✨│ai-help

[pure ledge]

please I will really appreciate any help . ive been stuck for a whole hour

[feral saffron]

https://huggingface.co/NoahZe/Frank/resolve/main/Frankie1.zip?download=true

[low shard]

U need to click the lowest point, get the step, and then in the weights folder you need to find the pth with the most similar pth in the file explorer

[low tangle]

is this normal on rvcdisconnected...? it is training, i can see the epochs in the logs, but the tensorboard says this. ik common question i just havent seen a clear answer

[brittle wing]

Yes but I can't I had it saved at 380 through are you sure it looks like overtraining If it does I'm concerned

[low shard]

Did u put it public

[feral saffron]

Uhh let me check

No I did not

My bad

[low shard]

wdym with you had it saved at 380? Also yes it looks like training as it got to the lowest point and only goes up now

[feral saffron]

Now it’s public

Do you need a new link

[brittle wing]

It was at 380 exported before I screenshot the tensorboard

[low tangle]

lmfao nevermind i just needed to reload it

[rare goblet]

Ayo? @low tangle level 3 !!!

[pure ledge]

please im loosing my mind

I tried multiple zip links and googledrives too , it just doesnt want to give me public url to use gradio

[heady valve]

Help i downloded an model and it says can't open file

[low tangle]

are you trying to open the .pth

or the .zip

sorry devil i really hope u get help 😭

[heady valve]

The .zip

[pure ledge]

can someone ping me the latest colab site ? maybe i have an old one

[pure ledge]

its been 2 hours already

[low tangle]

what are you using for inference? like to convert vocals

[low tangle]

hopefully more people will be awake throughout the day

[heady valve]

Elevenlabs

[low tangle]

im not an expert at all but sometimes u dont need to open the zip. however if you do, windows should just be able to unzip it (assuming ur on windows) by right clicking and clicking extract

the gui i used to use just wanted the zip archives but others want the files inside

[heady valve]

im on mobile

[low tangle]

ah ok

and elevenlabs wants the files inside the zip?

if u send me the file i can unzip for u

[heady valve]

Wdym? the model?

[low tangle]

yeah

[pure ledge]

anyone ?

[heady valve]

Alr

[rare goblet]

Ayo? @heady valve level 2 !!!

[feral saffron]

I hope u get help devil. I wish I knew how to help u bro

[rare goblet]

Ayo? @feral saffron level 5 !!!

[pure ledge]

thanks man... I just want a working collab link so I can use gradio

[rare goblet]

Ayo? @pure ledge level 2 !!!

[feral saffron]

Here

-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

[pure ledge]

kinda loosing my mind at the moment

[feral saffron]

Maybe this will help

I use llaria for making covers

And I trained a model on rvc disconnected last night however it’s not working for some reason

Hopefully if nick gets back soon he can help me some more

[low tangle]

u can also just use an online tool that unzips files, like idk this one looks fine https://www.ezyzip.com/unzip-files-online.html

[pure ledge]

you are a hero

[low shard]

i js checked it and its all fine, what ilaria are you using?

[low shard]

check our docs https://docs.aihub.wtf

[feral saffron]

I’m using the one from the -colab thing

[low shard]

the one from the docs right?

[feral saffron]

-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

[feral saffron]

I use the link from this thing

[low shard]

oh yea its the same one from our docs, could you try to click runtime, disconnect and delete runtime,and re run everything?

[feral saffron]

I will try that

[dusk oracle]

okay so, i tried making a model with rvc disconnected and it of course disconnected me for being out of time :/ i did the inspect element thing that it suggested me to do

if anyone knows easier ways to make models hmu pls cuz i am not capable of doing stuff like this

[feral saffron]

I tried it and it still didnt work

[proper shale]

https://docs.aihub.wtf/

[low shard]

Try to refresh the voices list

it should be there even if it says error

may be an error of ilarai rvc as its abbandoned now

[feral saffron]

Alright I’ll try that

[feral saffron]

BRO YOU DID THE IMPOSSIBLE

it works!!

