✨│ai-help

[finite galleon]

Dang

Bro is dedicated

[pastel oak]

Or well, yeah you could delete the 48k ones if you dont plan on but no need to really

[pastel oak]

I am even more annoyed that I couldnt find the actual problem to the answer. All I found is some guy from the server absolutely insulting the shit out of the one that experienced that issue 😭

[river sedge]

damn wtf

[rare goblet]

Ayo? @river sedge level 1 !!!

[river sedge]

ayoo

[river sedge]

Could it have to do smth with my dataset?

[pastel oak]

As I said, it could if you have many many files and they vary too much in file size

[river sedge]

Hmm, I have abt 70 files with a max length of 10 seconds

[river sedge]

So, I can try deleting some?

[pastel oak]

Put everything into one file

I am going to assume you watched a youtube tutorial that told you to split audios into 10 second files

dont do dat

just keep it as one file, see if it works

[river sedge]

they had an instant dataset maker lol, I'll try to keep it as one file

thanks

[tough fiber]

no i mean just dont want traing this data anymore ...

ill train new data so

is it need or no

[pastel oak]

Oh no no, the thing youre looking at is NOT your voice model

those are pretrains. You need those to train any voice model

If you want to delete your entire voice model the 230 epochs, delete the folder ixtivhmale in "logs" folder

[river sedge]

should the audio file be flac or wav? Just to be sure

[tough fiber]

oh okay

[pastel oak]

both are lossless so doesnt matter

[river sedge]

perfect, thanks

[tough fiber]

and is it good i have mp3 file data so if i rename this file mp3 to .wav is it work

[amber zephyr]

hey is it ok if use OV2 on a 6:30 dataset

[zinc anchor]

Locked in for the bag

He'll understand

[finite galleon]

[pastel oak]

Funnily enough, ever since that update, I changed my style into fewer but longer messages, so I am doing the exact opposite

[pastel oak]

no that doesnt work

[zinc anchor]

Yeah

If it's lossy it's lossy

No way around that

[tough fiber]

so how can i change

[pastel oak]

Convert

[tough fiber]

online convert websites is enough or

[amber zephyr]

https://www.freeconvert.com/ogg-to-wav/download

[pastel oak]

I dont know if theres any point in that because if its already mp3 theres not much to do right..?

[tough fiber]

should i change with a program

[pastel oak]

Probably if you dont want to download program

[amber zephyr]

anybody?

[tough fiber]

i have audacity already lol

but i dont know how to use

i think i did somehow -*-

well nice my data 507mb

[pastel oak]

Yes, wav files are very big

use FLAC instead

if its a problem for you

[odd shale]

-paperspace

[azure marsh]

RVC v2 Tutorial for Paperspace

💡 Click on the button below to open the guide.

Credits: LollenApe

[tough fiber]

so shadicti

[rare goblet]

Ayo? @tough fiber level 7 !!!

[tough fiber]

its stereo and 48hz

so is it okay?

i hope i dont get any error -.-

[split frigate]

which should i pick, in tensor board

g/total or g/mel or g/kl

i want pick lowest point, from which 1 i pick?

[tough fiber]

write on search bar g/t and check

[split frigate]

what about mel and kl

[tough fiber]

well i only check g/t

[split frigate]

some1 advise me to follow them alongside with /t

[tough fiber]

if its increasing its overtraining so

[split frigate]

idk anything about The Hz im always choose 48hz

kinda hard to understand

[zinc anchor]

@red kayak

[tough fiber]

he said g/t

at least for me

[pastel oak]

Dont know

[zinc anchor]

Yeah they show you if there's mode collapse

g and d total

[split frigate]

@proper shale advised me about /mel and /kl

[zinc anchor]

Good advice, I pinged Litsa because he can explain it

[pastel oak]

From my knowledge:

The easy, non complicated answer: Train until you are sure you hit the lowest point in g/total, then take the voice model before the lowest point in steps.

The more advanced answer: Even if g/total hit's the lowest point, monitor the graphs d/total ; mel and kl. If those are still sinking, you may keep training

[split frigate]

the only 1 i want to know is
if /mel and /kl going down and g/total going up

how can i pick lowest point

[pastel oak]

As I said, even if g total going up but mel and kl going down you can take anything after, you DONT HAVE TO pick the lowest point of g total if those are going down

most recent/newest one is probably ok

[zinc anchor]

Understandable

[split frigate]

i get it now, u mean if mel and kl going down just take newest point in g/total

[red kayak]

What do u need

[red kayak]

That simply isn't true

[split frigate]

what should i do?

[rare goblet]

Ayo? @split frigate level 7 !!!

[red kayak]

What shad is saying isn't entirely true however over training isn't ideal

If ur generators loss starts to increase that means that the model is slowly getting worse and worse at generalizing the data given

Even if Mel and kl are improving (Mel = clarity btw) that still doesn't enable he model to perform at its best cuz it's slowly starting to over exaggerate. The generators outputs start getting worse and worse the discriminator keeps on telling fake samples apart

[tranquil cliff]

hi litsa

[red kayak]

Hi :>

[pastel oak]

Good thing I disclaimered indirectly that my answer may not be the right one

[red kayak]

It's fine

[split frigate]

my EN is bad can u give me like short and clear answer of what should i do?

sry

[pastel oak]

we learn

[red kayak]

U can in fact over train for a little bit to get that extra bit of clarity but still it isn't advisable to do so unless u actually have a good dataset with no noise

[red kayak]

Just look at g/total and check Mel and kl from time to time

Ur main focus should be g total though

[split frigate]

and pick lowest pont?

[red kayak]

Yes set smoothing to 0.987 and ignore outliners on

[tough fiber]

literally ill punch my pc im so close

[split frigate]

g/mel and g/kl is hard to understand but i want master rvc and make completion model

[split frigate]

already did that

[red kayak]

B4 u master RVC u should learn how to make datasets and how to choose data

[split frigate]

still didnt get what should i do if mel and kl going down and total going up

in my case i will just ignore thats mel and kl going down and pick lowest point in g total

[red kayak]

Which all takes experience

[red kayak]

Yeah do that

[split frigate]

my datasets about 10min and clear as a broadcast

and i clear it more from audcity

[red kayak]

Podcasts are never really that clean

[split frigate]

so its like nothing clearer i cant get and some of it over 10min

[split frigate]

nothing clear as they imo

[tight ether]

[red kayak]

There's software u can use for audio repair

[split frigate]

which 1 ?

[red kayak]

Uvr and iZotope rx10

[split frigate]

i want get best thing i can

[split frigate]

tysm

[red kayak]

Good luck

[split frigate]

use both of them?

or u suggest one of them

[red kayak]

Both

[tough fiber]

.paperspace

what was the command

-paperspace

[azure marsh]

RVC v2 Tutorial for Paperspace

💡 Click on the button below to open the guide.

Credits: LollenApe

[proper shale]

u don't necessarily need to pick the lowest point, as i said b4

but you can pick the lowest mel/kl points

[split frigate]

are mel and kl equl in choosing lowest point?

[pastel oak]

You only ever choose the lowest point of g/total

[split frigate]

okay.. tysm

[brittle wing]

what do i do about this?

