asking for

[shy monolith]

looking at the code, the code for the cli literally does not have enough arguments for the function its calling

[sand iris]

lol

Applio moment

[shy monolith]

im facing either figuring out how to fix the function

or fix the web interface

neither of which is rather appealing to me

[rare goblet]

@brittle wing has reached level 3.

[shy monolith]

ok fuck this im just reinstalling applio

[sand iris]

https://cdn.discordapp.com/emojis/790214199544512573.webp?size=48&name=kek&quality=lossless

[finite galleon]

Wait give link

[patent stirrup]

https://colab.research.google.com/github/hinabl/AICoverGen-Colab/blob/main/Hina_Mod_AICoverGen_colab.ipynb?authuser=3#scrollTo=NEglTq6Ya9d0

[finite galleon]

Oh lmao it bypasses the cloning repository thing so it doesnt get detected

Also bypass the downloading models and webui launch

[golden elm]

alr after training is over what do i do with these files to make my model sharable?

[analog oxide]

can i use this format in making datasets?

[glad zealot]

yes

[stone dawn]

how do i fix thisss

i mean some links are working except this

[proper shale]

mega link

try downloading the zip and uploading it somewhere else

[stone dawn]

huggingfacE?

idk how to upload in huggingplace

[golden elm]

?

[proper shale]

or maybe g drive

[proper shale]

go to weights folder, download the latest pth

[stone dawn]

that doesnt support on the gradio tho

[proper shale]

and download index from logs

[proper shale]

then upload to huggingface

-huggingface

[azure marsh]

huggingface.co upload Guides

• Posting models to AI Hub by FDG

https://rentry.org/fdg_guide_new

[proper shale]

^^^

[stone dawn]

thankss

[golden elm]

oh so the .pth in the weights is the one that i share

i mean

use to share

alr

[proper shale]

yeah!

[golden elm]

thanks

[proper shale]

G/D are training files

[golden elm]

i can delete those afterwards?

they take up quite some space

[proper shale]

if you dont plan to retrain

yeah

[glad zealot]

iirc pixeldrain works on that

[analog oxide]

500 epochs for 15 mins of recordings enough?

[glad zealot]

yup

[winged comet]

Guys, how can help with Applio API?

[topaz scaffold]

has anyone got doja cat model ??

[remote mesa]

so uhh.. problem
the .pth files for all my installed model thingies won't show up

happened after updating RVC. Kept around the logs/weights folders

it manages to find all the index files, but it doesn't find any of the pth files

im using the RVC WebUI

is RVC/weights no longer the folder you're supposed to add all the pth files?

OH WEIRD

[rare goblet]

@remote mesa has reached level 1.

[remote mesa]

It's moved to assets/weights apparently

[slender laurel]

how to see how much epochs are in a model?

[echo creek]

This is the graph of the model I'm currently training. Should I stop it now?

[fluid slate]

Hello. Has Google Colab done changes that makes it pretty much difficult to do RVC covers using Colab?

[potent root]

they've made changes, but we adapted

-colab

[azure marsh]

Google Colabs

RVC

• AI-Cover-Gen-No-UI (inference only)
• RVC v2 Disconnected (training only)

🌐 Translated colabs

• PT-BR: -colabs_br

Audio Separation/Isolation

• MDX Net Colab
• UVR arch colab 5.0.2/c1.04
• KaraFan colab

[potent root]

ai cover gen is for covers

no more webguis

google hates em apparently

[fluid slate]

Ok. Thank you

[rare goblet]

@fluid slate has reached level 1.

[analog oxide]

ayo guys i have question, when i want to resume because it looks like it goes from 1 again

[stoic cloud]

You have to find the G and D of your model

Which should be in logs/name/here-are-they

Make sure to do a new name for the model

[analog oxide]

too late

my files didnt even save

i searched in drive

after i go disconnected i lost all

@stoic cloud i have one question, after getting .pth and .index files what should i do

[stoic cloud]

The model is done. Now you can synthesize using those two files

-colab

[azure marsh]

Google Colabs

RVC

• AI-Cover-Gen-No-UI (inference only)
• RVC v2 Disconnected (training only)

🌐 Translated colabs

• PT-BR: -colabs_br

Audio Separation/Isolation

• MDX Net Colab
• UVR arch colab 5.0.2/c1.04
• KaraFan colab

[stoic cloud]

AI Cover Gen No UI

[analog oxide]

then i should put them

in one file

like model/.pth , .index ?

zipped?

nvm i found i think

[stoic cloud]

You have to put them in a zip and then put the url in there

[shrewd bolt]

C:\RVC0813Nvidia>runtime\python.exe gui_v1.py --pycmd runtime\python.exe
2023-10-14 19:16:38 | WARNING | xformers | A matching Triton is not available, some optimizations will not be enabled.
Error caught was: No module named 'triton'
2023-10-14 19:16:38 | WARNING | xformers | Triton is not available, some optimizations will not be enabled.
This is just a warning: No module named 'triton'
Found GPU NVIDIA GeForce GTX 1650 , force to fp32
input device:5:Headset Microphone (Realtek(R) (MME)
output device:12:CABLE Input (VB-Audio Virtual C (MME)
using_cuda:True
gin_channels: 256 self.spk_embed_dim: 109
<All keys matched successfully>
Exception in thread Thread-1:
Traceback (most recent call last):
File "threading.py", line 980, in _bootstrap_inner
File "threading.py", line 917, in run
File "C:\RVC0813Nvidia\gui_v1.py", line 476, in soundinput
with sd.Stream(
File "C:\RVC0813Nvidia\runtime\lib\site-packages\sounddevice.py", line 1800, in init
_StreamBase.init(self, kind='duplex', wrap_callback='array',
File "C:\RVC0813Nvidia\runtime\lib\site-packages\sounddevice.py", line 898, in init
_check(_lib.Pa_OpenStream(self._ptr, iparameters, oparameters,
File "C:\RVC0813Nvidia\runtime\lib\site-packages\sounddevice.py", line 2745, in _check
raise PortAudioError(errormsg, err, hosterror_info)
sounddevice.PortAudioError: Error opening Stream: Unanticipated host error [PaErrorCode -9999]: 'Неизвестная внешняя ошибка.' [MME error 1]

Help

[wheat apex]

Мира, не туда написала

[atomic mulch]

Which software do we download if we want our gpu to power the ai creation?