I can’t thank you enough

Thank you so much bro

You helped me so much I really appreciate it

[low shard]

lmao yw, dont ask me why it says error but working lol, just know that ilaria rvc is not updated anymore and btw u can find other colabs too in our docs https://docs.aihub.wtf, anyways it should still work ilaria rvc

[feral saffron]

Thank you so much man

U need a promotion or something

[low shard]

youre welcome, i like helping people :)

[low shard]

So you got disconnected for inactivity or finished your daily free gpu ?

[proven hill]

reworking for the beta 3, beta 2 was unstable

[low shard]

btw abt that ilaria rvc colab that is in the docs some people get error in downloading, then refresh the list and it appears lol

should we update the docs to have ilaria mainline or its still in beta

[low shard]

if u scroll up i think u get what i mean

[zinc anchor]

The real test to see if it works is still inference

Do both the "model" (pth) and the index file appear?

[low shard]

Hey btw did the model work for inference? Like ur efreshed list, and it wokred all fine making the ai cover right?

[low shard]

yes the files are correct

[proven hill]

whats the error

[low shard]

‘the download link is invalid’ but like, if u refresh u find it here a d it works all for inference and all the files are correct and public

see down from here

#✨│ai-help message

[proven hill]

https://media.discordapp.net/attachments/1159290139609137264/1219666894672367676/ilariaerror.gif

[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

[feral saffron]

Yes!!

It’s all working perfectly thank you

[low shard]

no no, that was already done, its just for some models it displays the error download even if its downloaded for some reasons, not that big of a issue but could make the users confuse, but deleting and disconnecting runtime was already tried

[low shard]

yw was just asking to try to analyze the problem

[proven hill]

thats strange

maybe theres some random ass files confusing the program

in the download i mean

sometimes people put there .txt files or jsons

[low shard]

nope

thats why its really weird

[proven hill]

uh

only thing i can think of is because of the index being so big

literally the only thing

[brittle wing]

anyone knows how i can fix rvc lag in google colab?

my wifi is good

on some days, it works perfectly well, and on somedays i can't even say "hello"

it just cuts off.

[low shard]

kinda weird tho it does that only on ilaria rvc, could be bc is from mangio ig

[proven hill]

try it with mainline

maybe there the issue is non existant

[low shard]

oke i will and tell u

also is mainline stable to be putted in the docs ?

[proven hill]

the colab yes

[livid monolith]

/help rvc

where is the help bot

[verbal oxide]

what's the different BS roformer viperx 1296 and 1297?

[proper shale]

iirc 1297 is the best one

but test n see

[brave garnet]

Voicechanger Settings (Okada)

⠀

Settings for Nvidia GPUs

F0 Det.: rmvpe (suggested for all series)

RTX 40-series: 80-96 chunk | +16384 extra
RTX 30-series: 96-112 chunk | +16384 extra
RTX 20-series: 112-128 chunk | +16384 extra
GTX 16-series: 128-192 chunk | +8192 extra
GTX 10-series: 128-192 chunk | +8192 extra

Advanced Settings

Protocol : Sio or Rest
Crossfade: 4096 start 0.2 end 0.8
Trancate: 300
Silencefront: Off
Protect: 0.5
RVC Quality: Low

⠀

Made with love by Ilaria 💖

[verbal oxide]

Oh thanks..

[crude plinth]

help
Frequent errors occur. Please check if the model of the framework being targeted is loaded.

[brave garnet]

Voicechanger Settings (Okada)

⠀

Settings for AMD GPUs

Don't forget that your models needs to be converted in ONNX!

F0 Det.: rmvpe_onnx (suggested for all series)

7xxx XT cards: 112-128 chunk | +16384 extra
6xxx XT cards: 128-192 chunk | +16384 extra
5xxx XT cards: 192-256 chunk | +8192 extra

RX 580: 192-256 chunk | +8192 extra
RX 570: 192-256 chunk | +8192 extra
RX 560: 256-384 chunk | +8192 extra

Advanced Settings

Protocol : Sio or Rest
Crossfade: 4096 start 0.2 end 0.8
Trancate: 300
Silencefront: Off
Protect: 0.5
RVC Quality: Low

⠀

Made with love by Ilaria 💖

[proper shale]

Ignore that unless something's going wrong

[split frigate]

step is the number which I consider to take in the lowest point right?

and in weights it is the number beside the epochs right?