[glass drum]

for local running is using Mainline more recommended than Ilaria RVC?

[zinc anchor]

For training yes, because only Mainline does it out of those two

For inference Ilaria RVC is better

[glass drum]

I'm looking to transform text to speech or songs to a voice of a model so not training I think

is that what inference is ?

[zinc anchor]

Yes

You need a vocal track as the input

[glass drum]

cool, i'll go for that then. thanks

[fervent cargo]

can i manually change the language that rvc is using?

[glass drum]

but after cloning ilar rvc 3 in the readme it says to clone https://huggingface.co/spaces/mateuseap/magic-vocals/ but it doesnt exist
nvm i'm dumb

[rare goblet]

Ayo? @glass drum level 2 !!!

[brittle wing]

why is my command promt saying this?

AttributeError: 'RVC' object has no attribute 'tgt_sr'

and how do i fix it?

[pastel oak]

is it RVC gui realtime?

[brittle wing]

yeah

[pastel oak]

Did you download the CODE or the binary release?

[brittle wing]

idk

[pastel oak]

whats the folder called

[brittle wing]

its called

RVC1006Nvidia

[pastel oak]

Ok send a screenshot of your RVC

[brittle wing]

alr

[rare goblet]

Ayo? @brittle wing level 3 !!!

[glass drum]

does Ilaria RVC not work on windows?

[brittle wing]

caould it be because of my gpu?

[pastel oak]

Change your OUTPUT device to SAMSUNG (MME)

Also, the default index file causes issues, even if youre not using it. So load up any other index, it doesnt matter if its the same as your voice model, just any custom one you downloaded

[brittle wing]

there is no samsung (mme)

[pastel oak]

There should be one called SAMSUNG (Nvidia high definition audio) (MME)

it should be somewhere rather on the top

[brittle wing]

theres 2 and they are these

[pastel oak]

Is that even your output? Your headphones?

[brittle wing]

im not using headphones

[pastel oak]

Your speakers, same thing

[brittle wing]

yes thats my output

[brittle wing]

oh

[pastel oak]

Ok, then change your Input to the one with Windows DirectSound

[brittle wing]

alr

this is the list btw

[pastel oak]

However, from experience, anything other than MME causes issues but maybe youre lucky

[brittle wing]

so im using this model

https://discord.com/channels/1159260121998827560/1214061831866552380

[pastel oak]

I am not familiar with OV2 models but if its RVC it will work just fine

Select it, and select its index also

[brittle wing]

thats what i did before

[pastel oak]

After you have the same type of driver for input and output, try voice changer again

[brittle wing]

and it says that error

[proud elbow]

OV2 is pretrain for rvc models

[pastel oak]

Ah ok gotcha

[brittle wing]

so i press this right

[pastel oak]

Ye

[brittle wing]

alr

[pastel oak]

can you just not use input: microphone MME and output: line 1 MME?

since youll be running this as a voice changer anyway

[brittle wing]

ait what

this is what im using rn

[pastel oak]

Thats fine, output should also be something with MME for example the Line 1

[brittle wing]

theres is line1

but thats voicemod

[pastel oak]

Isnt it just Line 1 Virtual Audio Changer but you changed the name to Voicemod?

If not then ig download the real VAC Line 1

https://software.muzychenko.net/freeware/vac470lite.zip

run "setup64.exe"

[brittle wing]

it is installed like that

[pastel oak]

ah ok

[brittle wing]

alr

how do i set it to my microphone?

wait

is this how it supposed to be

[pastel oak]

thats the correct one, but select the (MME)

[brittle wing]

for output or input

[pastel oak]

output

[brittle wing]

there is none

OH

IM STUPID

i thought that was all of them

theres still a bunch

[pastel oak]

You didnt scroll up did you

[brittle wing]

i didnt lol

[pastel oak]

😂

Even told you its on the top 😦

[brittle wing]

i just noticed it 🤣

[brittle wing]

oh...

welll what do i do now

[pastel oak]

Now it should work

[brittle wing]

alr

[proven hill]

it works but you have to do some complicated things

[brittle wing]

still says this

[glass drum]

wsl I imagine right?

[proven hill]

i suggest you using ilaria rvc mainline directly but its currently on hold

[proven hill]

nah

[pastel oak]

Hmmm

[brittle wing]

what do i do

[pastel oak]

1 sec

[brittle wing]

alr

[pastel oak]

Open the folder "assets"

is there a "hubert" folder in there?

[brittle wing]

yes

[pastel oak]

and are there files in the hubert folder

hubert_base.pt
hubert_inputs.pth
.gitignore

[brittle wing]

yes

[pastel oak]

then screenshot the entire error code

[brittle wing]

what file

oh

[pastel oak]

was asking if these 3 files are in the hubert folder or not

[brittle wing]

there is

[brittle wing]

in the command promt or

[pastel oak]

ye

[brittle wing]

qlr

[pastel oak]

Wait did you extract the rvc1006nvidia folder out of the zip?

[brittle wing]

yeah?

[rare goblet]

Ayo? @brittle wing level 4 !!!

[pastel oak]

Confused why the folder path is listed twice lol

[brittle wing]

oh

lol

[pastel oak]

what gpu do you have?

[brittle wing]

NVIDIA GeForce GTX 770 :b

it does warn me of that

[pastel oak]

Quite honestly not sure if thats even going to be able to run it

[brittle wing]

wait

what

[pastel oak]

Its a weak gpu

[brittle wing]

is it?

[pastel oak]

its very outdated

[brittle wing]

yeah

[pastel oak]

it only has 2gb vram aswell

[brittle wing]

oh

[pastel oak]

But now I cant really tell if it wont start because of that or if there are any files missing

[proven hill]

THE 770

[brittle wing]

yeah...

[proven hill]

babe you will not be able to even make it run even if you pray

[brittle wing]

AAAA

NOOOOOOOOOOO

[proven hill]

use the colab

[brittle wing]

didnt that get deleted

[proven hill]

nope

[pastel oak]

Hes trying to run the RVC realtime. Dont think theres a colab for that, aside from wokada

[proven hill]

ohhhhhhhhhhh its realtime

[brittle wing]

what is this one

[proven hill]

but we are in help rvc

[pastel oak]

Let's backtrack quickly

https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/releases/tag/updated1006v2

Did you download your rvc from this?

[proven hill]

its for the normal stuff

[brittle wing]

no

[pastel oak]

It still technically counts as RVC, as it is hosted by RVC

[proven hill]

lame, there should be a voice changer help channel instead of okada

okada runs worse but its more update and have easier gui

[pastel oak]

Same. I was told there was one, but people still asked in this channel instead of clear help VOICE CHANGER 💀

so thats why it got changed to okada

[proven hill]

people are stupid

[pastel oak]

was that no towards me? where did you download it from then

[brittle wing]

uhh let me find the link rq

[glass drum]

Elevenlabs plugin on Applio can't use custom voices? (the ones you get from voice library)

[brittle wing]

https://huggingface.co/lj1995/VoiceConversionWebUI/blob/main/RVC1006Nvidia.7z

here

[proven hill]

yea its the same

[pastel oak]

Well yeah thats linked on the link i sent

[brittle wing]

oh

[frigid garnet]

i need the hxh narrator fr

[pastel oak]

Then I dont know why that issue is happening. It is caused by pre-trained models like hubert not being downloaded, but youre claiming it is so no clue.

