✨│ai-help

[analog obsidian]

now u can send pics here

is stuck?

is normal, is downloading the requires model to run the voice changer

[brittle wing]

oh i see and will it run without an gpu

?

[analog obsidian]

normal wokada no, but modded wokada can run in some IGPUs tho im not an igpu user and the only person who knows more than me is the maker of modded wokada

[brittle wing]

oh no i meant to say wait can i send a url

?

[analog obsidian]

what?

you have a gpu or not?

[brittle wing]

no i don't

ig i have igpu

[analog obsidian]

https://github.com/deiteris/voice-changer

this one is the modded version

[brittle wing]

its ez to run

?

[analog obsidian]

similar interface to wokada's

[brittle wing]

i see

[brittle wing]

check that out if it would run

[analog obsidian]

i was reading emojikage's comment

[brittle wing]

?

[analog obsidian]

he said intel uhd is too weak to run the voice changer

[brittle wing]

i see

so there are no options available for me

[analog obsidian]

if you don't meet both cpu only or* gpu only conversion minimum requirements, then no

[analog obsidian]

these

[brittle wing]

what is cpu minimum requirement

[brittle wing]

i have a cpu requirement

[analog obsidian]

cpu only conversion uses over 80% cpu and is slow, so is not really viable

[brittle wing]

the cpu requirement is fulfilled

[brittle wing]

i see

[brittle wing]

i mean the program ran smoothly

[analog obsidian]

o yea i mean, the gui is light, what is intensive is the voice conversion

[rare goblet]

Ayo? @brittle wing level 2 !!!

[brittle wing]

i see

can't i close that cmd

prompt

?

[analog obsidian]

you can try it, click cpu and increase chunk to around 400 ish

however keep in mind is gonna lag a lot

so my recommendation is not do it

[brittle wing]

how to

?

[analog obsidian]

first click any of the anime characters, then click chunk and change the value to 400 or higher
extra keep it at 4k
gpu(dml) click where it says cpu

[brittle wing]

ok

[analog obsidian]

after that click start

but remember, its gonna lag your system a lot

so be careful

[brittle wing]

how can i run

if it

doesn't work

?

[analog obsidian]

its most likely not going to work because the system is very weak

your best option is the colab

it runs the voice changer on the cloud

[brittle wing]

yeah i

quitted

the progrm

[brittle wing]

how to do colab

?

brb in 15-20 min4

[analog obsidian]

https://docs.google.com/document/d/e/2PACX-1vTIceEcBfS6Zqolv_QEysrFfI_EJikPxozWptP_EjkpLVl-l-gdo-ijBonQMTviAHEYm5emmd9k9TdC/pub

the whole guide is there, explaining everything

[brittle wing]

ok sure

please help me in need

is there any risks

?

for my desktop

?

[analog obsidian]

everything should be fine

you only download one driver

[main tinsel]

deiteris voice changer issue.
It seems like the modulation of deiteris' voice changer became strange at some point. What I checked was version b2174 later. When you play the latest version, even though it is the same model, the sound is a bit more bizarre and you hear more mechanical sounds.

[analog obsidian]

then the rest is in the cloud, your system resources arent used

[brittle wing]

ok please teach me how can i remove the one i downloaded

the official one

[analog obsidian]

just delete the folder

[brittle wing]

ok sure

then now teach me

[analog obsidian]

@wispy lodge (he's not online now so he might not answer now)

[brittle wing]

how can i install that

[analog obsidian]

read the guide, everything is literally there

[brittle wing]

please

[analog obsidian]

the guide isnt that long tbh

[brittle wing]

teach me

[analog obsidian]

https://docs.google.com/document/d/e/2PACX-1vTIceEcBfS6Zqolv_QEysrFfI_EJikPxozWptP_EjkpLVl-l-gdo-ijBonQMTviAHEYm5emmd9k9TdC/pub just do everything is said here

[brittle wing]

just teach me how to

download virtual audio cable @analog obsidian

[analog obsidian]

just click this

[brittle wing]

ok

[analog obsidian]

and follow those steps*

[brittle wing]

ok

we can use it same as the normal one with diffrent models right

[analog obsidian]

yes, you can upload your own models

[brittle wing]

which one is the best model for a boy

?

[analog obsidian]

uuhh this depends in your taste, how do u want to sound and such

[rare goblet]

Ayo? @brittle wing level 3 !!!

[analog obsidian]

https://colab.research.google.com/github/hinabl/voice-changer-colab/blob/master/Hina_Modified_Realtime_Voice_Changer_on_Colab.ipynb

[knotty moth]

try the latest b2245

[analog obsidian]

i didnt made the colab so i cant fix that

but if i have to guess, you probably didnt the ngrok step

https://docs.google.com/document/d/e/2PACX-1vTIceEcBfS6Zqolv_QEysrFfI_EJikPxozWptP_EjkpLVl-l-gdo-ijBonQMTviAHEYm5emmd9k9TdC/pub

read this

wat

[knotty moth]

try again, and don't dismiss the prompt or be too late to respond

[analog obsidian]

do step 2 of the google doc i sent u

click this

[knotty moth]

@brittle wing why delete ur comments? 💀

bruh automod?

[brittle wing]

hey @knotty moth

[knotty moth]

[brittle wing]

i can't listen

its connected to virtual audio card

what shall i do

?

[knotty moth]

idk cuz I haven't used it

only the local installation yet

[brittle wing]

oh

can i close the

collab file

after the

voice changer runs in cloud

[knotty moth]

it would disconnect the session

[brittle wing]

oh ok

i won't

[queen kraken]

I trained a model using klm till epoch 200 then I stopped. Later, I resumed the training but didn't realize that I hadn't changed the pretrain to klm. As a result, the training continued with the og till epoch 400. I thought I had messed up the model but decided to give it a try, and it turned out to be pretty darn good.

[analog obsidian]

because og is superior

xD

[queen kraken]

I really thought that changing the pretrain in the middle of training would mess up the model

[analog obsidian]

you did what o my i just read what u said again

im tired tbh xD

damn i mean uuhhh

[misty elk]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• UVR5 NO UI for Google Colab, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab

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

[analog obsidian]

yeah idk i have never tried to change pretrains mid training lol

[brittle wing]

@analog obsidian

[queen kraken]

I might try again xD

[brittle wing]

it is not really working

[analog obsidian]

idk i didnt made the colab nor the google doc guide, srry nothing i can do

[brittle wing]

what is the

settings required

@analog obsidian

[rare goblet]

Ayo? @brittle wing level 4 !!!

[analog obsidian]

input real mic
output line 1

[brittle wing]

and outpot

*output

[analog obsidian]

line 1

[brittle wing]

server io anaylzer

[analog obsidian]

your speakers

[brittle wing]

?

[brittle wing]

what about

[analog obsidian]

output device put your headphones

[brittle wing]

input

?

[analog obsidian]

line 1 on disc

real mic on wokada

[brittle wing]

wdym

i don't get it

@analog obsidian

[analog obsidian]

set your input device to line 1 in discord

[brittle wing]

which one

line

1

[brittle wing]

?

[brittle wing]

then it won't detect the mic

even if i am speaking

@analog obsidian

[analog obsidian]

where it says output put line 1

on wokada

[brittle wing]

yeah i have done it

[analog obsidian]

increase tune to 12

move the slider

[brittle wing]

index one

[analog obsidian]

i said tune

...

[brittle wing]

by how much

[analog obsidian]

12

[brittle wing]

done

then

[analog obsidian]

start wokada

and listen to yourself on discord

[brittle wing]

alr started

[misty elk]

-rvc

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.wtf/

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• Full AI Voice Model Training Guide (local) by Christopher Villanueva
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[analog obsidian]

can u stop pinging me

[brittle wing]

ok sure

i won't ping you

[analog obsidian]

if aint working then idk

i dont use colab

[brittle wing]

it seems to be working fine from my normal mic but it won't pick a voice from line 1

why is that

[analog obsidian]

idk it should work, the only person that can really answer this issue is someone that uses colab

[brittle wing]

who does

le me ping him

[pastel oak]

Can you screenshot the voice changer again with what you set up so far

[knotty moth]

double check the voice changer's output and also some other settings in the control panel

[brittle wing]

ok sure

[analog obsidian]

shad knows this so ask him for help

[pastel oak]

Ayo i dunno anything about colab 👀

[analog obsidian]

me neither but i dont have more patience lmao

[pastel oak]

the bottom too

[pastel oak]

Ok you didnt set it up correctly, youre gonna do the following:

[brittle wing]

?