[brittle wing]

my error is: "is not recognized as an internal or external command operable program or batch file"

[surreal bridge]

anybody know why this is happening? I alr tried reinstalling with pip https://pastebin.com/hcqfXd1X
(applio)

[half cove]

ima test if cover gen no ui gets disconnected rq

[violet heron]

RVC

[atomic mulch]

Thank you and what link specifically?

[violet heron]

-rvc

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English Documentation

https://us.aihubfrance.fr/

Credits: AI HUB France

🇺🇸 English FAQ

https://us.aihubfrance.fr/faq

Credits: AI HUB France

Guides

🍏 Applio-RVC-Fork: Guide

Link: Google Docs

Credits: IA Hispano

How to install RVC on windows

Link: YouTube

Credits: iroaK

RVC v2 Tutorial for Paperspace

Link: Google Docs

Credits: LollenApe

🌐 Translations

• 🇫🇷 French Documentation
• 🇮🇹 Italian Guides or -it_g
• 🇷🇺 Russian FAQ for RVC/Voice changer
• 🇧🇷 RVC Guides in PT-BR (-rvc_br)

[violet heron]

Doesn’t really matter

I recommend applio

Since it’s easy for beginners

[atomic mulch]

Are they offline?

So if it gets taken down

[violet heron]

Applio is offline

It is on your pc

They cannot take it down

Off your computer

[atomic mulch]

Thanks a ton for your help

[heavy wolf]

can you help with ai voice model problem

[heavy wolf]

pls

[rare goblet]

@heavy wolf has reached level 2.

[brittle wing]

my error is: "is not recognized as an internal or external command operable program or batch file"

[heavy wolf]

my error is
runtime\python.exe trainset_preprocess_pipeline_print.py "C:\data set up\dataset\dataset_raw\maskoy" 40000 16 C:\Users\e n iyi pc\Downloads\RVC-beta-v2-0528\RVC-beta-v2-0528/logs/maskoy False
Traceback (most recent call last):
File "C:\Users\en iyi pc\Downloads\RVC-beta-v2-0528\RVC-beta-v2-0528\trainset_preprocess_pipeline_print.py", line 20, in <module>
f = open("%s/preprocess.log" % exp_dir, "a+")
FileNotFoundError: [Errno 2] No such file or directory: 'C:\Users\en/preprocess.log'

[surreal bridge]

mine doesnt even work and nobody responds whenever i bring it up ;-;

[wheat copper]

where is the step count?

[stoic cloud]

Step count is not there, but that's fine since epochs are all that matter

Steps = your amount of wavs / batch size

[lean monolith]

I've been making AI Spongebob songs (originals not covers) I'm not very good at this AI stuff but using google colabs I havent had any issues just using uploading my raw vocals n turning them too characters but ever since the OG disc got nuked no matter what settings I use, nor what method the vocals always sound chopped with alternating volumes, I'm stumped does anyone have any ideas?? (sorry if this is the wrong chat)

[autumn hill]

i love ai cover gen (lie)

[frail cave]

anyone tell me when make covers ?

[rare goblet]

@frail cave has reached level 2.

[tawdry ruin]

you're trying to use the local AICoverGen, right?

[autumn hill]

yep

[tawdry ruin]

wanted to give it a try as well but I hate dependencies

[autumn hill]

it never works for me lol

[tawdry ruin]

I'll probably try it one day, but the first requirement is Python 3.9 I think

and I have 3.11 on my pc

[autumn hill]

i have 3.9 so idk what its issue with me is 💀

[wispy burrow]

You're using 3.10.

[frail cave]

https://www.youtube.com/watch?v=gCO0JLbaC5g&t=230s Does anyone have these models or can send me them?

[wispy burrow]

Not really the place to ask this.
But check weights.gg.
It has numerous Sonic Models.

[late comet]

what version of rvc do i get

how do you even use it 😭

[violet heron]

Today

[violet heron]

I would use applio if your a beginner

[dusty rune]

no help just this

[half cove]

tim really helping jus to beat mj

[wispy burrow]

I would suggest mangio though...

[violet heron]

Applio has mangio

[wispy burrow]

I mean RVC-Mangio.

[violet heron]

Why’s that?

[wispy burrow]

Many people had issues with the Applio installer(s).
Some still do with the latest installer.
That's the one reason I'd suggest Mangio for newcomers.
It's just Download zip, extract zip & and run Go-web.

[balmy prawn]

Anyone with good/decent knowledge of running RVC 2 locally ? Need some help real quick

[violet heron]

What’s the issue?

[late comet]

hell is that

[rare goblet]

@late comet has reached level 1.

[balmy prawn]

Keep in mind that I'm r_tarded in this context, when start Go_web I get this "C:\Windows\system32>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7897
The system cannot find the path specified.

C:\Windows\system32>pause
Press any key to continue . . .

"

[wispy burrow]

Applio-RVC.
An altered version of (regular) RVC

[late comet]

im installing it

kaspersky is deleting shit

[sand iris]

Are you running as admin or something?

If yes, don't

[balmy prawn]

Yep

I've tried regular, I'll try again

[violet heron]

You need to install python

[balmy prawn]

Wait I don't have Python ?

[violet heron]

You might not

Check

[sand iris]

It's using runtime so it shouldn't matter

[balmy prawn]

Btw all this is in a different SSD from OS one

[sand iris]

Wait is that folder really named ...AI VOICE

[balmy prawn]

No listen

It's no like that

[sand iris]

Spaces in the file name break it

[balmy prawn]

Fuck Im retarded

[sand iris]

New Folder as well

Nah it's alright lol

It shouldn't break to begin with. Just that it's somewhat wonky

[rare goblet]

@balmy prawn has reached level 3.

[violet heron]

Don’t know

[sand iris]

[late comet]

can yall help with this

[balmy prawn]

Maybe folder should be in the root of the ssd ?

[sand iris]

Anti virus being an issue?

[late comet]

yeah

says its a trojan

[sand iris]

Where'd you download it from ?