[brittle wing]

idk

[proud elbow]

try redownload the prezip?

[brittle wing]

is there any other voice changers that i can run that use rvc?

[pastel oak]

theres wokada but your gpu is not strong enough

[zinc anchor]

#1159289738314919936 is your best bet

Free or paid

[brittle wing]

oh

[pastel oak]

Is your internet good? theres a colab version

[brittle wing]

im using ethernet

wso idk

[pastel oak]

google speedtest and lmk your download upload

[brittle wing]

k

[heady talon]

does it make sense to train more than 300 epochs? I decided to train 1000 but I don't think there will be a better model

[pastel oak]

We use tensorboard to determine overtraining or not

[pastel oak]

Nice thats good

1 sec

[brittle wing]

alr

[proud elbow]

try target 500 then observe the tensorboard

[pastel oak]

#🔍│help-w-okada message

Check out these links, follow the steps

[brittle wing]

alr

[west summit]

i am getting error

module not found eror

[pastel oak]

Copy the entire error

[west summit]

JSONDecodeError Traceback (most recent call last)
<ipython-input-6-825849d8c545> in <cell line: 31>()
31 if os.path.exists(config_path):
32 # File exists, proceed with creation of creds and client
---> 33 creds = Credentials.from_service_account_file(config_path, scopes=scope)
34 client = gspread.authorize(creds)
35 else:

5 frames
/usr/lib/python3.10/json/decoder.py in raw_decode(self, s, idx)
353 obj, end = self.scan_once(s, idx)
354 except StopIteration as err:
--> 355 raise JSONDecodeError("Expecting value", s, err.value) from None
356 return obj, end

JSONDecodeError: Expecting value: line 1 column 1 (char 0)

[proud elbow]

more likely outdated colab

[west summit]

what should i do now?

[rare goblet]

Ayo? @west summit level 1 !!!

[west summit]

from where i get the latest?

[brittle wing]

do i use this to add the vox rvc model

[proud elbow]

-colab

[azure marsh]

Suggestions for @west summit

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[pastel oak]

^ check these messages above. You can use Ilaria for inference (ai covers), or RVC Disconnected to train voice models

[ornate wasp]

Anyone use tensordock? Im trying to run AiCoverGen or any other collab on Tensordock through jupyter notebook and its not working. Every collab notebook ive tried on there has run into issues an im tired of burning my money on there lol. Has anyone got it to work?

Probably just gonna call it a wash and switch to Paperspace next payday

Was also considering Collab Pro but Collab pro is such a rip off compared to the other cloud gpu providers

[light root]

Why is ai gen taking forever for a 10 second clip?

[brittle wing]

how do i use this

[proud elbow]

the installation step?

[ornate wasp]

use what?

[west summit]

thanks man

[meager iron]

i just installed so-vits-svc, but i get PermissionError: [Errno 13] Permission denied: error everytime i start, any help?

or should i ask this somewhere else

[light root]

The processing after inputting my audio file

[ornate wasp]

My advice would be to not use So-Vits, its really outdated and most models are made for RVC not So-Vits

[rare goblet]

Ayo? @ornate wasp level 4 !!!

[light root]

It's taking 20-30 minutes for the final output

[ornate wasp]

what are your hardware specs?

[proud elbow]

prob he meant ai covergen colab

[meager iron]

i see, do u know where to get rvc?

[ornate wasp]

yea google RVC1006Nvidia

then click the first huggingspace link

or you can look at the guides for other links

but RVC1006Nvidia is the one that works best for me

just download, unzip and run go_web bat

[meager iron]

same GUI as so-vits or nah?

[rare goblet]

Ayo? @meager iron level 1 !!!

[digital summit]

where do I put the CKPT files in the voice changer?

[ornate wasp]

no clue if its the exact same gui, I haven't touched sovits since last summer. But the gui is super simple and easy. I'm an idiot and I can use it no problem

what so-vits gui do you use?

[odd shale]

W-Okada?

On the w-okada gui, press the "edit" button and add the .pth file of your model.

[pastel oak]

You dont use ckpt files on the voice changer

[proud elbow]

.ckpt? it doesnt support gpt sovits models

[meager iron]

alr thanks i will try it

i just need to run the go_web then, no prerequisites?

[odd shale]

Yeah, it doesn't support GPT sovits models.

[digital summit]

I do add just that but when I click upload it's just blank

[pastel oak]

Like the others said, gpt sovits dont work

[digital summit]

so the voices without index files just don't work?

[pastel oak]

They do, if theyre RVC .pth files

[ornate wasp]

yea thats all I had to do

[ornate wasp]

if you run into any issues feel free to ping me

[glass drum]

I can't change my voice model on applio.
no matter what I do (unload, refresh, restart) it only uses the same voice model that I used at first

[odd shale]

Upload a screenshot there.

[digital summit]

how do i get it to work then 🥲 i am big dumb

[rare goblet]

Ayo? @digital summit level 1 !!!

[digital summit]

of what?

[odd shale]

Of your W-Okada GUI.

[ornate wasp]

even when you close down applio completly and reopen it?

[glass drum]

yes

[odd shale]

Do you have an RVC model downloaded?

[digital summit]

yeah

i tried butters from south park, the hand unit and kurzgesagt and just uploading the path file but they all just appear blank

not sure if i should be doing something else

[ornate wasp]

my advice would be to just use regular rvc. these cool fancy versions with all the crazy features are nice but they always have issues for me

[brittle wing]

how do i ge this to work?

[pastel oak]

INPUT - Your Microphone
OUTPUT - Virtual Audio Cable this is mandatory so you can actually use the voice changer on other apps.
MONITOR - Headphones if you want to have live monitoring, this will throw most people off as there is a delay, but its good for the initial test.

Virtual cable download:
https://software.muzychenko.net/freeware/vac470lite.zip

"setup64.exe"

[meager iron]

wait, this work with AMD?

[ornate wasp]

no you have to use the other version hold on

[ornate wasp]

google RVC1006AMD_Intel

[ornate wasp]

everything else will be the same, just use that version as it was meant for amd

[meager iron]

i see

thankfully havent downloaded anything yet i was still researchin

[safe wind]

hey i need help, i downlaoded phyton. but i still can't open the voice changer

[ornate wasp]

yea give that one a go, what gpu do you have?

[meager iron]

i got amd gpu

[pastel oak]

https://huggingface.co/lj1995/VoiceConversionWebUI/resolve/main/RVC1006AMD_Intel.7z

[ornate wasp]

i meant what amd gpu lol

[ornate wasp]

this is it, was not sure if we were allowed to send links so thanks for that

[meager iron]

RX 580 8GB

[pastel oak]

if its github hugginface etc links youre fine to send them

[ornate wasp]

one of the classics! But yea that link shad sent is the same thing I was trying to have you get. Thatll work for ya

[pastel oak]

should be a trustworthy page ya know

[ornate wasp]

Good to know, I appreciate the info.