[pastel oak]

on voice changer, the screenshot u just sent

Input: Microphone (correct)
Output: LINE 1

And on discord, you do:

Input: Line 1
Output: Headphones

Then try again, it should work

[brittle wing]

i alr did

u meant my voice changer need no change of setting right

?

[pastel oak]

You have your "output" set to none. It has to be Line 1

[brittle wing]

ok

then

[pastel oak]

Then on discord you do Input: Line 1 and try again

[brittle wing]

ok

[knotty moth]

note that if you have a decent Nvidia/AMD gpu, or a decent modern cpu, you can try the local fork, so I might help you more https://rentry.co/forkvoicechangerguide

[brittle wing]

nope

[brittle wing]

i dont

[pastel oak]

Send screenshot of voice changer again one last time to check if everything is ok

[brittle wing]

it was working in

[pastel oak]

Alright, now click on the green button "stop" then "start again"

[brittle wing]

ouy

out

that recorded one its working

[pastel oak]

If it still doesnt work.. then restart the app, and make sure you allow your browser to use your microphone

[pastel oak]

really?

Then I got no clue what the issue is rn

[brittle wing]

yeah

[pastel oak]

Dont know what else to do then

[brittle wing]

peace of a crap

i wanted to prank as a boy to my friends

[knotty moth]

if the internet latency isn't good enough, it won't be good as well tho

[quasi night]

can i get the latest version of rvc?

[misty elk]

-rvc

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.wtf/

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• Full AI Voice Model Training Guide (local) by Christopher Villanueva
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[misty elk]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• UVR5 NO UI for Google Colab, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab

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

[wise estuary]

Im trying to separate vocals but i get this error message everytime.

FileNotFoundError: [Errno 2] No such file or directory: 'C:\RVC1006Nvidia\TEMP/Timeline 1i85h47wb.wav.reformatted.wav'

I did the exact same thing thing as always, just a newer version

[knotty moth]

you need this? https://github.com/Anjok07/ultimatevocalremovergui/releases/tag/v5.6
the built-in one in RVC doesn't have good new model options tho

[main tinsel]

b2245 also. b2174 later version makes artifact

[wispy lodge]

Which settings? Could you also record the output?

I never really noticed 🤔

Well, yes, you should download directml version. It will show your gpu

[ripe peak]

i need someone's help

[solemn walrus]

whenever i open start http the app opens but its just a white screen

[wispy lodge]

Just in case, if it was you who created an issue on github, I double-checked and left the comment there with spectrogram and samples attached. I cannot confirm that issue on my side, but it would be great if you could demonstrate or clarify what exactly I should look at or how I can reproduce this
https://github.com/deiteris/voice-changer/issues/162#issuecomment-2266649144

[candid oracle]

这个要怎么玩啊

How should i use this app

[solemn walrus]

i went into the app and everything but everything is in another language and idk what to pick mmv sovits or rvc

[rare goblet]

Ayo? @solemn walrus level 2 !!!

[solemn walrus]

can anyone help?

[misty elk]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• UVR5 NO UI for Google Colab, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab

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

[merry eagle]

~rvc

Hey can some one give me Real time voice changer download link

[proper shale]

https://rentry.co/VoiceChangerGuide

[misty elk]

Does custom pretrains increase your training time

Like

Does it slow down training time

I js wanna know

[regal saffron]

As far as I understood, Beatrice_v2 is a new model? Can you convert RVC to Beatrice_v2 or am I comparing cats with dogs?

[knotty moth]

different architecture

[rare goblet]

Ayo? @knotty moth level 25 !!!

[knotty moth]

no converter yet

[regal saffron]

So is there only one Beatrice_v2 model or are there "more" somewhere?

[timid pier]

Yall, my RVC GUI.bat wont work. Im trying to make a ai cover song but the app won't work. Any help?

[ripe peak]

Can someone help me

[pastel oak]

Ask the question in here

[brittle wing]

im using Mangio-RVC-v23.7.0

is that good?

[pastel oak]

I dont know about rvc gui, but if you need an urgent alternative, use Ilaria RVC Zero (pinned message, the ine for "inference") with the guide also linked

[pastel oak]

Good

[pastel oak]

Never seen people create beatrice models ngl rvc is just the more popular choice.. So not sure where to find them, maybe theres a filter for beatrice on weights.gg?

[proven hill]

no theres not and you need a lot of vram to create beatrice models

[pastel oak]

^ @regal saffron

[pastel oak]

When it asks enter the path of audio you only selected the folder instead of audio

So it has to be C:\blabla\Downloads\audiofile.wav (whatever format it is) for example

[brittle wing]

OHH

can mp3 work

[violet heron]

Yes

[brittle wing]

for some reason only wav file worked

[violet heron]

Make sure the file name has no spaces

[brittle wing]

alright

btw do u know any good voice models

that are already trained

i only used appoilo before

[violet heron]

#1175430844685484042

[brittle wing]

some of them r really bad

[violet heron]

voice models moment

[brittle wing]

do you know how to add pth files to llaria rvc

[rare goblet]

Ayo? @brittle wing level 3 !!!

[violet heron]

model loader section

[brittle wing]

can i screen record

[violet heron]

Why do you hold every message

[brittle wing]

uh

[violet heron]

Why do you need to?

[brittle wing]

idk where to add index file

so applio is basically rvc? @violet heron

[brittle wing]

i tried appilo and it was trash

https://tenor.com/view/trash-garbage-dumpster-gif-22255810

[violet heron]

It’s a fork

It’s built off of it but there’s no real reason to use it over mainline

[violet heron]

On the model uploaded you can upload the pth and index

[brittle wing]

srry i forgot to restart rvc

i was being dumb

😼

[chrome shard]

i cant open the rvc it gives me too many errors

[bold magnet]

yo does the dataset needs to have specific audio files? like .flac .mp3 or smth?

[rancid folio]

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

[violet heron]

What’s the error

[violet heron]

Preferably wav or flac if the audio was originally that fomrat

[bold magnet]

ight thanks

[pastel oak]

It is explained on the docs.
https://docs.aihub.wtf/essentials/how-to-make-voice-models/

[rancid folio]

Can anyone help create a voice model?

[violet heron]

-docs

[azure marsh]

Suggestions for @rancid folio

Documentation

AI HUB Docs

https://docs.aihub.wtf/

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• Full AI Voice Model Training Guide (local) by Christopher Villanueva
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[chrome shard]

i cant send ss can i dm you ?

[chrome shard]

C:\Windows\System32>python.exe rvcgui.py --pycmd python.exe
Traceback (most recent call last):
File "C:\Windows\System32\rvcgui.py", line 4, in <module>
import soundfile as sf
ModuleNotFoundError: No module named 'soundfile'

C:\Windows\System32>pause
Press any key to continue . . .

[rare goblet]

Ayo? @chrome shard level 1 !!!

[violet heron]

Where did you download it from?

[chrome shard]

github

[violet heron]

Send the link

[chrome shard]

https://github.com/Tiger14n/RVC-GUI

idk whats going on

just not working at all

i think its not installing the requirements .txt

[violet heron]

RVC GUI is very outdated

[chrome shard]

have u got a better one

[violet heron]

-local

[azure marsh]

🖥️ Local stuff

• 🍏 Applio, by IA Hispano GitHub
• Mangio-RVC-Fork, by Mangio621 Huggingface
• RVC Studio, by SayanoAI Huggingface
• AICoverGen, by SociallyIneptWeeb GitHub
• Replay, by Replay Team Website
• Original RVC, by the RVC-Project team GitHub
• GPT-SoVITS, by RVC-Boss GitHub

Credits to Faze Masta and Antasma for compiling these links.

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

[violet heron]

I personally use orginal RVC

[chrome shard]

im trying apollo if it dont work ill use orig

[weak sandal]

is there any website that removes the blank parts of ur audio file and makes it smaller audios?