[late comet]

the github

lemme send

[sand iris]

I'm not even sure how the anti virus thinks that's a threat

[late comet]

https://github.com/IAHispano/Applio-Installer

[sand iris]

Yeah that's the right one

[late comet]

i just wanna make sand underman say the funny big shot like

line

https://cdn.discordapp.com/emojis/1012933784628363384.webp?size=48&name=Spamton&quality=lossless

https://cdn.discordapp.com/emojis/851305705062400010.gif?size=48&name=letrollisSans&quality=lossless

[frail cave]

-rvc

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English Documentation

https://us.aihubfrance.fr/

Credits: AI HUB France

🇺🇸 English FAQ

https://us.aihubfrance.fr/faq

Credits: AI HUB France

Guides

🍏 Applio-RVC-Fork: Guide

Link: Google Docs

Credits: IA Hispano

How to install RVC on windows

Link: YouTube

Credits: iroaK

RVC v2 Tutorial for Paperspace

Link: Google Docs

Credits: LollenApe

🌐 Translations

• 🇫🇷 French Documentation
• 🇮🇹 Italian Guides or -it_g
• 🇷🇺 Russian FAQ for RVC/Voice changer
• 🇧🇷 RVC Guides in PT-BR (-rvc_br)

[frail cave]

-rvc

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English Documentation

https://us.aihubfrance.fr/

Credits: AI HUB France

🇺🇸 English FAQ

https://us.aihubfrance.fr/faq

Credits: AI HUB France

Guides

🍏 Applio-RVC-Fork: Guide

Link: Google Docs

Credits: IA Hispano

How to install RVC on windows

Link: YouTube

Credits: iroaK

RVC v2 Tutorial for Paperspace

Link: Google Docs

Credits: LollenApe

🌐 Translations

• 🇫🇷 French Documentation
• 🇮🇹 Italian Guides or -it_g
• 🇷🇺 Russian FAQ for RVC/Voice changer
• 🇧🇷 RVC Guides in PT-BR (-rvc_br)

[frail cave]

-colab

[azure marsh]

Google Colabs

RVC

• AI-Cover-Gen-No-UI (inference only)
• RVC v2 Disconnected (training only)

🌐 Translated colabs

• PT-BR: -colabs_br

Audio Separation/Isolation

• MDX Net Colab
• UVR arch colab 5.0.2/c1.04
• KaraFan colab

[acoustic thorn]

@nocturne mural yo

does applio fork have proper support for m1/m2 mac?

even on mangio-crepe?

or is that considered unsupported

[rare goblet]

@outer epoch has reached level 1.

[nocturne mural]

I don't think so, since I don't have a mac and neither do the team members.

[acoustic thorn]

dang

i can try nvidia style install and hope MPS / metal passthrough works as intended

or, worst case, i just have my guy install official RVC since that should have a working install script

[heady sequoia]

it may work

the thing is that we don't really know how well it goes on mac

[prisma kettle]

can anyone help

[rare goblet]

@prisma kettle has reached level 1.

[prisma kettle]

command

[prisma kettle]

anyone know how to fix this?

[uneven mantle]

can someone help me
why the voice is always delayed to me
and the application is lagging
but my gpu is rtx 3050

[rare goblet]

@uneven mantle has reached level 1.

[naive fossil]

-colab

[azure marsh]

Google Colabs

RVC

• AI-Cover-Gen-No-UI (inference only)
• RVC v2 Disconnected (training only)

🌐 Translated colabs

• PT-BR: -colabs_br

Audio Separation/Isolation

• MDX Net Colab
• UVR arch colab 5.0.2/c1.04
• KaraFan colab

[cedar cedar]

Hi guys

I need help

when I want to make a voice, an error pops up

traceack error

traceback*

[violet heron]

Kalomaze

Should I mode my switch, I already have a steam deck

[cedar cedar]

is anyone able to help me?

[rare goblet]

@cedar cedar has reached level 1.

[balmy prawn]

God bless.

[astral perch]

What does steps mean in the name of the voice models?

[half cove]

Hmm

What can I make a Google doc on

[brittle wing]

how do i add dataset on the applio colab / gradio?

[rare goblet]

@brittle wing has reached level 1.

[round marsh]

@potent root yo does this look good?

[inland hill]

Anybody else have their voice sounding fine on their own ( recording or hearing it back ) but when you use it on a discord call your friends say it sounds awful/glitchy?

[potent root]

looks fine

[rare goblet]

@potent root has reached level 5.

[potent root]

remember to maximize smoothness to 0.999

[round marsh]

i had at 0.987

also do i stop training when its at the very top or as soon as it starts going up

[brittle wing]

when i try, and press "update list" nothing comes out

[round marsh]

@potent root rlly sorry for pinging btw

[dusk oracle]

hello, how many epochs should i do for 6 minutes of audio?

[round marsh]

depends on tensorboard

[dusk oracle]

sorry but i know literally nothing about this and im trying to clone my voice and i used a tutorial that was linked on this server and it said to put the epochs depending on the length of my audio files, and i didn't know what to do, sorry

[potent root]

stop training when it goes up for a while

[dusk oracle]

i tried with 500 which was the already set number but the only thing that came out in my drive was a logs zip

also im using rvc disconnected

if that helps

[round marsh]

alri. thanks man you've been incredibly helpful

[glad zealot]

Try getting the pth file from around the 2k mark since that seems to be the lowest, but you are free to continue training it

[round marsh]

its supposed to be like that

[brittle wing]

and i use wav, so can anyone help??

[round marsh]

flyleaf tought me this g/total thing on tensorboard

so its supposed to be low like that

[glad zealot]

Ohh aight then continue on

Mine keeps going lower after 2k but okay

[analog oxide]

How does exactly resuming work

It saves G and D in logs

Not on gdrive

[round marsh]

I think if you stop the training, it should be able to resume it

i haven't tried it myself tho

also are u using colab?

[analog oxide]

Yeah

I want to save G and D + config to drive

[glad zealot]

I forgot exactly but you're using colab you need to save the G and D first then move them back to where they are from when you need to resume iirc

[analog oxide]

So I will be able to get it

[glad zealot]

I forgot where the guide for that is sadly

[analog oxide]

Ugh

I have 300 epochs

I'm in half

That will be huuge loose of time

[round marsh]

how long is your dataset

[analog oxide]

7min

[round marsh]

you could prolly go for abt 450 - 500 epochs

[high night]

350

[dusk oracle]

i only find rvcLogs.zip in "My Drive > rvcDisconnected > (my model name)" what do i do?

[rare goblet]

@dusk oracle has reached level 1.

[dusk oracle]

ill just try redoing everything again

[prisma kettle]

anyone can solve this?