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

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

🇪🇸 Spain

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

🇫🇷 France

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

🇩🇪 Germany

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

🇯🇵 Japan

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

🇰🇷 Korea

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

🇵🇱 Poland

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

🇷🇺 Russia

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

🇮🇹 Italy

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

[meager iron]

same as before? with what file should i start it with?

[rare goblet]

Ayo? @meager iron level 2 !!!

[ornate wasp]

should be the exact same. just unzip the zip and then run go_web bat

[pastel oak]

go-web-dml.bat

[ornate wasp]

i assume you know where to put the index and path

[pastel oak]

for AMD, you always look for anything with DML at the end lol

[ornate wasp]

listen to shad he knows his stuff

[pastel oak]

I sense the power of a big Helper inside you, so I shall let you do your thing and correct wherever needed prayge

[meager iron]

alright thanks

it works now

still figuring it

[ornate wasp]

aw thank you 🙂

[ornate wasp]

awesome! glad its working. no errors or anything when you try to convert a file?

[meager iron]

how do i do that 😭

i wanna make like ai song covers

ive done it before but not with rvc

[ornate wasp]

oh sorry just saw this, so do you already have the model and the song you want to convert?

[rare goblet]

Ayo? @ornate wasp level 5 !!!

[meager iron]

yea man

but like i have AMD so theres not much options

[ornate wasp]

okay cool, so do you mind sending me a screenshot then

so that way I can see what options you do have

[meager iron]

what kind of screenshot?

[ornate wasp]

a screenshot of your screen. I want to see what options your webui has

its probably the same as the one I have but i would like to make sure before I give you the wrong advice

[meager iron]

hmm ok holdon

[ornate wasp]

sure

[meager iron]

is like dis

[ornate wasp]

great thanks

[ornate wasp]

so thats the realtime coversion gui. Its used for changing your voice in realtime, not for making song covers

[meager iron]

yeah ive noticed

i was searching for the song covers but it seems like theres none for amd cards (?)

not sure

[ornate wasp]

hold on a sec ill see

[meager iron]

aight

[rare goblet]

Ayo? @meager iron level 3 !!!

[full turret]

--colab

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[split frigate]

Any1 know how to use UVR ?

for clean vocals and model voice

[ornate wasp]

okay so try opening a command prompt in the rvc folder and then run the follwing command

runtime\python.exe infer-web.py --pycmd runtime\python.exe --dml

no clue if this will work but its worth a shot

[pastel oak]

https://docs.google.com/document/d/17fjNvJzj8ZGSer7c7OFe_CNfUKbAxEh_OBv94ZdRG5c/edit?tab=t.0#heading=h.7bakw3ajb3ii

good luck xd

[meager iron]

ok imma try

[pastel oak]

section " for vocals"

[split frigate]

i cant understand 1 thing

[pastel oak]

go to whatever you need on the left side

[split frigate]

i ment this

[meager iron]

it opens this

[pastel oak]

?

[split frigate]

how can i fix it now i havent see any solution

[pastel oak]

Cant help

just use a single process method instead of ensemble

[split frigate]

@proper shale
can you share me best settings u got for clear vocals?

i cannot understand this

[ornate wasp]

YES! thats it!

[split frigate]

idk anything litreally

literally

[pastel oak]

It literally shows the models you should use

[surreal bridge]

I am trying to use voices on the song "we didnt start the fire" but get bad results because on some parts of the song there is a high pitched and low pitched voice overlapping, which returns bad results. Is there any way to fix this, or is it not possible

[pastel oak]

MDX23C-InstVoc HQ if you want to separate a song instrumental - vocals

[meager iron]

but it seems like i have trouble changing the voice tho im looking for some ways

(ive downloaded some new voices)

[pastel oak]

Kim Vocal for blabla this that

[pastel oak]

https://rentry.co/ExtractGuide

else look at this

[ornate wasp]

did you put the pth file for the model in \RVC1006AMD_Intel1\assets\weights ?

[meager iron]

ok i just refreshed it and its there

imma try do convertin

[ornate wasp]

go for it! let me know how it goes. im rooting for you!

[split frigate]

my mind about to blow up

[pastel oak]

bro on the top left

CHOOSE PROCESS METHOD
ENSEMBLE METHOD

change that to MDX or VR

depending on what you need

And then select ONE model, run it

Then when youre done, select that new file again, and apply the rest of the models one by one

[split frigate]

what about settings ?

[pastel oak]

First, on the bottom left "choose vr model" select the model

[odd shale]

Nice job, i didn't know you knew about UVR usage.

[pastel oak]

Settings you can most likely leave default

[pastel oak]

bs-roformer go brrrrrrr

[split frigate]

what if i havent find MDX23C-InstVoc HQ

[pastel oak]

Its on CHOOSE PROCESS METHOD: MDX

You selected VR-Architecture

they are categorized

[split frigate]

sry about that mate

[zinc anchor]

If you don't find it there, you might need to download it

[split frigate]

YES I LOVE ERRORS

[pastel oak]

Lol no I am using CAPS LOCK to make it a bit easier, maybe it makes it easier to understand what to look for. I am not trying to sound angry if you think it sounds like taht 😅

[zinc anchor]

Do you have a Nvidia GPU?

[split frigate]

i do

[zinc anchor]

Oh

[pastel oak]

Which GPU

[zinc anchor]

I get that error because mine can't do GPU Conversion

[split frigate]

it isnt working on it?

[pastel oak]

which gpu

[split frigate]

i hope 3060 is the correct answer

[pastel oak]

Hm you should be able to run it just fine

No idea

[meager iron]

IT WORKSS THANKSSS

but it still needs a lot of tweaking but its fine i can experiment 💯

[proper shale]

choose the process method, dont enable the ensemble algorithm stuff unless u wanna ensemble stuff ig

ensemble is to use two voice sep stuff at once

so

yeah

anyways, id say, use mdx-net stuff

or try bsroformer

[pastel oak]

Thats an old screenshot

check the latest

[split frigate]

@pastel oak explained to me untill this 1

[proper shale]

select saved settings rq

[ornate wasp]

Glad I could help man! Make sure to pay it forward, if you see someone having the same issue that you did then teach them what I taught you. And if you need any more help feel free to ping me

[split frigate]

rq ?

[proper shale]

real quick

[split frigate]

same error if in saved settings

[proper shale]

wait did u even select the output thing

folder

[split frigate]

yes i named the folder
UVR ouput

[proper shale]

id recommend renaming that so it has no spaces and then u try again

or just export it to ur c: drive instead.....

[split frigate]

same error

[split frigate]

same erorr

this is annoying

[meager iron]

for sureee

btw, i keep getting this error

[split frigate]

i will try change the file it self'

[ornate wasp]

Weird I've never gotten that on local RVC. Does it happen every time you try to convert something?

[meager iron]

first time no, but the second ones yeah

[ornate wasp]

Is the command prompt still running?

Closing the command prompt also closes RVC so if you closed it that'd cause it

[meager iron]

after the error, yes

gave something like this before the error

[ornate wasp]

Try running that command I sent to open RVC again and see if it fixes it

[meager iron]

so run the command again after error?