[fiery raptor]

applio acting weird rn

what is this

nvm

it works

[odd shale]

I'm not sure, but you can manually cut out silent parts on your audio using Audacity.

[weak sandal]

oh i thought it was some sort of ai website that cuts out immediately

[red kayak]

Breathing yes a couple of breathing samples should be enough but laughing doesn't really work @crude flame

[woeful copper]

where do i download rvc lol

the one where you put ai models in it and then vocals and it makes them sound like the model

[red kayak]

https://discord.com/channels/1159260121998827560/1159513888199540817

[woeful copper]

thanks

[rare goblet]

Ayo? @woeful copper level 1 !!!

[quaint dock]

hey guys, is that a problem if I dont have 5 samples for the new model?

like for example I only have 1.wav and 2.wav?

[odd shale]

Do you mean, samples for posting a model you made?

I think 2 or 3 are fine.

[quaint dock]

no, like making a model

sorry for not being clear

[worldly totem]

guys

[rare goblet]

Ayo? @worldly totem level 2 !!!

[worldly totem]

help, onnx not working

i export to onnx

but not working

need help

[odd shale]

It's fine.

Which is the lenght of your samples you wanna use for making your model?

[sharp glacier]

sadly today i doubt i can get it to work i'll try tommrow i got it working before on here but thanks for helping me get this far.

[quaint dock]

one is 6:31, the other is 20 sec

[odd shale]

I think 6:31 mins is enough.

Make sure the audio is clean btw.

[storm kernel]

-help

[azure marsh]

Command has expired

[sturdy apex]

I can't find any voices from French YouTubers

[pastel oak]

People have to create these, so theres probably nobody that created them yet, which you could do yourself

[sturdy apex]

ok

thanks

[quaint dock]

alright, thanks. yeah its clean

[odd shale]

You're welcome mate.

[quaint dock]

I make sure to have some noise supression on the sample so I think its fine

[odd shale]

Alright buddy.

[slate harbor]

I have a 2060 super.. whats the best route for AI voice changer? Currently having cutting out issues on okada

[brittle wing]

Cna anyoone help me out? i got a problem with the AI voice and it says Pipeline is not Initialized and i cannot use the models voices

[rare goblet]

Ayo? @brittle wing level 1 !!!

[brittle wing]

I been having this problem for like a few days. im pretty sure i got python and pytorch installed aswell

but not comepletely sure on it

anyoone got solutions to these?

[gilded harbor]

How do i use the ai voice models 😭

[storm kernel]

-hf

[azure marsh]

<:huggingface:1179800228946268270> Hugginface Spaces

• UVR5 UI, by Eddy and Ilaria Huggingface Spaces
• Ilaria RVC Zero, by Ilaria Huggingface Spaces
• RVC⚡ZERO, by r3gm Huggingface Spaces
• Applio, by IA Hispano Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

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

[quaint dock]

sorry if I ask too many questions, does the quality depend on the epochs?

[thin stump]

https://docs.aihub.wtf/rvc/resources/datasets/#introduction consider the clarity of the voice, consistent mic proximity, and see if the frequencies can stay above 15khz

[thin stump]

oh nice, I was right all along. It just took me longer to say the same thing this guide says in the FAQ https://docs.google.com/document/d/1wTJ_wutDqEtsA99BJOXDDGax25pPIDE84O5E2Rio5Qk/edit

[quaint dock]

alright, thanks!

[molten fog]

does anyone know if i have a bunch of acapellas and i combine them into one 20 minute audio file using audacity and export it as wav and then use isotop rx11 to clean the extra noise, will that affect quality as oppsed to doing it 1 by 1?

[thin stump]

If the frequencies are flunctuating, ig that's okay. Just keep in mind of the consistency of the accent for the singer because people do ERA models for a reason

Taylor swift 1989 era etc

[molten fog]

yep im careful to eliminate things like when the singer is using more of a yelling singing tone as opposed to casual singing

[solemn walrus]

how do i put the voice model on the rvc new version?

cause i press upload and it doesnt work

[molten fog]

also should i remove every single silence in my dataset? like silence between words that the singer is singing as well?

[rare goblet]

Ayo? @molten fog level 1 !!!

[thin stump]

that gets removed during the audio labeling method eventually

assuming you have the rx11 guide?

[molten fog]

i have the rx11 guide yes

but how do i audio label

and does it take a long time

because this dataset is projected to be around 20-30 minutes

so manually removing everything is not really effective and id rather truance silence at that point

[thin stump]

thats near the very end. After cleaning dataset > resampling to 40k or 32k on RX> auburn noise gating > audio labeling https://rentry.co/RVC-dataset-RX11#noise-gating-and-audio-labeling

[molten fog]

oh its not a manual process i just saw that

thank you

[thin stump]

yup it makes your the graph on the tensorboard much smoother on the tensorboard, mostly

[thin stump]

yeah those are only suggestions in the guide. It just won't capture everything accurately if we auto-declick or de-ess

I used to do that and my models were still fine

[molten fog]

understood

[errant pier]

does anyone know why when i try to use the voice changer and i talk into the mic the outcome is just gibberish and i non stop hear whispering from the voice changer

[violet heron]

Turn on sup2

Make sure f0 det is RMVPE

[ornate ore]

Question,what does sup1 and sup2 mean in RVC?

I haven't messed with most of the settings in RVC. Just the latency and chunk

[teal ridge]

What has been going on for my new models on weights.gg? They come off as a verification error.
First, my Goblin Caught on Tape model, ||which has been reuploaded already because I can't fucking upload any more on this site,|| and now my Tattletale Strangler model.

[thin stump]

I think you should contact Beatsboy this time. He's going to ask a lot questions just for a simple troubleshoot

[teal ridge]

thx m8
I already let him know.

[solemn walrus]

does anyone know how to put the models i put the index and path file and i uploaded it in the slots and i cant use it

[proven gate]

alguien en español :c

[molten fog]

yo when i use spectral denoise, do i have to select certain parts to denoise? or can i just press learn without selecting any specific parts and it tries to denoise the whole dataset?

[thin stump]

you have to select the noise. they look like this

if you press learn without selecting specific parts, it won't denoise anything

[molten fog]

oh alright well i already ran pure stem acapellas through bs roformer so i dont have much noise

[thin stump]

you have blue noise left probably if you zoom in

[thin stump]

you can get little noise profiles

now if there's blue noise inside an audio then you'll have to RX dereverb that

[molten fog]

when i export do i do wav 32 bit int or float

or does it not matter

[thin stump]

wav float

[rare goblet]

Ayo? @molten fog level 2 !!!

[molten fog]

do you think these 2 different styles of vocals should be trained in the same dataset?

not fully cleaned yet just sending a sample

[thin stump]

these are two different voices imo. Like I wanted to do a Deadpool model and when I focused more on Ryan Reynolds, it doesn't sound like deadpool

i got lost in the sauce because I didn't prioritize accent

overall I liked the dbtuavocals because its so clear already

[molten fog]

Problem is I don't have enough from that era in music to make an accurate model compared to the others

I have stems for the era I'm currently doing and I have a lot of songs for the dbtua vocals era but most of the songs are layered and its hard to produce a clean separated on the parts that aren't layered anyway

[thin stump]

yup then you have no choice then to mix both of them. It might sound like them but be a bit off

[molten fog]

I have abkut sxactly 10 min of data without mixing the eras

[thin stump]

oh then 5-10 minutes is fine

[molten fog]

7 minutes worth in stems (removed some because he was yell singing) and 3 minutes from vocal separation

Can you confirm that the artist singing harder / kind of yelling would not mix well with his normal singing styyle because I know that it probably doesn't

[thin stump]

he can do yelling if you want to keep that in. We're just cautious about screams because it can ruin the model

grunting is bad too

[molten fog]

It’s not screaming it’s just different like lemme send an example

[molten fog]

compare those 2 to icarusvocal

different singing style

[thin stump]

imagehook is basically ruined because you can't remove that noise. I don't think Bandit V2 will help out either, test it out

stephook still has a little bit of backing vocals left. A bit questionable because of quality so just use UVR BVE

[molten fog]

In terms of singing style do you think it is still ok to put in the dataset

Because I do not think rvc can differentiate between vocal tones like casual singing and more energetic almost yell singing

Those are just 2 examples of vocals I have cleaner ones tho from stems

[thin stump]

rvc just distributes it to a pitch so I think you should leave it out for casual singing. It's not for realtime anyways

[molten fog]

That’s what I thought

[thin stump]

ppl just want to push the boundaries with their ai covers and they're going to be disappointed anyways because rvc isn't all that

if they're doing a metal cover on a casual voice etc

[open stag]

• colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• UVR5 NO UI for Google Colab, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab

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

[molten fog]

Yea I wish I had enough data to make a casual singing and then more energetic singing voice model

But I feel like just putting more energy into the vocal sample that you convert will produce a similar effect

[brittle wing]

is it possible to do the opposite? 40k to 48k?