[proper shale]

delete pretrain and model_dir folders

and restart start http

[raw mirage]

guys

[ocean torrent]

i need some help

Exception Traceback (most recent call last)
<ipython-input-1-159ffb84dd67> in <cell line: 48>()
47
48 if not os.path.exists(dataset_path):
---> 49 raise Exception(f'I can't find {dataset} in {os.path.dirname(dataset_path)}.')
50
51 os.makedirs(final_directory, exist_ok=True)

Exception: I can't find audios1.zip in /content/drive/MyDrive/rvcDisconnected.

it says that?

[nimble rock]

Quick question, hearing myself with the realtime voice changer in the mangio software works just fine. But when I use it through discord, everyone says it sounds very choppy and robotic. Does anyone know what the issue could be? I turn off the noise gate and not have noise supression on in the discord settings

[oblique mason]

whats the best settings

[violet heron]

Which RVC

[oblique mason]

idk

[violet heron]

That really doesn’t help

[oblique mason]

do you know if theres a way i can hear myself?

[violet heron]

what program are you using

[oblique mason]

real time voice change

[rare goblet]

@oblique mason has reached level 2.

[violet heron]

Okada or RVC Real-time

[oblique mason]

real time

[violet heron]

Send ss

Before I help you, does the voice changer even work for you

[oblique mason]

yes

[violet heron]

Monitor should be your headphones I think

See if that works

[oblique mason]

when i do that my voice changer echos other peoples voice and changes it too

[violet heron]

That’s because the microphone is picking up on the headphones

Or monitor

If your not using headphones

[oblique mason]

Im using headphones do you know how to fix that?

[violet heron]

Turn down in maybe?

[oblique mason]

I tried it still copies people

[violet heron]

Don’t know then

[oblique mason]

alr thanks for trying

[ionic yew]

has anyone having issues with voicemy.ai cus mines got jammed

i have a error model stuck and i can't delete it

and i was gonna train another rvc model which that also having errors

[analog oxide]

• why pitch is making audio robotic and a bit destroyed

coverGen

[vivid pasture]

can you train on linux?

[open ruin]

THIS IS THE VOICE PLS ANYONE TRAIN A MODEL WITH THIS VOICE AND SHARE ME

this one

HOW CAN I ADD HUGGING FACE LINK

[rare goblet]

@open ruin has reached level 2.

[pliant crescent]

Is this local

Cuz I'm pretty sure you have to download the model and put it a certain folder

[analog oxide]

huh, why pitch is killing everything\

[open ruin]

its local

[pliant crescent]

Yeah I think you should look at the local guide
It should tell you how

[rare goblet]

@pliant crescent has reached level 10.

[open ruin]

https://drive.google.com/file/d/1OJPN9g359vrmw1qRhkarU319OLo6h2D0/view?usp=drive_link this is the link pls anyone create me a model with this voice and share me index and path file pls

@pliant crescent i just want to train

[pliant crescent]

Oh yeah
There's also a guide for that

[open ruin]

@pliant crescenti did train model locally it trained fully but i only have 4 gb graphics card so its not making me a path file its getting stopped which i need its only giving index file

[pliant crescent]

Get a better GPU then
Or use the No Webui Colab

[open ruin]

its constantly disconnects

[rare goblet]

@open ruin has reached level 3.

[frank marsh]

Maybe someone uses Applio and got this error? An error occurred in the feature extraction, make sure you have provided the audios correctly.

[peak barn]

C:\Users*\Documents\ROD-v1.0.1_rod.py:10: SyntaxWarning: invalid escape sequence '_'
print("""
Traceback (most recent call last):
File "C:\Users*\Documents\ROD-v1.0.1_rod.py", line 1, in <module>
import gdown
ModuleNotFoundError: No module named 'gdown'
When trying to run ROD.bat for the Automatic Overtraining Detector I'm getting this error message even though I have gdown installed.

[oblique mason]

How do I make it so I can hear myself?

[steady ruin]

Can I continue training a model on a different Colab than the one I started training on?

[rare goblet]

@steady ruin has reached level 2.

[ember crow]

idk you gotta try it yourself

[steady ruin]

ok

[ember crow]

haven't tried it before but

[steady ruin]

https://colab.research.google.com/drive/1r4IRL0UA7JEoZ0ZK8PKfMyTIBHKpyhcw?authuser=1#scrollTo=Vc01O8jLgcFW cause I was using this one

and now I'm trying to resume training with RVC Disconnected

(linked in the info channel)

[proper shale]

ehhhh nk

it aint gonna work

i tried

[steady ruin]

shit

[steady ruin]

how do I continue training with the one I used first?

[proper shale]

Insert the same settings

Except for stuff like save frequency and total training epochs

You could change those

But everything else is the same

And don't process data or feature extract again

Thing is, that Colab is probably restricted, from my knowledge.

[steady ruin]

well I got disconnected after training for 3 hours and 12 minutes

so idk

[proper shale]

Oh so you have Pro or something?

Because uhm

It shouldn't be working

[burnt agate]

@proper shale sorry for the ping but do you use UVR?

[proper shale]

Nope

[burnt agate]

ahh ok no worries, I need someone to go through the small uvr guide i made to confirm or add anything lol

[proper shale]

oh

[steady ruin]

uh no

[burnt agate]

had to repost it here again lol

@low shard I had to retype all of this again but this is what I got from my testing with UVR

ree

forgot discord deleted edited messages lol

Please let me know if you feel like anything needs to be added or removed from this, I will do my own testing and see how the results are

UVR (Ultimate Vocal remover)

Click on the wrench icon next to start processing to open the settings of UVR, go to the download center and download the needed stuffs

MDX-net MDX23C InstVoc HQ is only available on UVR 5.6

Vocals

Removing instrumental from vocals, try one of the following (Check Vocals only)

• MDX-net MDX23C InstVoc HQ
• MDX-net Kim Vocal 2
• MDX-net Voc FT

Removing harmonies from vocals, try one of the following (Check vocals / No others only)

• MDX-net Karaoke 2
• VR-Arch 5_HP-Karaoke

Removing reverb from your vocals, try one of the following (Check no others only / No reverb only)

• MDX-net Reverb HQ
• VR-arc DeReverb / VR-arc DeEcho-DeReverb

Removing weird noise/hissing/artifacting sounds after removing reverb/echo, try the following (Check no others / No noise only)

• VR-arc DeNoise

Instrumentals

Removing vocals from instrumentals, try one of the following (Check Instrumentals only)

• MDX-net MDX23C InstVoc HQ
• MDX-net Voc ft
• MDX-net Inst HQ 3

Retaining Original Samplerate (tested on a 20khz sample rate audio and the audio was still 20khz after seperation)

• MDX-net MDX23C InstVoc HQ or MDX-net Kim Vocal 2
• MDX-net Karaoke 2
• MDX-net Reverb HQ
• VR-arc DeNoise lite

Please note that some of these might work better than others so test and see which works best for your audios

sorry for being kinda spammy lol

[low shard]

Thank you for all this

I wilk put the important ones in my guide later

[astral perch]

I have the graph thingy

How can I see if its overtraining

[reef marten]

which option is the best for vocal removal?