[split frigate]

like
Name_e45_s2205.pth

s is the step in lowest point right?

[rare goblet]

Ayo? @split frigate level 6 !!!

[proper shale]

yea, steps are what compose epochs

u shouldn't really always take the lowest point, most of the time the better model is the latest, unless its overtraining

and yes, they are besides the epoch count

[split frigate]

how can i know if its overtranining

u mean by latest like the last epoch?

[proper shale]

yeah the latest epoch

[split frigate]

i putted to 1k

[proper shale]

now this, is complicated

you need to check g/mel and g/kl, see if they're still going down, even if g/total is increasing and those two are going down your model is getting better

g/total is just the average model

[proper shale]

you won't train to 1k in one normal colab session tbh

but idk, if you're training locally...

[split frigate]

i just want best result i can get thats why i choose 1k

[proper shale]

fair enough

i suggest testing every 50/100 epochs

[split frigate]

u mean let the g total and seek after these 2 right?

[split frigate]

i think its huge gap

[proper shale]

yeah u should check those alongside g/total

[split frigate]

check point every 5 epoch

[proper shale]

eh it's not a huge gap tbf

model improvements are really hard to notice every 5 epochs for example

every 50 epochs you can tell it's getting better

[split frigate]

what would u choose

50 ?

[proper shale]

for save frequency? I'd keep at 5 because you can pick a point closer to overtraining tbf

but I'd still try out every 50 epochs

25 might be acceptable too idk

[split frigate]

at first i didnt get it

now i get it

can i test it while traninig?

[proper shale]

don't think so

i tried it on colab once and

cursed

crashed the pc lol

[split frigate]

how u would then?

my EN is bad so hard to understand sometimes

like how can u test the model every 50 epoches

[split frigate]

predictable mate

[proper shale]

after you're like... done with the training

thats what i did cuz colab was limited af

you can stop after a while and check out tho

[split frigate]

and start from beginning?

i dont know what does colab mean tbh

1k epoches looks hard to test it every 50e or 100e and just start from begining

is it like that ?

i want to master RVC as long as i got u

[proper shale]

no you can resume

[proper shale]

that's the google service that gives u a gpu, for research

but eh we do it for funny ai cover

[proper shale]

you don't start from the beginning, as i said earlier

[split frigate]

how can i make it resume

i didnt know that

[proper shale]

just insert the same settings as before and it'll resume

don't preprocess or feature extract again tho

[split frigate]

thats my settings is it good tho

resume like i stopped at 50e

it will continue as 51e or from 1

[tough fiber]

why it gives errno20

[proper shale]

decent, i just wouldn't use 48k all the time tho

[proper shale]

it will start from 50 iirc

btw, when u stop training, make sure it hasn't stopped on an epoch that saves the d/g files

the saving epochs

[tough fiber]

i dont know which part is wrong

[split frigate]

all my models at 48k idk i just feel its better

[proper shale]

folder path, not audio path

[tough fiber]

oh it just want folder path my bad -*-

thx

[split frigate]

i didnt get it

[split frigate]

same happend to me

[tough fiber]

so its just want the folder path

but my voice data will be in this folder right

[split frigate]

yes

folder path that have your voice in or whatever

[proper shale]

when it saves, just make sure u train for another epoch so it doesnt screw up the training files

[split frigate]

the way to stop it force close right?

[proper shale]

yeah or u can click on stop training in the GUI

[split frigate]

im not
and i fell ashamed of saying still i didnt get wdym

sry

this point
"u train for another epoch so it doesnt screw up the training file"

[proper shale]

ok, example:

ur at epoch 100, which is a saving epoch (when the stuff saves)

if you intend to stop, u train until epoch 101 to make sure everything saved properly

[split frigate]

i get it

[proper shale]

also, next time, i suggest clicking on yes to "Save only the latest ckpt to save disk space"... unless you're on an SSD

[split frigate]

u mean completely save the epoch and it moves to another

[split frigate]

m.2 ssd

[proper shale]

yeah... that kinda kills your ssd

[proper shale]

yeah

it will resume from epoch 100 tho for some reason haha

[split frigate]

is it better to leave it on or no?