[thin stump]

check it out 👀 http://www.noiseaddicts.com/2009/03/can-you-hear-this-hearing-test/

[brittle wing]

Regarding to ai dataset training, does it actually make a huge difference? also is it possible to mix 40k and 48k datasets together? if it doesn't make a difference?

[thin stump]

If you want do true 48k then you have to remove the 40k audios but if you dont care about perfection then just mix

[knotty moth]

min(40k, 48k) = 40k

[brittle wing]

Thanks

[misty elk]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• UVR5 NO UI for Google Colab, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab

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

[brittle wing]

It's shows it will last till 2 hrs is there any way I could increase?

[odd shale]

Buying Colab Pro /j

Or just making alt Google accounts

[supple inlet]

https://rentry.co/RVC-dataset-RX11
Is RX 11 Elements sufficient to follow the tutorial above?

[odd shale]

You can use either RX11 or 10.

[brittle wing]

11$ thing? Per month? How much it will like I want it for at least 7/8 hrs at least

[odd shale]

Buddy, i did say that as a joke.

[odd shale]

Actually you can just create some alt google accounts or use Hina's Kaggle notebook that got 30 hours of free usage.

[sleek fern]

Can anybody help me? When I enter a call in discord with the modifier, I start giving feedback to all the voices on the call and they all start to suffer the effect

[brittle wing]

Oh okje okie no like I can buy it XD NOT much of deal so

[lethal galleon]

Anyone were knows how to make a speech to speech translation with the original voice? Or using a model

[brittle wing]

can u help me out with the kaggle thing

[odd shale]

I don't remember if there's a kaggle notebook for W-Okada.

[brittle wing]

oh okok

[brittle wing]

so colab it is! i guess need to make alt accounts :0

[odd shale]

Yep, of course.

[brittle wing]

i used the free one today it gave me around just 45 mins

so..... what should i use like cpu or gpu on colab?

[rare goblet]

Ayo? @brittle wing level 4 !!!

[brittle wing]

cuz gpu gave me 45mins

[odd shale]

On colab you should use always GPU.

[brittle wing]

if cpu then? does it affect?

[odd shale]

I mean, on colab you should be using both normally.

[brittle wing]

I am kinda running right now

on this

[mint crag]

Where the heck do I start

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.wtf/

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• Full AI Voice Model Training Guide (local) by Christopher Villanueva
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[odd shale]

Do you wanna use a model or make a new model?

[brave grove]

Hello i have quest

[rare goblet]

Ayo? @brave grove level 2 !!!

[brave grove]

what i need do

i dont have my gpu

[merry eagle]

guys why it's taking to much time

[rare goblet]

Ayo? @merry eagle level 4 !!!

[violet heron]

Open that link

It’s waiting for you to open it

[pastel oak]

Delete "stored_settings.json" and try again

Im more worried about the log that says cuda call failed

[unborn geyser]

where can i find snowie pretrain?

[stiff talon]

Can anyone who uses an AMD GPU help me? I would like to improve the application settings

I left it this way and the voice that already appears in the program remains ok

btw what gpu(dml) i select?

[brittle wing]

I am getting similar errors on other programs that I'm trying to run

[rare goblet]

Ayo? @brittle wing level 1 !!!

[brittle wing]

"No module named 'utils'"

[pastel oak]

If you have AMD gpu you should download the Fork (modified version) of wokada instead, it has better performance

https://rentry.co/ForkVoiceChangerGuide

[stiff talon]

thank you brother, i download it right now and testing!

[rare goblet]

Ayo? @stiff talon level 1 !!!

[violet heron]

#1235952130855010365

[regal temple]

How long would pre processing take on an 11 hour audio clip?

[pastel oak]

Are you trying to train a voice model or make your own pretrain lol

I dont have the answer though

[violet heron]

For inference or training

[regal temple]

This is to train a voice model - I have a 11 hour clip to train off of, but during the pre-process over the course of 9-ish hours only one thread has completed processing

[pastel oak]

11 hours is overkill for training a voice model

I would advise not to try it

[regal temple]

Alrighty - 😅 That makes sense. What is the maximum amount of data you would recommend for a really robust model?

[rare goblet]

Ayo? @regal temple level 1 !!!

[pastel oak]

If the dataset is clean af then you can already get the best results with 15 minutes. If you want to be sure, you could go with 30 minutes. Its more important the dataset includes all kinds of vowel, alphabet and sound pronounciations i guess than quantity

A bit of breathing but not too much is good so it doesnt artifact on breathing sounds. Idk how much of breathing it should have though, just a few included, like seconds

[regal temple]

Brilliant - this is really helpful and gives me a clearer frame of reference for training, thank you!

[hallow coyote]

do u guys know how to fix the voice cutting out on discord?

[wise estuary]

How do i know what epoch the lowest point is?

Sorry for the picture

[pastel oak]

all model files, assuming you enabled the save every frequency thing, has steps and epoch number in the title like testmodel_e600_s47729

so you go to the lowest point, it gives you the steps, you search for that

[zenith shore]

so when i went to the applio website to download and i was doing the compiled version it says download the exe file but like which one-

do u want me to like show the options

[tame mica]

download whichever ya want :p

[zenith shore]

mks

[rare goblet]

Ayo? @zenith shore level 1 !!!

[zenith shore]

so download any of the exe ones?

[tame mica]

yep

[zenith shore]

ty

[unborn geyser]

hey if i didn't found in here snowie?

then can i get similar?

[brittle wing]

I'm trying to find a fix for the "No module named 'utils'" issue I am having

[regal temple]

quick follow up - how many epochs would you run a 30 minute sample for best results?

[odd shale]

You would have to check tensorboard.

You can set the training to probably 500 or 600 epochs.

[pastel oak]

You should use tensorboard to determine when the model is ready
https://docs.aihub.wtf/rvc/resources/epochs--tensorboard/

[regal temple]

Thank you both, looking through this now 🙂

[wise estuary]

I just downloaded the codename rvc fork but i cant intall it. I was able to run "install requirements for RVC ( CUDA118 )" but the other things dont run. Can anyone help me?

[pastel oak]

is codename still in the server?

[analog obsidian]

no

[proven hill]

lasted long time

[analog obsidian]

if u have used non compiled applio before then u need to uninstall torch and reinstall a certain version which i dont remember lol

let me search for it

pip install torch==2.0.1+cu117 torchvision==0.15.2+cu117 torchaudio==2.0.2+cu117 -f https://download.pytorch.org/whl/torch_stable.html

this one

and then u need to run pip install -r rvc-requirements.txt in the folder where requirements.txt is located
edit: if this doesnt work then idk, the only person who can really help u is codename since he made the fork and he's the only one who knows how to fix errors related to it (but he left the server again)

[soft stratus]

I had a problem: Module not found

[pastel oak]

easygui is not working anymore if i remember correctly

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• UVR5 NO UI for Google Colab, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab

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

[soft stratus]

Will it fix

[pastel oak]

Not sure

it probably will but idk when

refer to this announcement

#📰│dev-updates message

[low shard]

easygui google colab is broken

[low shard]

last time rejekts spoke here was in may

the creator of the colab seems not being active since months, a fix would be that you make a cell at the top with !pip install pip==23.1 but the colab is outdated overall and seen theres some other outdated things too soo your choice

[soft stratus]

When will it fix

[low shard]

I just told you a fix and that the creator isn't active since months anymore

[soft stratus]

Ok

[low shard]

i'd personally suggest to use the alternatives but your choice

[soft stratus]

When will there be a new update for it... Can you tell me

[low shard]

I dont think there will be an update for it tbh

[soft stratus]

Ok

[rare goblet]

Ayo? @soft stratus level 10 !!!