[prisma kettle]

anyone knows how to delete pretrain and model_dir folders ?

[sand iris]

What's the issue?

You just delete them

[prisma kettle]

where tho lol

"pretrain"

[potent root]

i use MDX Net models the most

but the other options are good too

isolation quality isn't based on the option you choose from that list, but whatever model you choose

voc ft is the best model for vocal isolation & removal i think

[sand iris]

You using the realtime voice changer right ?

[prisma kettle]

yup

[sand iris]

You unzipped a folder named MMVCServerSIO, and the folders to delete are within that

[prisma kettle]

thats all ?

[sand iris]

Yeah

[prisma kettle]

wow ...never thought it will be that easy lol

Big thanks!

[sand iris]

👍

[astral perch]

Anyone know how this works?

[mint stag]

is it normal if the index makes the voice come out more robotic ?

[golden elm]

hold on @red kayak does training with v2 mean training with this or this option

cause i feel like i confused it a bit

[red kayak]

yeah haha

[golden elm]

former or latter

[red kayak]

huh?

ur using an old gui from what im seeing

[golden elm]

does it mean training with RVC-beta-v2-0528 or the v2 model arch

[red kayak]

RVC-beta-v2-0528 is just the version of the build

[golden elm]

yeah i haven't updated it in a while

[golden elm]

so training with the v2 model arch

means v2

gotcha

[red kayak]

v2 model arch is basically the v2 weights

[golden elm]

where can i download the latest build again

[red kayak]

yeah but make sure to check v2

[red kayak]

-rvc

[azure marsh]

Hey, @golden elm!

👇 Here are some useful resources to help you learn how to make ai covers

Documentation

🇺🇸 English Documentation

https://us.aihubfrance.fr/

Credits: AI HUB France

🇺🇸 English FAQ

https://us.aihubfrance.fr/faq

Credits: AI HUB France

Guides

🍏 Applio-RVC-Fork: Guide

Link: Google Docs

Credits: IA Hispano

How to install RVC on windows

Link: YouTube

Credits: iroaK

RVC v2 Tutorial for Paperspace

Link: Google Docs

Credits: LollenApe

🌐 Translations

• 🇫🇷 French Documentation
• 🇮🇹 Italian Guides or -it_g
• 🇷🇺 Russian FAQ for RVC/Voice changer
• 🇧🇷 RVC Guides in PT-BR (-rvc_br)

[golden elm]

thanks

so i should use applio over the original rvc branch

gotcha

[burnt agate]

hope it helps out

[rancid knoll]

wrong graph, check the loss/g/total graph

[rare goblet]

@rancid knoll has reached level 1.

[rancid knoll]

overtraining is when that value goes up

[astral perch]

I cant find that one i only see d total

[rancid knoll]

there should be a search bar at the very top

if there isn't, go to the "scalars" tab

[astral perch]

Found it

[rancid knoll]

that's the graph you should look at to see overtraining

in this graph for example ~10500 steps is where overtraining began

[rancid knoll]

is the smoothing at .999?

[astral perch]

yes

[rancid knoll]

wow that's a weird graph

[astral perch]

do i need to change it

[rancid knoll]

but see the dip at around 17k steps?

[rancid knoll]

nah its good

[astral perch]

yes

i see it

[rancid knoll]

after that dip the loss value starts climbing

that's the point of overtraining

[astral perch]

wow so its already overtraining now?

[rancid knoll]

yeah it seems

[astral perch]

how can I see what epochs that is

[rancid knoll]

you can train for longer to see if it goes down

are you running locally?

[astral perch]

yes

[rancid knoll]

ok so

at the end of each checkpoint file (its in the "weights" folder) it ends with _s(step count).pth

[astral perch]

yes

[rancid knoll]

find the file closest to that step count

[astral perch]

so the dip is at 16800steps

so I need to find one below that

right

[rancid knoll]

yeah pretty much

[astral perch]

okay

but that is only 190 epochs

what happens if i train this model more?

[rancid knoll]

that's fine

[rare goblet]

@rancid knoll has reached level 2.

[rancid knoll]

i submitted a model at 165 epochs and it sounds fine

[rancid knoll]

you might overtrain or the loss might keep going down

i assume you have a long* dataset?

[astral perch]

9 minutes and 31 seconds

[rancid knoll]

ok reasonable

if you have gpu time to spare try training more

if it still goes up though cut your losses

a dip like that might not mean overtraining maybe

[astral perch]

so if the graph keeps going up its still overtraining?

[rancid knoll]

yes

[astral perch]

so it needs to keep going down

got it

[rancid knoll]

like the graph here

[astral perch]

I see

[rancid knoll]

it starts gradually going up

[astral perch]

And how do you wich dips indicates the start of overtraining cause I see multiple dips in my graph

[rancid knoll]

hmmm

the graph is a bit shaky yes

just look at the average

[mint stag]

What about this?

[rancid knoll]

cook this a little bit more

[mint stag]

Alright, thanks

[rancid knoll]

also can you turn the smoothing up to .999?

bit easier to spot it that way

[astral perch]

what does the smoothing do?

[mint stag]

middle seems to be overtrained but thenfalls below

[rancid knoll]

basically smoothens the graph to make it easier to find stuff

[rancid knoll]

thats what also caught my eye

[mint stag]

This scenario isn't mentioned in the FAQ

[rancid knoll]

yeah train this more

[mint stag]

so I have no clue

[mint stag]

alright

[rancid knoll]

can't really locate overtraining if the graph looks like a heartbeat graph

[astral perch]

what are those big blue lines in the background?

[rancid knoll]

that's the graph without smoothing applied to it

[astral perch]

okay

[mint stag]

@rancid knoll is it normal if index at 1 makes the voice sounds more AI'ish compared to 0.5?