[ornate wasp]

Oh also, you might want to copy that command I sent earlier to the notepad, and then save it as a bat file in the same folder. That way you can just run it from the bat instead of putting in the command each time

[ornate wasp]

Yeah just run the command again after you close out the first command prompt. Then test it to see if it gives another error

[meager iron]

still error

didnt close the console tho

or am i supposed to not touch anything

[split frigate]

@proper shale@pastel oak
its worked after i change audio file it self
tysm

[ornate wasp]

Nope you don't have to close the console. The consoles gotta stay open while using it. If it's still not working try restarting your PC. When I get errors that's how I fix it most of the time.

[meager iron]

ight imma try later, thanks for the help before, i will ping you if i come up with any new problems 👍

[ornate wasp]

Sure, glad I could help. You take care 😁

[tough fiber]

how can i fix model inferer issue ...

can anyone help me

[glass drum]

is this over training?

[pastel oak]

Youre at the start of the model, keep training

click the blue button on the right side

to expand the graph

[glass drum]

i'm gonna run out of disk space didn't know it took that much 💀
it's already at 80gb and there is 25gb left on the disk

[tight ether]

It seems a bit early to judge. let's train it a bit more.

[glass drum]

the dataset was very short btw less than 3 mins

[pastel oak]

You enabled ckpt saved at every frequency, turn it off for the next time

delete pth files that you dont need of older model versions at 1k steps or whatever

it should never be 80gb

[glass drum]

yeah it's the pth files. 800mb each and it made 100 so far

[pastel oak]

delete the ones you dont need

everything under 2.8k you can throw out safely

everything above 3.2k until recently you can delete aswell

[glass drum]

what's the differnece between Dxxx.pth and Gxxx.pth?

[pastel oak]

You use the most recent one of both D and G to continue training the voice model if you stop now

for those you should also only have one of each

[glass drum]

yeah the d and g path are the ones i was referirng to

I'll delete older ones

[pastel oak]

yea you enabled that one setting you shouldnt have

unlucko

[glass drum]

rip

which option is it for next time?

using applio for training btw

[pastel oak]

I dont know if applio is the same but on mainline its called something like "save only the latest ckpt file to save diskspace" that setting is turned off by default, shoudl be turned on

[glass drum]

found it

[rare goblet]

Ayo? @glass drum level 3 !!!

[glass drum]

you recommend turning both off?

[pastel oak]

nono the second one is important to be enabled

so you can select a certain X steps version of the model

since when you overtrain you wont know until you trained further, so you want the older pth files of that OT point

[glass drum]

got it, thanks for the help boss

[glass drum]

the file names has the epoch number not the step. how can I find which step it is?

since the graph shows steps

[pastel oak]

Would wait for someone who uses Applio to reply, I thought it always shows steps

[glass drum]

seems like it's ~12 steps per epoch so gonna go with that

unless someone knows a better way

[rustic trellis]

Hey i have a question, what is the best pretrain_type for singing models?

[rustic trellis]

which one of these

@helpers

[zinc anchor]

I'd say original is pretty good in general

[rustic trellis]

i heard something about OV2

[zinc anchor]

OV2 and Rin take care of datasets shorter than 5 mins better than the originals

[rustic trellis]

yeah my dataset is 6+ minutes

[zinc anchor]

ItaIla should be good for that too

Snowie is the russian pretrain

[rustic trellis]

oh ok

so Italla or original?

[zinc anchor]

I'd say go for it with the original

[rustic trellis]

alrighty thanks!

[zinc anchor]

Np

[rustic trellis]

also what is a good amount of epochs for a dataset of about 6 and a half minutes?

it isnt that good of quality aswell

[rare goblet]

Ayo? @rustic trellis level 3 !!!

[rustic trellis]

@helpers

[zinc anchor]

You could use 200-250

[rustic trellis]

thanks!

[frank olive]

hello link rvc google?

[zinc anchor]

-colab

[azure marsh]

Suggestions for @frank olive

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[frank olive]

other model italian?

[strong pagoda]

If I want to train a french speaking model, should I use the same model as suggested in the Training RVC v2 models Guide ?

[zinc anchor]

Italian models? Your best bet is to go on weights.gg and search for models with the "italian" tag

https://weights.gg/invite?inviteCode=e7c90

[zinc anchor]

Yeah, there's not a specific french pretrain right now

[strong pagoda]

Ok I'll try some stuff, see if it still works

[thin cliff]

Im not sure if this is the right place to ask but Im trying to learn how to make ai music and I have gotten up to easyui mangio-rvc, but when I paste the link to download my model I just get an error every time

I tried it in the google colab, and in a localhost

Both give me an error, Ive tried models on drive and huggingface and they dont work

Is there no way to just upload it myself since I have the file already downloaded on my pc?

[topaz basin]

Greetings, any idea on how to remove distortion in RVC voice.ai app because when I use it it sounds really laggy and like a robot?

[thin cliff]

It doesnt seem to actually matter what I put in it just always errors, even on the default demo link

[pastel oak]

Could be that the colab easyguo mangiorvc you use is outdated.

I'd suggest you use Ilaria's RVC

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[thin cliff]

Thank you!

[rare goblet]

Ayo? @thin cliff level 1 !!!

[pastel oak]

Choose one of the ilarias both should be the same(?)

[thin cliff]

I was using one from a video I found on youtube thats pretty old so yeah that makes sense its outdated

[pastel oak]

We do not offer support for voice.ai

[pastel oak]

Yeah all of them are outdated ngl unless its from members of this server

Theres also guides on how to use these all in this server

Check out the docs in the pinned message, go over to Cloud -> ilaria for proper guide

[thin cliff]

I just tested it out and Ilaria's works 🙂

How long should it take to process?

Its on 5 minutes now

Im converting a 2 minute vocal track

Nvm it finished

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

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

🇪🇸 Spain

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

🇫🇷 France

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

🇩🇪 Germany

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

🇯🇵 Japan

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

🇰🇷 Korea

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

🇵🇱 Poland

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

🇷🇺 Russia

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

🇮🇹 Italy

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

[thin cliff]

Thank you so much for the help Shad, I finished making my first song and it worked perfectly!

[light flint]

-collab

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[low tangle]

🤨

[dawn nebula]

mode collapse ggs

[low tangle]

is there anything i can do about it

[rare goblet]

Ayo? @low tangle level 4 !!!

[dawn nebula]

how long is ur dataset

[low tangle]

90min

[dawn nebula]

rvc does not work well with long dataset, try to get the best 25 minutes from your 90 minutes audio

[echo estuary]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[proud elbow]

quality over quantity, no matter how long it is, you might have probably used poor quality audio or improperly processed it (e.g. truncating it too short instead of splitting using audio labelling in Audacity)

[low tangle]

its very high quality i promise, and i made sure each file was long enough

i guess i should probably cut it down

i just worry about getting all the right phonemes and pitches even though ik it doesn't necessarily need it

[dawn nebula]

ye

25 minutes max

[low tangle]

what would an ideal dataset consist of in terms of pitches, phonemes, vocal qualities, and the combinations of such?

[dawn nebula]

just trying to have a variety on all dat

length should be below 30 minutes

then its all g

[low tangle]

i added the google drive link to ilaria and made it public but theres no models in the dropdown (i refreshed)

is it looking for a .zip?

if so which files does the archive need?

ill try pth and index and see if that works

yay that worked

[wet wren]

just got done extracting features, then i trained my model, where do i go from here?

to use it and stuff

[strong belfry]

Load up the model in inference and convert acapellas

[wet wren]

how do i do that

[strong belfry]

Are you using a webui?