[tight ether]

the pre-trained model is loaded only when creating the initial model. after that, it loads the g/d from the last save point of the model you were training, so changing the pre-trained model in the middle doesn't have any effect.

[median monolith]

I have been seeing this guide: https://rentry.co/RVC-dataset-RX11 but there are some things that even after reading it, i still dont understand. like, what is the orange and blue side of this image supposed to represent?

[zenith bear]

anyway to make it sound less robot?

[odd shale]

Using an studio quality dataset, removing harsh esses and any kind of noise.

[thin stump]

you don't have to understand everything. I have the older demographic that told me they liked that kind of stuff because they're audio engineers etc.

it was there in the previous RX10 guide so I kept it in

to sum up the guide

• keep the audio as natural as possible
• least audio processing as possible
• no compression
• fidelity/quality/mic proximity/frequencies/accent consistent

there's soft compression but I trust that people will discover this sooner or later #✨│ai-help message

[zenith shore]

okay so i downloaded the applio exe and it gave me an extracted folder, what do i do now?

[violet heron]

Open it then open the go applio bat

[zenith shore]

mk ty

[brazen skiff]

Hello, does anyone know if the slower training is related to why my dataset is larger? Or nothing to do with it, he has only been training for 1 hour and has barely 100 eponch, and my dataset is

[knotty moth]

yes, it depends on the dataset length and batch size, as well as the GPU performance

[brazen skiff]

And what is the ideal batch size? What change is there in that variable?

[knotty moth]

it may vary, but you can try between 4, 8, 16

[brazen skiff]

And what is the difference? The higher the value, the better? Or the lower the training, the slower it is?

[knotty moth]

the higher value, the faster and the more vram usage, but also the more "stable" (how the graph fluctuates), but higher quality and/or shorter datasets might probably benefit from lower batch size

[brazen skiff]

I understand, thank you very much :))

I have another question, in case I no longer want my model to reach the amount of eponch that I place, can I save one of these pth files, or are these not useful?

[knotty moth]

those G and D files are training snapshot for resuming training later, not for inference
so you can delete all but one of them

[brazen skiff]

It's just that I didn't activate the wardar weight option and I already wanted to stop the training but I see that if I do it it will be of no use to me

[native rain]

Hey i wanna dub a video with multiple characters into english while still using the original characters voices.

where do i even start

[zenith shore]

can anyone help me with my simple problem 😭

[wise estuary]

I need all of these three?

[viral shore]

hey guys is there a software like wokada but for text to speech?

[gray tinsel]

wait high or low pitch for deep/high voices? i completely forgot

[pastel oak]

If you deep and model female -> higher
If you deep and model male -> ~0
If you female and model male -> lower

[gray tinsel]

👌 perfect, 10/10 you said it just right for my circuits thanks

how do i properly test it's working in games? the voice part of audio is not going at all

[storm wadi]

whats the best gui to use rn

like which one has the most features

[ripe grove]

Applio.

[low shard]

that one doesn't have the bypasses, don't use it

If you wanna use rvc mainline, use the Kaggle or Google Colab guide, else you can find some forks guides in https://docs.aihub.wtf like Applio and RVC Disconnected

oh you wanna do it locally?

Whats your GPU?

google colab won't help your local GPU

its different

Google Colab is a Cloud Computing service, you use in remote a Google PC GPU

It means you still have to do it on Cloud, not on your local PC

Also, to check your PC GPU, you need to open task manager via ctrl+shift+esc and then go to performance tab and see the gpu tab name

is there an issue with those ? because you should use those if your PC GPU is bad

dw

tell me the name of your GPU so we can see if its enough

yea

dw, and nah thats not good for training nor inference

[low shard]

Its better you use one of those

[rare goblet]

Ayo? @brittle wing level 6 !!!

[low shard]

you have the google colab subscription for better gpus soo you can use the google colabs

yea the easygui google colab is dead, a fix would be creating a new cell at the top with !pip install pip==23.1, but the problem is i seen theres other outdated things too and the creator isn't active since may

Also, the quality doesn't matter, easygui is a fork of rvc

so its all the same quality

that's not possible, could be related to the default settings being different in 2 different forks

because at the end they are all the same tool, just with differences in the ui and some things added but the quality doesn't change

for training right?

well, its kinda your choice, i personally used rvc disconnected that has no ui but you can choose whatever you want tbh

wdym higher? like make the voice pitch sound higher when used?

oh you mean quality? that all depends on how you clean dataset and train that you can find in our docs https://docs.aihub.wtf/essentials/how-to-make-voice-models/

that shouldn't be possible technically, did you use the tensorboard?

was it 2 different epochs for the 2 forks?

then it could be because of that, as you used 2 different epochs

yw

[scenic arch]

hello guys

is there a way to use rvc with okada remotely? like i want my friend to be able to convert his voice in real time (way too delayed obv) using my hardware and stream on his end if that makes sense

[dusty rune]

help me.

trying to make a sonic model.

[north socket]

Hello, I tried to re-install Mangio-RVC 23.7 on a different drive after running into some issues that I wasn't able to resolve. The error message (same as the previous one from the issue I had) given is attached. I tried asking chatgpt for help and I've uninstalled other versions of pythons along with its modules and made sure that the PATH and environment variables didn't point to the incorrect version of Python (3.12 which i uninstalled, Mangiorvc uses 3.9 in the runtime folder). However, it still gives me this error:

E:\Mangio-RVC-v23.7.0_INFER_TRAIN\Mangio-RVC-v23.7.0>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7897 Found GPU NVIDIA GeForce RTX 3060 Ti Set fp16_run to true in 32k.json Set fp16_run to true in 40k.json Set fp16_run to true in 48k.json Use Language: en_US 2024-08-05 20:01:20 | ERROR | root | Accept failed on a socket socket: <asyncio.TransportSocket fd=5592, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=6, laddr=('0.0.0.0', 7897)> Traceback (most recent call last): File "asyncio\proactor_events.py", line 819, in loop File "asyncio\windows_events.py", line 817, in _poll File "asyncio\windows_events.py", line 566, in finish_accept OSError: [WinError 10014] The system detected an invalid pointer address in attempting to use a pointer argument in a call 2024-08-05 20:01:20 | ERROR | root | Task exception was never retrieved future: <Task finished name='Task-7' coro=<IocpProactor.accept.<locals>.accept_coro() done, defined at asyncio\windows_events.py:568> exception=OSError(14, 'The system detected an invalid pointer address in attempting to use a pointer argument in a call', None, 10014, None)> Traceback (most recent call last): File "asyncio\windows_events.py", line 571, in accept_coro File "asyncio\proactor_events.py", line 819, in loop File "asyncio\windows_events.py", line 817, in _poll File "asyncio\windows_events.py", line 566, in finish_accept OSError: [WinError 10014] The system detected an invalid pointer address in attempting to use a pointer argument in a call

sorry for the huge formatting, i wasn't able to send as a .txt file

Additionally, upon running the .bat file, I was prompted to allow firewall access which I did, and I have also given Python access in firewall settings

[wispy lodge]

Yes, if you have a public IP address or set up a VPN so they can connect (if you're talking about streaming over internet)
https://rentry.co/VoiceChangerGuide#opening-on-multi-pc-setups

Setting up a VPN would be preferable (Radmin VPN as a simple option would work), because in case with public IP, anyone will be able to connect

[scenic arch]

yes, would nordvpn suffice?

[wispy lodge]

Probably yeah. I see they provide Meshnet that allows for end-to-end connections, so it could an option I guess

[north socket]

I've read that mangio is kind of abandoned but I don't think it would affect how the scripts run

[rare goblet]

Ayo? @north socket level 1 !!!

[forest latch]

Does the program need a powerful graphics card to work properly?