[proper shale]

I'd rather keep it on, but suggesting you'd increase the save frequency if you intend to do so... else it will quite literally kill your SSD after a while

[split frigate]

i got one has like 900gb empty

which im using

[trail trout]

i have some ai made singing vocals but they are full with chorus and harmonics. since it is almost impossible to remove them all, does it means i cannot make voice changing with even best model?

[proper shale]

hm

[proper shale]

yeah no sadly

[trail trout]

but i see that rvc does made the model resing it like it is not changing the voice, it is resinging

so it comes out as solo sound

it looks like there is some way to do it

[proper shale]

yeah cuz rvc is single speaker only

it tries to fit the audio into one person only ye

[trail trout]

so it does not looks like impossible

maybe there is some setting for that

[proper shale]

well, it is

but you could maybe get a decent result

depending on the type of harmonies

[trail trout]

do you know the types of harmonics?

i mean like how many persons

or what type of harmonics

[proper shale]

like, if it's in unison and everybody's singing the same way in pitch, yeah it can work

but if it's like Michael Jackson harmonies...

[trail trout]

there is high and low harmonics

[proper shale]

yeah that's not gonna be good probably

but you can always test and see

[trail trout]

damn my prompts cannot block the harmonics

[trail trout]

why uvc or what ever its name is, is not creating the folders and other outputs anymore?

do you know which setting is stops doing that?

[proper shale]

honestly idk, I don't really use UVR

[trail trout]

if someone knows, pls tell me

[primal barn]

whats a good delay (mic input to cable out)

[trail trout]

that crackling was also not happening on uvr with the exact same settings i get crackling on sound on same songs i did exported before

i think crackling happens because of this so i reverted it back but which one is best you think?

[lyric obsidian]

hey guys

what is model training?

do they like get stronger

https://tenor.com/view/dragon-ball-z-pushups-goku-gif-11726458

[proper shale]

...yeah

you can only do it with ur own models tho

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

[proud elbow]

32-bit float

[trail trout]

idk why but i just reset the settings of uvr5 and re do all the same settings included 32bit, it is no more crackling

what i see is that uvr and also rvc does not do sometimes the settings we do

[stiff goblet]

Batch size 4 or 8 ?? Dataset length is 8 minutes. I'm using Google Colab @proper shale @half cove

[brittle wing]

anyone knows how i can fix rvc lag in google colab? my wifi is good
on some days, it works perfectly well, and on somedays i can't even say "hello" and it just cuts off

[proud elbow]

how many layers of harmonies? anyway UVR BVE is more likely working, or sometimes MDX karaoke 2 works

BVE and/or 5HP also sometimes peel a layer of vocal in some choirs, e.g. in We will rock you

you can try infer some remaining harmonies/chorus in backing vocal stem with crepe hop length 128, or leave it as is/throw in another new track

[trail trout]

yea i know that some models does removes it but then the sound just getting broken.

there is maybe 4-6 person harmonics

this is how it sounds like

[upbeat cloud]

i dont know if i am using the right voice.ai xD

[trail trout]

high sounds low sounds

and middle sounds

[upbeat cloud]

i dk why i cant send a picture of how it look

[rare goblet]

Ayo? @upbeat cloud level 1 !!!

[upbeat cloud]

is this the right one XD or am i wrong?

[half cove]

try 8

for some reason that didnt show up on my pings

[half cove]

we dont support voice.ai here

try w-okada

[upbeat cloud]

how to download and use w-kada?

[proud elbow]

pinned guide in #🔍│help-w-okada

[foggy folio]

what is the best uvr or mvsep model for instrumental?

I want to make some instrumentals of stupid music

?

[broken sky]

do i need to download pytorch again if i had w-okada before (still installed)

[trail trout]

@proud elbowbecause of harmonics it does jumps to up and down and my question is: do you think auto tune can fix this?

[thin stump]

BS Roformer on MVSEP is by far the best one. You need an account to skip queues

[trail trout]

it does not what i did setup why the fkn hell does it happens?

[foggy folio]

yep using that

[quartz moon]

How do I make an AI model without using google collab? It doesn't work for me.

[trail trout]

https://github.com/Mangio621/Mangio-RVC-Fork this one has train option

or you buy here in this server

[quartz moon]

Ok. So if I download it then it I can make them offline as well right?