[wet wren]

yes

applio

[strong belfry]

Go to the first infer tab

Select your model (whatever epoch) and select the index

Might need to click refresh

[wet wren]

it has a dropdown button but it doesnt do anything

[rare goblet]

Ayo? @wet wren level 3 !!!

[strong belfry]

You might need to click in the middle of the box

I think it was buggy?

And also refresh/restart if no models show up

[wet wren]

doesnt work, it lets me type in it tho

[strong belfry]

Check if there are models in the logs folder where you installed applio

[strong belfry]

There should be more models pth files in the same logs folder

Something is wrong cause there is no trained models below that folder?

[wet wren]

a mute folder with random folders with no pth

[wet wren]

am i supposed to train or generate an index first

[strong belfry]

There is no difference

I usually do index first cause it’s faster

[urban berry]

-collab

-collab

[strong belfry]

-rvc

[azure marsh]

Documentation

📚 All-In-One English documentation

https://docs.aihub.wtf/

❔ English FAQ

https://docs.aihub.wtf/faq

Credits: AI HUB FRANCE

✨ More guides

Full AI Voice Model Training Guide (Local)

Link: YouTube
Credits: Christopher Villanueva

Model training with Mainline RVC

Link: Rentry
credits: Raven (ravencutie21)

AICoverGen Colab Guide

Link: Google Docs
Credits: Eddy (Spanish Helper)

Create a model with RVC disconnected (colab)

Link: Google Docs
Credits: Angetyde

How To Make an AI Cover With Ilaria RVC

Link: Rentry
Credits: 👽 Julia (ailen2091)

[strong belfry]

eh I don’t remember

[wet wren]

yeah it still didnt work and theres none of that

[strong belfry]

You must not have actually trained the model

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

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

🇪🇸 Spain

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

🇫🇷 France

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

🇩🇪 Germany

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

🇯🇵 Japan

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

🇰🇷 Korea

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

🇵🇱 Poland

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

🇷🇺 Russia

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

🇮🇹 Italy

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

[split frigate]

is this sample with encoding ok?

[zinc anchor]

Yeah

[split frigate]

ty

[rare goblet]

Ayo? @split frigate level 8 !!!

[split frigate]

when training i choose 40hz right ? as long as its 44hz

[zinc anchor]

Yeah

Probably a better option than 48k in this case

[split frigate]

tysm

[pastel oak]

Yes, but the thing is that windows always lies as to what your ACTUAL sample rate is

It might say 44.1hz but your spectogram is actually at 36k, in that case you need 32k..

with the program "spek" you can see it

[split frigate]

in taining page there is 2 options only

[pastel oak]

I dont want to confuse you even more, you can try training at 40k like others do but if you want to be very very technical xd

[pastel oak]

Click on v1, then click on v2

Its bugged

[split frigate]

i want to be

nice! tysm

i will see the spek program

from where can i download it ?

[pastel oak]

https://www.spek.cc/p/download

Windows Installer

When you put your audio file in there, send a screenshot and I can explain what you need to look for

[split frigate]

40hz right?

[pastel oak]

You always look on the left side at the kHz and then you multiply the number by 2 that you see

So your specto is at around 15 kHz, times 2 = 30 khz

it doesnt even reach 32 so you should definitely try 32 for now

And next time try to see if you can get your files from somewhere else

[split frigate]

And next time try to see if you can get your files from somewhere else

wdym?

i get it mostly from youtube

[pastel oak]

what youtube downloader did you use?

[split frigate]

https://tuberipper.cc/45/save/flac

mostly

if u got better one i would use it]

[pastel oak]

Youtube is inconsistent - sometimes the audio is rlly good with 44khz but sometimes its literally 20khz

so idk if its the videos or the tuberipper.cc

I can show you what I use later when im on pc

[split frigate]

okay
tysm

[split frigate]

if it 44k i will choose 48hz better right?

[pastel oak]

Ehh i dont understand but ill explain what you should do

When you download file from youtube, open that file in spectogram. Check for its khz top. If its 24khz * 2 = 48khz then when you edit for cleaning vocals, use 48khz

If EVERY file you have is 16khz * 2 = 32khz, then you do 32khz instead
Is that understandable?

[split frigate]

yessir i get it already

[pastel oak]

Theres UVR models for that, I dont know off the top of my head. Never used drum removals

Its also on mvsep probably

[split frigate]

i ment 44k is between 40khz and 48khz

which is better?

[pastel oak]

Ahh

40

Because:

44khz doesnt have the frequencies like 48, so the model will have noise, so you always go down

Same thing gor 36, 37, 38khz = you choose 32

[split frigate]

i get it

tysm

[split frigate]

are stereo and 24bit are great with 32khz or there is better?

[pastel oak]

It should work just fine

I dont know if theres any better

[split frigate]

tysm u r great

[split frigate]

Wav or flac what do u prefer?

[pastel oak]

both are good so it doesnt matter

If I have source raw footage of audio at high quality and sample rate id use wav. Everything else like from youtube, instagram etc, flac is enough. If you want you can choose wav but not needed

[split frigate]

tysm

[proud elbow]

normalize below 0 dB before exporting to any PCM, unlike 32-bit float that won't get clipped
also better let the format sample rate 44.1k for compatibility (not confused with the sample rate used for training)

[tough heron]

what'sthis?

[queen escarp]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[dawn nebula]

what are you using

[light trench]

can someone help with installing applio?

i clicked install on colab

i clicked run, now idk whats happening

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

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

🇪🇸 Spain

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

🇫🇷 France

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

🇩🇪 Germany

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

🇯🇵 Japan

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

🇰🇷 Korea

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

🇵🇱 Poland

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

🇷🇺 Russia

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

🇮🇹 Italy

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

[tough heron]

dml

[dawn nebula]

?

[hot lagoon]

Yo, what’s the best settings for covers? I know it’s a personal take, but I still want to have some recommendations from others.

[zinc anchor]

There isn't a best setting, when you use like AICoverGen or Ilaria RVC you can stick with the default settings or play with them if you're not happy about how the result sounds

[hot lagoon]

About AI cover gen what models do they use?

[zinc anchor]

You mean for vocal/instrumental separation and stuff like that?

[hot lagoon]

Yea

[zinc anchor]

Voc_FT for vocal/inst separation, MDX Karaoke 2 for lead/background vocals separation, Foxjoy Reverb HQ for de-reverb

[pastel oak]

Do you think foxjoeys reverb is better than the standard uvr dereverb one?

[proper shale]

it's less aggressive

and doesn't have the 17khz limit

[glass drum]

is there a way to have applio add the steps number in the trained model?
currently it's only saving the epoch number for me, for example test_520e.pth

[proper shale]

[hot lagoon]

When useing vocal split mode shound or shound not derever the back vocals?

[proper shale]

doesnt applio have a steps counter or anything similar

[proper shale]

dereverb the backings? not really, unless u want to

[hot lagoon]

k

[zinc anchor]

Reverb HQ only works on stereo reverb, while DeReverb-DeEcho also removes some mono reverb

IIRC

[proud elbow]

AMD/intel dedicated GPU or iGPU?