[odd shale]

RVC or W-Okada?

[forest latch]

RVC

[odd shale]

Well, depending on what you want to do, you can either download it locally or use Hina's Kaggle mainline notebook or Applio Colab

[forest latch]

I don't know why this happens, but the voice can only be heard after a few seconds.

[odd shale]

Are you talking about W-Okada?

[pastel oak]

Sounds like you downloaded MMVCServerSIO (wokada) i assume

Whats your gpu name, the f0 fet you selected, chunk and extra

[odd shale]

Yes he did

[forest latch]

rmvpe_onnx
chunk 24000 0.5 sec
extra 3840 0.08 sec
1050ti

[pastel oak]

You are using the newest alpha, which we havent benchmarked yet since it can be unstable at times.

1050ti is not the strongest these days, there could be up to 1 second delay

I recommend you download normal wokada instead of using the alpha

https://rentry.co/VoiceChangerGuide

Settings to use are also listed on the guide

[forest latch]

Okay, I'll try to download another version. Is there an uninstaller file to remove the current one?

[rare goblet]

Ayo? @forest latch level 1 !!!

[north socket]

I got the same error when I tried using applio 😦

[pastel oak]

Just delete the folder thats all

[pastel oak]

try putting it on your main C drive

[north socket]

yeah ive got it there

[pastel oak]

Make sure you dont have an antivirus when extracting the zip that may delete files

Ok, not sure then unfort

[north socket]

I would assume i don't need python installed on the C drive already because the environment folder has it

[pastel oak]

if you got the prebuilts it comes with python yes

on the github, go to the Issues section and try to find out if anyone had the issue with the error code or any key factor words

[north socket]

unfortunately i couldn't find anything on the github

[pastel oak]

You can try out Mainline RVC aswell to see if it happens

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

[north socket]

I had the same issue with Mangio but I think the reason why I got it was because I tried installing Python 3.12 and some modules (which didn't install properly). I only had this issue happen after that but I uninstalled the other version of Python completely but it still doesn't work

[pastel oak]

Ye, think you need 3.10.x for rvc

[north socket]

Wait really? The py version in mangio rvc was 3.9

As in inside the runtime folder

[pastel oak]

Anything after 3.10.x (or 3.11.x, idk which one) doesnt work is what I meant

I dont know after which version exactly, but any 3.12 definitely doesnt work afaik

[north socket]

I see

I'm not sure why mine wouldn't work though considering i removed the 3.12 version

[north socket]

It still happens with mainline, as well as the firewall popup on the first time the .bat is run

[rare goblet]

Ayo? @north socket level 2 !!!

[plush compass]

What algorithm do you use to extract uvr drums?

[sturdy stratus]

What is index in realtime voice changer program ?

[pastel oak]

No clue, im not specialized on rvc like that:( would wait on other helpers

[pastel oak]

forces the accent of the voice model (if too high, it can cause autotune)

[north socket]

no problem, thank you for trying to help though

i might just try again after i factory reset my pc, have been wanting to for a while

[sturdy stratus]

Should it be adjusted?

[pastel oak]

Usually no need to use index at all, i never use it

[odd shale]

Demucs or HT Demucs.

Also, you can use any drums model on MVSEP

[wise estuary]

I was told yesterday, to get better models i should try mangio-crepe and before i do something wrong, do i click this on the pitch extraction section, or do i also need to change something other?

[analog obsidian]

who told u that? pitch extraction methods (rmvpe, mangio) barely changes the outcome of the model

all what matters is your dataset

tho rmvpe might sometimes add a metallic sound in small datasets, mangio doesn't do this and can make the model sound a bit more natural but you can only notice this if you inference mangio and not rmvpe

[wise estuary]

Shad said that

Yes he said that with the metallic

And after i retrained the model, without absolute overtraining i also was able to hear that metallic sound

[analog obsidian]

yes rmvpe does this

but also kl values

[odd shale]

That's not really true.

[wise estuary]

So did i do everything the right way?

[analog obsidian]

both mangio and rmvpe are viable
mangio doesnt do the metallic thingy at the cost of being slightly less pitch accurate, and you can only notice this if you inference crepe

rmvpe is more pitch accurate and better for speech

and handles noise better

so its up to you

[wise estuary]

No, i meant the setup. Do i really just need to pick it at the pitch extraction?

[analog obsidian]

it depends, mangio works better for soft and delicate voices because it makes them sound more natural

[wise estuary]

Im just trying, got a fast graphics card

[analog obsidian]

they're both equally good

but like i said, if you give rvc a hq quality dataset, you're gonna get a hq model

no matter if u used rmvpe or mangio

[wise estuary]

Oh also, does the batch size per GPU matter? Or does it just make it train faster?

[analog obsidian]

batch size is how the ai is gonna learn your dataset
lower is gonna add more noise to the samples and the model will be more versatile at the cost of sounding a bit robotic
higher batch sizes sound more closer to the dataset, and are more stable at the cost of versatility

you can use any batch size in any dataset

[wise estuary]

It is, i sat down for 5 hours to get every noise out of the dataset. Even compared all vocal remover to get the best results.

Funny thing is, that i had once a much more worse dataset and it turned out better than my try on getting this voice model

[analog obsidian]

mangio captures more details of your audio than rmvpe i forgot to say that, which leads to slightly better fm values as well

fm handles mostly breaths and Sibilances

kl handles the likeness of the voice and how robotic is gonna sound

[wise estuary]

But my actual question was, if i set it up correctly. I was wondering if this is the only button i need to press to use mangio

[rare goblet]

Ayo? @wise estuary level 5 !!!

[analog obsidian]

yea hop length 64 is also good

u can train that

[limpid reef]

does anyone know what the index does?

[pastel oak]

Forces the accent of the voice model

[grizzled cove]

I'm trying to use UVR5 UI via Colab, but it keeps erroring out and asking for really suspicious permissions to my Google Drive. How do I fix this? The HuggingFace one just errors out when trying to separate audio.

[low shard]

@viscid moss hey check this out

[viscid moss]

That's because you didn't accept Drive permissions

[grizzled cove]

Why does it need permissions to create, edit, delete, download, and read everything in my Google Drive though?

I gave the perms, I really hope it's not gonna try to blackmail me in a month with whatever is on my Google Drive.

[viscid moss]

To get the files you're going to sseparate from your Drive and then save the already separated files back to Drive

[viscid moss]

U can check the code anyways

[low shard]

Oh nvm sorry I didn't see the upper part as I was in the car 😭

[grizzled cove]

Thanks for the answer, it's processing now

[viscid moss]

dw

[grizzled cove]

Bruh

[viscid moss]

wait what

On my colab?

[grizzled cove]

On the UVR_UI.ipynb colab

The HuggingFace Colab also doesn't work.

[viscid moss]

https://huggingface.co/spaces/TheStinger/UVR5_UI
This one?

[grizzled cove]

Yes

[viscid moss]

show me the settings

[grizzled cove]

Wdym "Settings", I use default

[rare goblet]

Ayo? @grizzled cove level 3 !!!

[viscid moss]

lemme test rq

[grizzled cove]

It does pop up the "Looking for GPU" thing, but no "Found GPU" thing

[viscid moss]

[viscid moss]

ya u re right, HF is not providing GPUs

[viscid moss]

And this...

Google is probably blocking my gradio tunnel even when encrypted

for now u can use the NO UI one

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• UVR5 NO UI for Google Colab, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab

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

[pliant hawk]

How should I cover song with Notti Osama voice

Pls help

[grizzled cove]

Is there an ETA for it to be fixed? I'm using the NO UI one like you recommended for now.

[viscid moss]

I don't know why this happens in HF, payments are up to date. Maybe I need to send an email to support.

Colab.. I need to check the code and maybe use another encryption method. But first I have to figure out what exactly the problem is.

So... I'm not sure maybe a few days

[brittle wing]

-rvc

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.wtf/

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• Full AI Voice Model Training Guide (local) by Christopher Villanueva
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[vivid pewter]

why is it doubling the time

[fathom schooner]

idk if i installed pytorch correctley bc i get this error whenever i try to to import torch

it says that the module isnt found

[hazy hedge]

Guys, excuse me, but what is the difference between the latest versions of wokada and fork version? (for Nvida RTX gpu, what should be better to install or no major difference?)