[rancid knoll]

index at 1? sorry i'm not really that good at the inference side of things

[mint stag]

index at 1 with okada

[rancid knoll]

ah i've never used okada before

[mint stag]

oh I see

my bad

[rancid knoll]

folks at #🔍│help-w-okada will help you though

[astral perch]

I shouldnt use the d or g model right

only the model that saves to ''weights''

[rancid knoll]

yeah

the inference model won't accept the d/g models iirc

even if they do you will immeadiately hear why

[astral perch]

ok Im training it a bit more since that graph didnt really go up or down at 20k steps

[rancid knoll]

alright happy training! 🙂

[astral perch]

Thanks for helping me!

When I train from a checkpoint it says

Grad strides do not match bucket view strides. This may indicate grad was not created according to the gradient layout contract or that the param's strides changed since DDP was constructed. This is not an error but may impair performance.

My epochs are also just training how they normally would

[rancid knoll]

ah i just ignore that

it doesnt change anything for all i know

[astral perch]

Okay thank you

I used the model on rvc without index to try it out but it sounded really bad

[rare goblet]

@astral perch has reached level 5.

[astral perch]

But on the realtimevoicechanger it soundded pretty good

[rancid knoll]

huh thats weird

[rare goblet]

@rancid knoll has reached level 3.

[rancid knoll]

back then i almost never used index files and they sounded normal

[astral perch]

They control accent strength right

I never use it

[rancid knoll]

could be

im not really knowledgeable in the inference stuff 😅

[mint stag]

try 0.5

it seems to be the best

at least for me

[astral perch]

Okay!

[mint stag]

Should I still continue?

[rancid knoll]

it seems to have started overtraining at around 19500

can you turn up the smoothing a bit?

[rancid knoll]

hmmm

go on a little bit more

[mint stag]

alright

[stoic lintel]

hii i was wondering if someone could help me..
im unable to set my gpu setting to actually use my gpu and its stuck using cpu any help would be appreciated !

[rancid knoll]

right here does it list your gpu?

and does it say that it has found your gpu in the terminal?

[stoic lintel]

mine looks nothing like that..

[rancid knoll]

uhm ok

can you send a screenshot of the training page?

[stoic lintel]

in dms? im unable to send imgs in here lol

[rancid knoll]

thats weird

did you select roles from id:customize?

[stoic lintel]

yupp

[rare goblet]

@stoic lintel has reached level 1.

[stoic lintel]

now i can?

ok ig

[rancid knoll]

ah i see

head over to #🔍│help-w-okada

[rancid knoll]

folks over there will help you

[stoic lintel]

ok tyy

let me just copy and paste lol..

[mint stag]

the graph is so weird

it can't decide if it wants to go up or down

[astral perch]

Isnt it going down at 24

[mint stag]

it went up again

[rancid knoll]

i advise keeping smoothing at .999 or .998 until you are absolutely sure its overtraining or the graph seems awfully straight

[rancid knoll]

for now though you can keep training it seems

how long is the dataset?

[astral perch]

I saw a tutorial on youtube where someone said that you should pick the model that's the lowest on the graph

[mint stag]

12 mins but audio is ripped from game files so its high quality

[rancid knoll]

thats also correct yes

[rancid knoll]

alright thats good to know

[rancid knoll]

but training until the point of overtraining guarantees the lowest loss value anyway

[astral perch]

If thats the case shouldn't @mint stag use the at 24k since its the lowest there?

[rancid knoll]

there's no guarantee (yet) that that will be the lowest in the future

[mint stag]

Yep

[rare goblet]

@mint stag has reached level 3.

[rancid knoll]

it might go lower

[mint stag]

That's why I set it to 800

I thought it would be done around 500-700

[rancid knoll]

if it doesn't though that will be the one they should use

[mint stag]

yes

[astral perch]

Ahhh I see that's why people train it to like 1000 epochs

[mint stag]

Exactly

[dense marsh]

Is there any google colabs that work at this point

[rare goblet]

@dense marsh has reached level 1.

[astral perch]

But then its endlessly because you never know if it will go any lower

[rancid knoll]

some of my models needed 1200 epochs to overtrain but my most recent one needed 165

its a waiting game really

[dense marsh]

Cuz I can’t do local and I havent made a model in a while

[rancid knoll]

here's the fun part

if it starts overtraining it will never go lower

[rancid knoll]

try the one in the pinned message

i have no idea if it works though

[dense marsh]

Bet

[astral perch]

But when can you absolute sure its overtraining

[rancid knoll]

the point at loss/g/total rises up without going down

you can never be %100 sure but you can be pretty confident

i run it for an extra 150-200 epochs to be really sure but thats because i have a lot of time on my hands

[astral perch]

So the higher the epochs doesnt necessarily mean better quality

[rancid knoll]

yeah pretty much

some of my models sound the same at 300, 600 and 1200 epochs

its just they are more fine tuned

if a model is overtrained though you can definetly hear it

[astral perch]

I think I'm gonna stop training cause I dont have much time left

[rancid knoll]

thats fine

[rare goblet]

@rancid knoll has reached level 4.

[astral perch]

Also I think it keeps going up

[rancid knoll]

just take the lowest loss at the highest step count and you're good to go

[rancid knoll]

then you should take the model at the point of overtraining

[astral perch]

but the lowest is at 16k right

[rancid knoll]

wow that's a wild graph

set the smoothing to anywhere between .97 and .999

[rancid knoll]

yeah that sure seems like it

take the one closest to the dip

[mint stag]

show your d loss

[astral perch]

total?

[mint stag]

yes

[rancid knoll]

thats right that exists too 🤦‍♂️

[astral perch]

what is d loss

[mint stag]

yep overtrained

[astral perch]

How can u see that

[mint stag]

d loss should mirror g loss

[rancid knoll]

discriminator loss as opposed to generator loss

[mint stag]

your gloss is going up while d is down

thats a big nono!

[astral perch]

Wait lemme process thiss lol

[astral perch]

this was gloss total right

[mint stag]

yas

[astral perch]

Where is it going down on the d loss graph

[mint stag]

wdym?

[astral perch]

Nvm i see it

so dloss going down means overtraining

and gloss going up means overtraining

[rancid knoll]

not exactly

dloss going down while gloss is going up is overtraining

[astral perch]

so they both need to go up and down for it to mean overtraining

[rancid knoll]

yeah pretty much

[astral perch]

but on the dloss graph 20k is still going down

[rancid knoll]

but the gloss graph is going up

[astral perch]

Oh yeah right I see

[mint stag]

@rancid knoll

[astral perch]

so I should get the model thats a bit after the dip or before?