[pastel oak]

As in, standard one only dereverbs up to 17khz right?

[proper shale]

yeah

[fluid topaz]

use applio, it has steps + epoch on the output

[proper shale]

while foxjoy's reverb does 22khz

aka 44k

[proud elbow]

foxjoy dereverb removes stereo reverb only, yet the UVR one can remove mono reverb as well, though the former is preferrable for stereo reverb
(note that if foxjoy one fails to remove, it is mono reverb)

[pastel oak]

Thankss

[glass drum]

I am using applio tho as i said lol. i'm using last version and it's not saving the steps

[proud elbow]

tick "high end process" in VR arch settings to get fullband

[fluid topaz]

no idea

[glass drum]

😦

[fluid topaz]

use 3.1.1

[glass drum]

uh that must be it

can i update or have to reinstall?

[fluid topaz]

yeah, replace/overwrite the files

[pastel oak]

oo ty

[verbal oasis]

Hi guys, is it possible to train a tts model and regulate the speed of it so that it reads very slowly without glitches in voice like it's in slowmo?

[proud elbow]

Edge TTS API actually has rate (speed) parameter but Applio doesn't use it, so might need to modify if you know how to code

[proper shale]

ig if you take a tts result and slow it down you could maybe do that... wouldnt be realistic

or do what @proud elbow said

[verbal oasis]

Do you know if there's a google colab where I can upload my trained voice model (trained with easyGUI and uploaded in huggingface) and use it to read with the possibility to adjust the speed of reading?
I'm currently using Soni Translate in colab the RVC Custom voice, it works great but I cant' slow down the readin in a natural way, it's all forcing the model

[verbal oasis]

So my question is, how do I train a tts?

is there a easyGUI version to train TTS not voice models?

how does it work?

[proper shale]

welp, you can't train tts models on RVC

because, audio conversion

ig you could use GPT-SoVITS

[verbal oasis]

any colab that you know?

[proud elbow]

some colab links there
#1229627925658337360 message

[verbal oasis]

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

[rare goblet]

Ayo? @verbal oasis level 5 !!!

[verbal oasis]

founded this inside GPT-soVITS

[proud elbow]

there's also no webui one

[proper shale]

dunno if this works but try it

[verbal oasis]

yeah, let's see

what it is

anyway do you guys can tell me how does it work?
I' m new to all this, so it looks like first comes the TTS reading model and then the Voice, is it correct?

if yes then my question is how's a TTS model trained? Is it based on Audio? how it determines the pace/speed of reading?

[proper shale]

you give it audio files, ask it to transcribe said files, then it does the training part

[verbal oasis]

so it starts from audio and it the speed of reading is automatically detected?

[brittle wing]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[verbal oasis]

thanks, anything to train tts?

[proper shale]

yeah uh... you give it a reference audio, and also the audio it's been trained on affects the pace too

[brittle wing]

is training audios easy?

[rare goblet]

Ayo? @brittle wing level 1 !!!

[proper shale]

https://docs.aihub.wtf/tts/gpt-sovits/ in case you wanna know better

[proper shale]

i mean... yeah? just takes some getting used to

https://docs.aihub.wtf/ in case u wanna learn more

[brittle wing]

ty

[verbal oasis]

This is way too complicated for me unfortunately.
Do you have know any slow tts reader ready to use?

[proud elbow]

try this little girl voice

[verbal oasis]

Is this easy gui?

[proud elbow]

Applio

[verbal oasis]

thanks, let me try it

Is it possible in Applio uploading a custom TTS and a custom voice?

[proud elbow]

no custom voices here but prob some plugins for custom TTS

[verbal oasis]

plugins?

[proud elbow]

idk much but there might be

[verbal oasis]

I'm getting Error when hitting Convert on Applio

Do you have any ideas why?

[proud elbow]

the RVC voice model required

[zinc anchor]

Look at the Colab notebook box, it usually shows more info

[zinc anchor]

Yeah that's true

[verbal oasis]

But then I need to upload a model right?

I have a huggingface link with my model

if I go into Download

it gives me error

[tranquil cliff]

does colab work?

[proud elbow]

alternatively, put the pth & index file in logs\yourmodel, then click refresh in inference tab

[tranquil cliff]

the runtime?

any signs of life?

[verbal oasis]

ah someting happened

even if Error appear

it downlaoded the model

[flint vapor]

It is working now?

[verbal oasis]

Yes it is working now

but I have the problem that the TTS is a female and my model is Male so when converting it mixes them somehow and makes a very cringe voice

any workaround?

[proud elbow]

-12 pitch or might be different value

[verbal oasis]

I'm trying let's see the effect

I'm aware it could sound very weird

why it mixes the voices?

shouldn't it use only my voice model?

and use the TTS solely as a style of reading?

[half cove]

@proud elbow I see you active in help chats loads

do you have any interest in joining staff?

[proud elbow]

I might doubt my activeness though I have chance to check the chats, but sure as needed

[vestal lion]

thats it

i dont know what this means

i tried reinstalling

and doing different versions

[rare goblet]

Ayo? @vestal lion level 1 !!!

[vestal lion]

they all do this

btw for the people with message logger go to #🔍│help-w-okada i sent to wrong channel

[pastel oak]

Which GPU do you have, and which version of wokada did you download (onnxgpu-cuda or DirectML)

Answer in #🔍│help-w-okada

[vestal lion]

Rtx 3050 ti laptop and an i5

oh k

[half cove]

ngl i think you would be accepted if you apply for staff

[proud elbow]

limited vram, only using it alone

[proud elbow]

prob later

[brittle wing]

do you guys know why i keep getting this error in rvc gui?

[half cove]

[proud elbow]

rvc gui is outdated, try following instead

-rvc

[azure marsh]

Documentation

📚 All-In-One English documentation

https://docs.aihub.wtf/

❔ English FAQ

https://docs.aihub.wtf/faq

Credits: AI HUB FRANCE

✨ More guides

Full AI Voice Model Training Guide (Local)

Link: YouTube
Credits: Christopher Villanueva

Model training with Mainline RVC

Link: Rentry
credits: Raven (ravencutie21)

AICoverGen Colab Guide

Link: Google Docs
Credits: Eddy (Spanish Helper)

Create a model with RVC disconnected (colab)

Link: Google Docs
Credits: Angetyde

How To Make an AI Cover With Ilaria RVC

Link: Rentry
Credits: 👽 Julia (ailen2091)

[brittle wing]

oh

i didnt think it was

is the Ilaria RVC link broken?

?

whenever i click on it it says this

[rare goblet]

Ayo? @brittle wing level 5 !!!

[proud elbow]

-colab

[azure marsh]

Suggestions for @brittle wing

☁️ Google Colabs

• Applio, by IA Hispano 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
• RVC Disconnected, by Kit Lemonfoot Google Colab
• easyGUI, by rejects Google Colab

🤗 Hugginface Spaces

• Applio, by IA Hispano Huggingface Spaces
• Ilaria RVC, by thestingerx Hugginface Spaces
• RVC-HFv2, by r3gm 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, -local, -overtrain, -realtime, -rvc, -help

[pastel oak]

https://colab.research.google.com/drive/16LkwvFZeudTpUOsE_6bMjOq2qkxFo8Hr?usp=sharing

works for me

[brittle wing]

alr

[vagrant mortar]

I'm trying to download the new Japanese-speaking hubert, just download the checkpoint .pt and replace it in the applio folder, or do I have to put something else? For example, this code is mandatory, where is it placed?