[odd shale]

The Emojikage fork's version has some performance improvements over the OG W-Okada repo.

[hazy hedge]

Ok thanks thats sound good

[waxen sierra]

Sorry, what is UVR5UI?

This is for covers, right?

[odd shale]

UVR5UI is an colab for isolating vocals and cleaning audio. (Also, UVR stands for Ultimate Vocal Remover)

Altho you got MVSEP for that too.

[odd shale]

Nope.

[teal ridge]

Hold the F up, it HAS something to do with breaking the Content Policy!

Anyone know how to bypass this fucking thing?

[odd shale]

Which Colab are you using?

[teal ridge]

I'm trying to upload a model on weights.gg.

[odd shale]

I can't help you, sorry..

I don't know too much about Weights.

[teal ridge]

BeatsBoy couldn't respond to me for one second.

[misty elk]

-rvc

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.wtf/

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• Full AI Voice Model Training Guide (local) by Christopher Villanueva
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[misty elk]

-huggingface

yarr

wait

-hf

[azure marsh]

<:huggingface:1179800228946268270> Hugginface Spaces

• UVR5 UI, by Eddy and Ilaria Huggingface Spaces
• Ilaria RVC Zero, by Ilaria Huggingface Spaces
• RVC⚡ZERO, by r3gm Huggingface Spaces
• Applio, by IA Hispano Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

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

[regal temple]

What is the best pre-train model for screaming and laughing, a lot of the models I have trained struggle wiith higher pitch and laughter, could this be improved with a better pre-train model barring optimizing my dataset?

[odd shale]

If you can't laugh or scream naturally using a model, it's due to Hifigan limitations.

You can use the OG pretrain that is always on RVC.

[regal temple]

Would a dataset that consists of more laughter and high pitch yelling mitigate those hifi-gan limitations?

[analog obsidian]

no

[regal temple]

it is what it is in that regard, i see

[sonic night]

-help

[azure marsh]

Command has expired

[open stag]

• colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, modded by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], credits to Eddy, Hina and Gdr for translating and fixing Google Colab
• Ilaria RVC, by thestingerx Google Colab
• UVR5 NO UI for Google Colab, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab

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

[flint seal]

i'm currently attempting to train a model with a 15 minute dataset, 147 MB, everytime i train off applio on colab it crashes at around 100 epochs. is there any alternative or will this almost always happen because of google colab usage limits?

[thin stump]

-kaggle

[azure marsh]

Suggestions for @flint seal

Kaggle Notebooks

• UVR5 NO UI by Eddy
• How to use RVC Mainline Kaggle by Cauthess
• 🍏 Applio RVC by IA Hispano
• ✨ RVC Mainline by Hina

More commands: -colabs, -hf, -local, -rvc, -help

[thin stump]

applio kaggle or mainline

[flint seal]

thank youuu 🙏

[rare goblet]

Ayo? @flint seal level 4 !!!

[flint seal]

much appreciated

[thin stump]

Oh wait it's gone @low shard

the link for applio

[flint seal]

oh i just realized it is gone

[thin stump]

use mainline, it's the same thing and shares a few similarities with applio

it's a full tutorial too so thats a +

[flint seal]

ok thanks

[flint seal]

ah ok, just saw the tutorial lol ty

[knotty moth]

https://tenor.com/view/sad-bruh-no-why-gif-18375451

[thin stump]

Blaise

[thin stump]

I redacted some images if you reload the page but it wouldn't matter that much

[onyx ferry]

[violet heron]

Banned.

[onyx ferry]

[azure marsh]

💻 (Local) Realtime Voice Changer

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[brittle wing]

Is Applio based off of Mainline?

Or its own thing?

[thin stump]

I'm not sure about the whole lore behind that but it's the base of rvc https://docs.aihub.wtf/rvc/local/mainline/

[analog obsidian]

applio uses the latest pytorch version because is faster
besides that, its just a ui edit of mainline (with added mangio, fcpe and the other hybrid f0 they made)

[violet heron]

Hop length already existed on Mangio
There have been 0 plugins made for it currently
Voice blender already existed
Pretrains were already changeable, you just had to do so manually, the default is still better then their titan anyways
TTS is just edgetts which is also available on Ilaria RVC
Discord Presence is forced and at multiple times before the program refused to launch if discord wasn’t installed, this is now fixed but there is no toggle for it and they rejected a pull request for one

Multiple times they have forgotten to credit other peoples work or declined their feature only for them to add it later as their own (most uncredited people are now credited due to people complaining)

“Simplified installation” their recommended install method is the same as normal RVC because their install has broke multiple times most notably on V3’s launch

A lot of new features applio has were already there, they just removed it before, such as GPU usage for preprocessing iirc

[brittle wing]

Interesting, what’s fcpe?

I’ve seen it but don’t know the difference between the others.

[steep elk]

dataset doesn't have a laughing sound, so I added someone else's laughing sound, but it's so awkward. Are there any tips for making it fit well? lol

😆

[violet heron]

It’s faster then rmvpe and it’s really only good for speech

Rmvpe is still recommended

[brittle wing]

Thanks

[gray tinsel]

after installing Line 1 (Virtual Audio Cable) for o okada, now RVCNvidiaLatest no longer works in discord test and can't hear myself 🤔

[pastel oak]

Can you send a screenshot of how you set up RVC?

Line 1 works on RVCNvidiaLatest prepackage that I put up for me

[pastel oak]

And on discord you selected Line 1 as your input just like you did before with CABLE if you were using that?

Have you also opened up your sound devices and made sure to select your normal microphone and normal headphones as primary device AND communications device?

Maybe also a PC restart could be helpful if your windows is accessing the device, which sometimes causes for it to disable using microphone and/or line 1 on other apps such as voice changer

[gray tinsel]

i'll have to restart later, cannot lose progress on ai work, everything else is set properly

[low shard]

Wait what @nocturne mural applio Kaggle is gone

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.wtf/

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• Full AI Voice Model Training Guide (local) by Christopher Villanueva
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[storm wadi]

can someone help im getting an error with applio ValueError: Invalid value for parameter type: filepath. Please choose from one of: ['file', 'binary', 'bytes']

[rare goblet]

Ayo? @storm wadi level 1 !!!

[brittle wing]

hi guys i need help

[odd shale]

Are you using Applio Local?

[nocturne mural]

I'm going to ask blaise

[low shard]

alr lmk

[timid valve]

is it still worth using mangio crepe over rmvpe for pitch extraction of female voices?

[rare goblet]

Ayo? @timid valve level 6 !!!

[timid valve]

because i've seen people talk about how it's better, supposedly

[analog obsidian]

Mangio is better for softer voices because it softness the audio, mangio is also a bit higher quality than rmvpe (it doesn’t add that metallic effect like rmvpe does)
But is slightly less pitch accurate

And also u have to inference mangio to have the best results

[timid valve]

ooohhh i see

appreciate the much detailed response, thank you 🙏

[pastel oak]

From what I remembered from testing and other peoples words, wasnt rmvpe better for inferencing? The other parts I agree with, especially for realtime mangio crepe seems to be delivering better results

[analog obsidian]

Rmvpe inference is the best because is more pitch accurate, but mangio models sound a bit better if you inference them on mangio

For realtime both works just fine tho for crepe u need mandatory a noise removal app

[quick jungle]

https://cdn.discordapp.com/attachments/1160576561636966491/1165372018414342284/warning.gif

[analog obsidian]

Inferencing mangio on mangio models increases the softness/realistic sound mangio does when pitch extracting

U can try it

[pastel oak]

Ooo yea it makes sense that way, I misunderstood

[analog obsidian]

Yeah if u inference them on rmvpe theyre gonna be more pitch accurate but the softness effect is gonna be a bit less strong

[pastel oak]

tyyy will definitely try around when im back from vacay

[analog obsidian]

np, this only affects mangio models, rmvpe models sound exactly the same no matter if you inference them on mangio or rmvpe

[storm wadi]

yes

[earnest sphinx]

Hi, which tts tool do you use to get the best quality possible? (with rvc models). I use applio and I need to choose between different tts-voices like "en-US-MichelleNeural" for example. The voice is quit good but its monotonous. So I think the problem is the tts-voice model. But I dont know how to chance it and dont know also how to create one, so maybe there are other tools which are easier and also deliver a high quality. Someone has an idea? Thx

[rare goblet]

Ayo? @earnest sphinx level 1 !!!