[mint stag]

done training

which one do i take

[rancid knoll]

i'd take the lowest around the 24k mark

[mint stag]

I already showed it to u

[surreal bridge]

when trying to open applio i get the error "No module named 'rpds'" anybody know a fix?

[astral perch]

i mustve missed it, sorry

16k right

[rancid knoll]

applio?

[rancid knoll]

yeah that one

[mint stag]

Yes

[mint stag]

applio is extremely confusing, I recommend using mangio

[surreal bridge]

ah okay

[mint stag]

applio is like a cheap knockoff to mangio to be honest

[surreal bridge]

i was just reinstalling stuff because tensorboard on mangio didnt even open for me and nobody was helping

[rare goblet]

@surreal bridge has reached level 2.

[mint stag]

nobody?

[surreal bridge]

but isnt it a mangio fork...

[mint stag]

yes, it's a version of it but many people say that mangio performs better than applio

[rancid knoll]

though to answer your question you should run pip install rpds in your terminal

[surreal bridge]

it was giving errors and people either ignored or didnt know how to fix it

[surreal bridge]

alr tried it, didnt work

and tried uninstall and reinstall rpds

[rancid knoll]

im out of solutions then sorry

[mint stag]

what's ur python version? @surreal bridge

[surreal bridge]

i think latest, why? pretty sure applio docs said python was optional as well for some reason

so like 3.12

[mint stag]

big no-no

let me give you a better version hold on

[surreal bridge]

ah okay, so thats where i messed up ;-;

[mint stag]

3.10

uninstall 3.12

and install 3.10

[surreal bridge]

tbh i think applio recommended 3.9.8 lol

[mint stag]

don't use applio....

[surreal bridge]

ok xd

[mint stag]

install 3.10 and let me know if tensor board issue is fixed

[surreal bridge]

alr will do

[rancid knoll]

ah i was just about to respond 😅

[surreal bridge]

ill prob just reinstall everything as well because my entire install of mangio is scuffed because of how much i had to change to fix it

[rancid knoll]

you didnt have to delete

[surreal bridge]

;-; i just hope that uninstalling and reinstalling all of my python stuff doesnt break my code too much

[astral perch]

Im sorry

i was worng g is going up and d is going down

i used the 16,600 step

[surreal bridge]

hey so is this the right repo? https://github.com/Mangio621/Mangio-RVC-Fork

[rancid knoll]

seems like it

[rancid knoll]

try testing it to see how it sounds

[astral perch]

I will now

[rare goblet]

@astral perch has reached level 6.

[rancid knoll]

and if its good enough you can upload the model here lol

[surreal bridge]

yay mangio install script instantly detected as a virus by bitdefender

;-; i hate my antivirus sometimes it has so many false positives

[rancid knoll]

havent had a ransomware attack yet 😄

i'll try using applio to see what's the fuss about it

[surreal bridge]

3.10.0 or 3.10 latest?

[astral perch]

where

[surreal bridge]

or ms store version

[rancid knoll]

3.10 latest should be fine

[rancid knoll]

#1159514067187277865

after it gets approved you can post it to #1159289700490694666

[surreal bridge]

are there any differences between installer version and ms store version?

[rancid knoll]

i dont know of any

but i prefer using the installer

[mint stag]

use instalelr

[sand iris]

Where's the installer from ? I'd trust Microsoft more for that purpose

Since the python org doesn't provide installers for newer security updates of older python versions

[astral perch]

rmvpe is the best pitch extractor right?

[rancid knoll]

yeah ive had the best results with it so far

there are some cases where others can be better though

[astral perch]

for me crepe is best with songs and rmvpe for speaking

[surreal bridge]

well this is just great, when using mangio as well it gives the rpds error

[mint stag]

If you data set is less than 10 I would say mangio crepe would be betetr but if its more than 10 then definately rmvpe (Just my opinion based on experience, not factual)

[rancid knoll]

that's weird...

[mint stag]

with the new python version?

[surreal bridge]

with 3.10

[astral perch]

the voice is good just some artifacts in the qualty

[mint stag]

did you add it to path when installing?

[surreal bridge]

yep

[rancid knoll]

(happened to me before) are you sure mangio is using the right python installation?

[mint stag]

have you restartted pc?

[mint stag]

oh yes that as well!

[surreal bridge]

oh, no i havent

[surreal bridge]

theres only one version installed

[rancid knoll]

mangio has a runtime/python installation embedded into it

[surreal bridge]

it does?

[rancid knoll]

yeah

[surreal bridge]

is there a way to change it?

[rancid knoll]

in the go_web.bat file replace any "runtime/python.exe" with just "python.exe"

that worked for me

[surreal bridge]

by the way i think when I install the dependencies in requirements.txt it installs the cpu version of torch, do i need to manually install the cuda version?

well i think it does

[rancid knoll]

i had to manually install cuda yeah

[surreal bridge]

rip

[rancid knoll]

it didn't recognize my gpu the first time

[surreal bridge]

how do i know which cuda version to select when installing torch

[rare goblet]

@surreal bridge has reached level 3.

[rancid knoll]

been a while since i installed it

so i forgot sry

[surreal bridge]

are you kidding me ive already gone up 2 levels just trying to fix all of the issues im getting xD

[rancid knoll]

thats the rvc experience 😄

[surreal bridge]

its all good ill just guess xd

[mint stag]

@surreal bridge you haven't installed cuda manually?

[surreal bridge]

hey i mean a few more levels and i might even get verified so im happy

[surreal bridge]

i have cuda installed but the dependencies.txt installs the cpu version of torch

[rancid knoll]

back then i had some models start out with ~490 loss

[rare goblet]

@rancid knoll has reached level 5.

[rancid knoll]

spent a while fixing that 😄

oh look im verified now

[surreal bridge]

congrats

[rancid knoll]

note to self: never delete the pretrained values

[burnt agate]

also dont delete the folder called mute in the logs folder

[brittle wing]

can anyone fix this please? i cant make voice models.....

[rancid knoll]

havent made that mistake before (yet) 😬

[rancid knoll]

edit the "go_web.bat" file and replace any "runtime/python.exe" with "python.exe"

[rancid knoll]

what happens if you do?

[brittle wing]

what did i do wrong

show me please a video of hou you did it

the program just closes

[rancid knoll]

you didnt do anything wrong, its just that mangio is fucked upon first install

[burnt agate]

cant really recall but it will mess up a lot of stuff lol

[brittle wing]

how do i edit the bat file @rancid knoll im so confused

[rare goblet]

@brittle wing has reached level 5.