Or just placing the new "hubert_base.pt" in the folder where I have applio installed is enough?

help me!!!

[finite galleon]

Replacing it should be enough

They both load using the same code

[vagrant mortar]

ok

[brittle wing]

how do i use this

[restive violet]

hellooooo

how do i solve this

what is nvidia

[brittle wing]

its your gpu

[vagrant mortar]

And when will Hubert perform? You have to do something else (preload some configuration or something). For example, when I start training, is the hubert applied when I perform the tone extraction or when I load the dataset, or, at the time of training? Because in the new version of applio, I don't know if it is using the new hubert file or the old one?

[brittle wing]

you probably dont have a driver installed

[restive violet]

how do I install a driver

[brittle wing]

from the link

[restive violet]

now what

[brittle wing]

oh uhhh

i forgot you need to find out what gpu you have

[restive violet]

how do I find out

[brittle wing]

its in setting

then

[proud elbow]

its cloud not local, and make sure the runtime uses T4 also have run the first installation cell

[restive violet]

what do you mean

[finite galleon]

Hubert extracts the features of an audio. When training the hubert is only used on the pitch and feature extraction step and not during the actual training process.

[restive violet]

how do i put the runtime

[proud elbow]

check if the status in top right corner has T4, or Runtime -> change runtime type

[restive violet]

can you send an image

[brittle wing]

how do i add the model in the hugging face spaces

[proud elbow]

not rn 💀

[brittle wing]

how do i add voices to the hugging face spaces

do i like drag it in or

-help

[azure marsh]

Command has expired

[vagrant mortar]

So I don't need to use the code separately or cite it in some way, or use it with the "transformers" and "soundfile" libraries? To train my models with the new Hubert, wouldn't this be applicable in that way?

[finite galleon]

The code uses fairseqs checkpoint utils function which is made for hubert

[ocean beacon]

Excuse me, does anyone know how to resolve this error?
"JSONDecodeError Traceback (most recent call last)
<ipython-input-11-320ee20c9227> in <cell line: 31>()
31 if os.path.exists(config_path):
32 # File exists, proceed with creation of creds and client
---> 33 creds = Credentials.from_service_account_file(config_path, scopes=scope)
34 client = gspread.authorize(creds)
35 else:

5 frames
/usr/lib/python3.10/json/decoder.py in raw_decode(self, s, idx)
353 obj, end = self.scan_once(s, idx)
354 except StopIteration as err:
--> 355 raise JSONDecodeError("Expecting value", s, err.value) from None
356 return obj, end

JSONDecodeError: Expecting value: line 1 column 1 (char 0)"

[pastel oak]

Send the colab link

[low shard]

its an outdated one from yt lol

i solved this already in general chat

[fluid topaz]

anyone have link for rvc1228?

[azure marsh]

RVC Guides (How to Make AI Cover)

Documentation

🇺🇸 English (main)

https://docs.aihub.wtf/

​

Translation by country

🇧🇷 Brasil (PT-BR)

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

🇪🇸 Spain

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

🇫🇷 France

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

🇩🇪 Germany

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

🇯🇵 Japan

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

🇰🇷 Korea

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

🇵🇱 Poland

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

🇷🇺 Russia

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

🇮🇹 Italy

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

[fluid gyro]

Um hi, paperspace not working with wget https://huggingface.co/lollenape/LollenApeRVC/resolve/main/install.py ??

[odd shale]

-rvc

[azure marsh]

Documentation

📚 All-In-One English documentation

https://docs.aihub.wtf/

❔ English FAQ

https://docs.aihub.wtf/faq

Credits: AI HUB FRANCE

✨ More guides

Full AI Voice Model Training Guide (Local)

Link: YouTube
Credits: Christopher Villanueva

Model training with Mainline RVC

Link: Rentry
credits: Raven (ravencutie21)

AICoverGen Colab Guide

Link: Google Docs
Credits: Eddy (Spanish Helper)

Create a model with RVC disconnected (colab)

Link: Google Docs
Credits: Angetyde

How To Make an AI Cover With Ilaria RVC

Link: Rentry
Credits: 👽 Julia (ailen2091)

[fluid gyro]

We still can train RVC Mangio with Paperspace right?

[tranquil cliff]

Yes

but you pay

[fluid gyro]

yess i know

[rare goblet]

Ayo? @fluid gyro level 1 !!!

[tranquil cliff]

anyway now the things you did on paperspace are also on colab

[fluid gyro]

okayy, im new, If I have any doubts I'll ask

[tranquil cliff]

In my opinion it's better to applio anyway

[fluid gyro]

Im at uni so no GPU, thinkpad carbon x1 jajj

[tranquil cliff]

colab does not need your computer's GPU

it's online

[fluid gyro]

yeah but I ran out

[tranquil cliff]

the only flaw is that it lasts 4 hours a day

[fluid gyro]

anyways I already paid

[tranquil cliff]

ah

okay

[fluid gyro]

what do I upload? in data source, the zip?

cz the repository aint working

so I downloaded the og file 4gb to now upload it manually

[tranquil cliff]

wait while I tag people more expert than me ahhaha

[fluid gyro]

thanks dude!

[tranquil cliff]

@red kayak sorry if I tag you but I don't understand much about paperspace

[fluid gyro]

neither do I hahh so if u could give me a hand, Im obsessed to pull a model thru just cz the ones who made some where horrible

[tranquil cliff]

@zinc anchor do you use paperspace?

[zinc anchor]

Nope

Might experiment with it in the future

[red kayak]

-paperspace

[azure marsh]

Suggestion for @tranquil cliff

RVC v2 Tutorial for Paperspace

💡 Click on the button below to open the guide.

Credits: LollenApe

[red kayak]

If they need guides for RVC here

[fluid gyro]

thankss!!!!!

[red kayak]

It's only how to set up thouh

[fluid gyro]

well at least its something hahha

[red kayak]

I can help later. What do u need to do

[fluid gyro]

i only have the perfect .flac , nothing more hahh

[fluid gyro]

Just create a rvc2 model

for me to use it in a song

cz the ones created are shit

[red kayak]

Okay then follow the guide

[fluid gyro]

thanks a lot, seriously!

[fluid gyro]

the one u sent right?

[red kayak]

Mhm

Should have all u need to set up

[odd shale]

Hello.

You're the one who needs help with Paperspace?

You can read the guides tho.

But in case of any other issue let me know.

[fluid gyro]

thanks everyone!

[rare goblet]

Ayo? @fluid gyro level 2 !!!

[latent sorrel]

69 #Rename the folder to experiment_name and move it to /dataset/.
70 print("Dataset Type: Single Speaker")
---> 71 fi = os.path.join(temp_directory, experiment_name)
72 os.rename(directories[0], fi)
73 shutil.move(fi, final_directory)

NameError: name 'experiment_name' is not defined

any help?