[analog obsidian]

i personally like elevenlabs tts quality

[earnest sphinx]

for free ofc 😄

[analog obsidian]

oh... i see, sorry i don't know

[earnest sphinx]

thx anyway

[low shard]

Applio TTS is just Edge TTS, for free Bark TTS would be good for emotions but its kinda lower quality

Like using Bark TTS to generate a TTS, then use it as an Input (which is what Applio Does with edge tts api), ofc it won't be as good as gpt so vits bc rvc is for speech to speech

[earnest sphinx]

sry didnt understand the answer. So applio tts should be fine, and then I should use the applio file as input for which tool?

[low shard]

What i meant was that applio tts uses edge tts which is why its not good for emotions, if you want to use emotions, its better you use Bark TTS which has lower quality but better with emotions, use it to generate a TTS audio, then put that as an input in RVC and it could be better

OFc it won't be as good as gpt so vits, as rvc is meant for speech to speech not text to speech

[earnest sphinx]

yeah i know its not tts but for example I noticed differences between the tts tools in general (applio and tortoise). tortoise seems not to be as good as applio. But then I realised, since tortoise allows to chance the voice model (not rvc model), maybe its the better choice for a better result. Maybe I'm wrong and the voice model doesnt chance the result for quality, but I noticed differences when I choose different models (voice model)

[low shard]

applio is using microsoft edge tts, maybe Bark TTS could help as its good for emotions, btw if u want there is also xtts2 that is a forked improved version of tortoise

[earnest sphinx]

ok thx ill check it out

[low shard]

yw

[timid valve]

is there any way i can train a model on even more epochs, even if ive already finished training? e.g. i trained a model on 300 epochs but the optimal amount would be around 400. do i have to train it from the start? or could i just continue training where i left off?

by using the same settings with the only difference being that i set the amount of epochs higher?

[odd shale]

Yep, you can start where you left off.

[odd shale]

Yep.

[timid valve]

oh cool!

thank you, i will do that then

[odd shale]

But make sure to erase everything from the dataset path

I mean, on the GUI

[pastel oak]

Use the G and D paths where the "pretrain" would be placed in to continue training

[timid valve]

im not sure what you mean

[odd shale]

I mean, on the training folder path there will be the path to your dataset folder.

[timid valve]

and i have to empty that patch? why's that?

[odd shale]

Yep, you can now empty the path.

The dataset is not necessary if you already got the D and G files you'll use to continue training

[timid valve]

oh okay

thanks

[odd shale]

You're welcome.

[humble pelican]

yo can i send a ss per chance?

[pastel oak]

Need lvl 1 on the server for image perms

[humble pelican]

Oh damn 😭

Gonna have to yap it up to get some help I guess

[pastel oak]

Few messages away

[humble pelican]

Interesting

[pastel oak]

very!

[humble pelican]

Indubitably

[rare goblet]

Ayo? @humble pelican level 1 !!!

[humble pelican]

YAY

this is for Colab CoverGen_NO_UI_v2

[misty elk]

-kaggle

[azure marsh]

Kaggle Notebooks

• UVR5 NO UI by Eddy
• How to use RVC Mainline Kaggle by Cauthess
• 🍏 Applio RVC by IA Hispano
• ✨ RVC Mainline by Hina

More commands: -colabs, -hf, -local, -rvc, -help

[humble pelican]

GUYS HELP ME

[low shard]

Be sure you are using the T4 daily limited gpu

If you got no avaible GPU, you can use an alt google ac

[misty elk]

eljoy aint working

i tried to reload it

and wait

but it still aint showing

fuck do i do

[violet heron]

What do you need help with

Nvm

[violet heron]

Make sure your connected to a GPU runtime

[patent pasture]

Hadn't Applio been fixed yet?

[golden veldt]

I have a local web UI voice training program. How can I open TensorBoard? How can I set it up to automatically stop training to prevent overfitting?

[odd shale]

I think you can't set it to automatically stop.

[golden veldt]

Ok. I need to open TensorBoard to check if the model is overfitting during sound training. How can I do that?

[odd shale]

You can check tensorboard's g/total graph.

on your RVC installation there should be a .bat file titled "tensorboard" that you must click.

I'm not sure.

[golden veldt]

bat and python files related to tensorboard close immediately as soon as they are opened

[rare goblet]

Ayo? @golden veldt level 1 !!!

[slim wadi]

so i have a 45 minute audio set of clear vocals with no backround, what is the next step exactly? id like to use the web trainer as im on my laptop rn

[odd shale]

Your 45 mins are already split on segments and on a folder?

-rvc

[azure marsh]

Documentation

AI HUB Docs

https://docs.aihub.wtf/

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• Full AI Voice Model Training Guide (local) by Christopher Villanueva
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[odd shale]

Also, read the docs too.

[slim wadi]

trying to understand the docs currently lol, wdym by split on segments?

[odd shale]

For example, splitting your 45 min file on 10 or 5 secs clips.

[torn seal]

Yo

[young trout]

guys how can i transfer ai voice to discord

[open berry]

you need an app to create a new mic

guys how i make my voice stopping cut?

idk what conf can help on that

[odd shale]

You both can go to the #🔍│help-w-okada channel.

Shad will help you both.

[pearl bramble]

How to train an ai model free online apart from weights

[thin stump]

Whenver you stop the training or stop the cell from running, you need a fresh new link for the Imjoy to work

[thin stump]

-kaggle

[azure marsh]

Suggestions for @pearl bramble

Kaggle Notebooks

• UVR5 NO UI by Eddy
• How to use RVC Mainline Kaggle by Cauthess
• 🍏 Applio RVC by IA Hispano
• ✨ RVC Mainline by Hina

More commands: -colabs, -hf, -local, -rvc, -help

[pearl bramble]

I'm still confused with how to train an ai model online

[thin stump]

I provided a lot of images like a picture book for the mainline kaggle. You might want to take a look at rvc disconnected instead https://docs.aihub.wtf/rvc/cloud/rvc-disconnected/#rvc-disconnected

[pearl bramble]

with the google colabs stuff it disconnects after a while

like when training

[thin stump]

Yeah thats why I recommended kaggle. It took me 16 minutes to setup and probably will take you longer since you're new

[pearl bramble]

does it work on laptop like for windows

[thin stump]

and the guide because I forget steps all the time. It works on the cloud so you don't need a GPU or anything

[pearl bramble]

alright i will read through it and see

[vivid pewter]

inference

-inference

[violet heron]

-local

[azure marsh]

Suggestions for @vivid pewter

🖥️ Local stuff

• 🍏 Applio, by IA Hispano GitHub
• Mangio-RVC-Fork, by Mangio621 Huggingface
• RVC Studio, by SayanoAI Huggingface
• AICoverGen, by SociallyIneptWeeb GitHub
• Replay, by Replay Team Website
• Original RVC, by the RVC-Project team GitHub
• GPT-SoVITS, by RVC-Boss GitHub

Credits to Faze Masta and Antasma for compiling these links.

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

[violet heron]

-spaces

[azure marsh]

<:huggingface:1179800228946268270> Hugginface Spaces

• UVR5 UI, by Eddy and Ilaria Huggingface Spaces
• Ilaria RVC Zero, by Ilaria Huggingface Spaces
• RVC⚡ZERO, by r3gm Huggingface Spaces
• Applio, by IA Hispano Huggingface Spaces
• AICoverGen, by r3gm Huggingface Spaces
• Advanced RVC Inference, by r3gm Huggingface Spaces
• RVC v2 Huggingface version, by Clebersla Huggingface Spaces

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

[brittle wing]

I ont have the lv to send my problem ahhh

[rare goblet]

Ayo? @brittle wing level 1 !!!

[brittle wing]

NVM NOW I DO

YEY

huh.. nvm the ai convert is working now too

my god Im having such a good day