[rancid knoll]

hang on im sending a video

[burnt agate]

also install python version 3.9.8

[rancid knoll]

im really tempted to delete that folder

[burnt agate]

make a backup and test it

[rancid knoll]

also god damn applio installs a ton of stuff on first run

[brittle wing]

it still did not work

[rancid knoll]

hmmmmm

what error did you get now?

[brittle wing]

i give up

@rancid knoll

[rancid knoll]

hmmm

[burnt agate]

do you have python installed?

[brittle wing]

yea

@burnt agate yes i have

[burnt agate]

what version?

[brittle wing]

3.111

3.11*

[burnt agate]

uninstall it and downgrade to the recommended stable version 3.9.8

[brittle wing]

okey?

[mint stag]

why does python use double \ \

[burnt agate]

doubt thats whats causing your issue tho, remove any spaces in the names of the folders

[rancid knoll]

since some special chars start with \

like \n \t \r etc

\\ escapes that \

[brittle wing]

show me a video of you fixing the isssue

@rancid knoll

[surreal bridge]

okay now my mangio is even more screwed

[brittle wing]

how???

im super confused

[rancid knoll]

uhm

[rancid knoll]

just move the folder to your desktop

[brittle wing]

i did but i still get the same error message @rancid knoll

[rancid knoll]

hmmmm

sorry man im out of ideas

[brittle wing]

so what now?

@burnt agate @rancid knoll

[rancid knoll]

try moving the folder inside the folder that ends with (3)

[surreal bridge]

diff error this time, reran the requirements.txt install in pip

[rancid knoll]

ah yeah this error now

just manually install the packages from this point on

[surreal bridge]

alr

[rancid knoll]

thats how i got it fixed

[burnt agate]

is python added in your system environment variables?

[brittle wing]

how show me a video im super confused

@burnt agate

[surreal bridge]

alr it opens now but its still using cpu

now i just gotta change it to gpu version of torch right?

[rancid knoll]

yeah i guess

[surreal bridge]

cuda*

[rancid knoll]

why is mangio so fucked at first start lol

half of the dependencies missing and all that

[proper shale]

(applio is even worse)

[surreal bridge]

why cant someone make an easy thing to install

[rancid knoll]

true

currently working with an applio dependency issue 😄

[surreal bridge]

applio also had a dependency issue for me

rpds or smth like that

[proper shale]

because it has 999999999 dependencies (joke ofc) and most people wont even use all of em

[rancid knoll]

where do i place the model in applio lol

dont see a weights folder anywhere

[surreal bridge]

yay my mangio finally works

now i need to figure out how to get tensorboard to work ;-;

this is where it all broke last time

[rancid knoll]

there should be a TBI.bat and a tbL.bat in the folder

run TBI.bat once to install and tbL.bat every time you need to run tensorboard

[surreal bridge]

nah neither are there

[rancid knoll]

if those 2 files aren't there i can send them

just simple 1 line files really

[surreal bridge]

wait it says intel

is it installing the cpu version

[astral perch]

@rancid knoll Thanks for helping me i submitted my model

[rancid knoll]

uh

idr if that was the correct one

[stoic cloud]

Oh intel and amd stuff is needed, since sometimes in AI stuff just runs on CPU

[rancid knoll]

sorry wrong message

[rancid knoll]

np!

[stoic cloud]

Tensorboard is not something that contributes to learning, because it just shows you what the AI has been doing, so it's completely fine if that runs on CPU

[surreal bridge]

ah okay

[rancid knoll]

also is there a support server for uvr5 models?

[surreal bridge]

hey so idk if this is an issue or not

[rancid knoll]

need a model to take just the piano out of a song

[rancid knoll]

well it seems it worked in the end so i'd just ignore the errors 😄

[sand iris]

Not sure why it's installing tensorflow though

I'm 90% sure it's useless

[surreal bridge]

is there a guide for learning to make models that anybody would recommend? or a sample dataset to learn making them with

[rancid knoll]

https://docs.google.com/document/d/13ebnzmeEBc6uzYCMt-QVFQk-whVrK4zw8k7_Lw3Bv_A/edit#heading=h.bjzhhhcn3f69 i used this one

the dataset making is the hardest part really

[wispy burrow]

For real!

[rancid knoll]

after the dataset is done you can just get up from your computer and go away while it does the hard work 😄

[surreal bridge]

is there like a minimum dataset size or anything

[rancid knoll]

i recommend at least 5 minutes

try to get at least 10-15 though

more is always better

[wispy burrow]

a Minimum of 10 minutes is recommeended.

[surreal bridge]

ah okay

;-; i just realised the audio I have is mixed with a bunch of voicemod soundboards and stuff, that will be annoying to cut out

[rancid knoll]

welcome to model making

[surreal bridge]

is there any sort of maximum dataset size? xd

or does it just take a long time

[rancid knoll]

nah you can go as long as you like

[rancid knoll]

epochs just take longer yeah

but because more information per epoch gets saved it cancels out

leaving you with a higher quality model

[surreal bridge]

does the dataset being stereo/mono/surround sound make any differences? or is it all converted before training

[rancid knoll]

i mean you can even use mp4 files for the dataset so im pretty sure it doesn't make a difference*

[surreal bridge]

what are good settings for truncating silence in audacity? i have like 2 hours of audio i gotta cut down

or is that the wrong thing

oh noise gate

[rancid knoll]

i havent used audacity much sry :/

[surreal bridge]

oh what do you use?

[rancid knoll]

i usually manually cut audio files

which is bad i know 😄

most of my experience comes from debugging mangio rather than using it

[rare goblet]

@rancid knoll has reached level 6.

[wispy burrow]

Truncate silence is certainly something you want to do.
Don't have silence that's longer than 0.5 seconds.

So far that's the only thing I've used Audacity for.

[surreal bridge]

alr thanks

[rare goblet]

@surreal bridge has reached level 4.

[wispy burrow]

Something like UVR (Ultimate Vocal Remover will be very useful too.
It's for isolating vocals and removing echoes/reverb.

[surreal bridge]

yea i alr have it

unfortunately it cant remove soundboards ;-;

[wispy burrow]

Never knew there was a W-Okada colab...

[half cove]

💀

[wispy burrow]

Maybe ask in #🔍│help-w-okada

[surreal bridge]

is there any sort of text to speech software that is good for using with voice models?