Ok thank you very much 😄 What are the requirements for running it locally? I have an old GPU.

[trail trout]

i dont know but nvidia of course

[rare goblet]

Ayo? @trail trout level 5 !!!

[trail trout]

i did bought model with that reason

[quartz moon]

Will it run on just the cpu?

[trail trout]

yes but very slow

[quartz moon]

I know its slower but my gpu is a 760...

And wdym by buying the model?

[proud elbow]

inference is ok but not training

[quartz moon]

Will training work on my gpu?

[finite galleon]

Idk which fork ur usin but thats probably just it loading as a part of the whole script. Basically you are still using mangio crepe.

[finite galleon]

Mangio fork havent been updated since last year so use Applio

[trail trout]

does it also effects on quality of export?

[finite galleon]

I mean

There shouldnt be any difference in quality really

Its just that it has more features and more stable

[tough fiber]

when i started it was good but now after 65 epoch

when i start training it trained 100 epoch in 10 min

and why graphic is like this

[rich crown]

yo ixtivh u mind if i ask for some help in dms?

cant send images anywhere here

[rare goblet]

Ayo? @rich crown level 1 !!!

[tough fiber]

im pretty new too so

[rich crown]

o np

[tough fiber]

@proper shale if u free can u check this pls -*-

[trail trout]

these features are also included?

[finite galleon]

Yes

• more

[trail trout]

but it does installling sooooo many things like anaconda etc and later if i want to clean, how to know what to remove?

[brittle wing]

Hello, where can I find instrument models?
Are there only vocal models here?
Can't RVC GUI be used to convert human voice into any instrument voice?

[thin stump]

I believe mode collapses happen because it gets hung up on silences so it doesn't read through the dataset completely. And also dependent on batch sizes

[dusk oracle]

it said something about gpu runtime

[thin stump]

If 8 is giving you those valley's, refresh and try 4 or fix your dataset

[dusk oracle]

and im lacking money so i can't really pay for it

[tough fiber]

it isnt collab

[low shard]

Then it means that you finished your daily free gpu, dont put too much high epochs, and also u can just use a google alt acc, or wait for tmr, u have around 4 hours daily for free of colab gpu

[trail trout]

are these now installed to the folder of applio?

or these ones?

[dusk oracle]

this is too much for me

i dont have this amount time

i gave up

[low shard]

you can #1159289738314919936 or ask in #1191429836321849435 so someone makes it for u

[dusk oracle]

for free?

[trail trout]

@dusk oracleif you wanna buy models, i can suggest you my seller here having helper role which is im happy with him, im also new here btw

[trail trout]

im just saying

[low shard]

in #1159289738314919936 you can do a free request yea

[dusk oracle]

even if i had money i wouldn't pay for this

respectfully

it's just for some random joke

[dusk oracle]

ty

probably won't do it as i wanted to do it myself though

ty for the help

[low tangle]

what's the difference between mangio-crepe and rmvpe? do people prefer mangio-crepe?

[thin stump]

I'm just going to assume mangio crepe is for extremely clean datasets then do they inferences in rmvpe. You should still train with Rmvpe though

[low tangle]

i do have extremely clean datasets that i honestly am not really supposed to have lmao

[thin stump]

#1210625778895552592 message An example of someone using mangio crepe

[low tangle]

i own the files but they weren't intended to be extracted into wavs

ok thanks

[thin stump]

but, I don't know about the specific hop length for a 90 minute dataset so that's why I said to stick with rmvpe

[low tangle]

ah right

[brittle wing]

Hello, where can I find instrument models?
Are there only vocal models here?
Can't RVC GUI be used to convert human voice into any instrument voice?

[trail trout]

you destroyed my life with this suggestion. lol

absolutely nothing is working .d

[finite galleon]

?

[brittle wing]

Is there anyone who can help me?

[trail trout]

only 1 example of 100 errors

[tough fiber]

anyhone have who using paperspace?

[finite galleon]

You prolly did something wrong

https://docs.aihub.wtf/

Just follow that guide

[trail trout]

how can i do something wrong if there is 2 files to run only?

1 click install bat
2 click run bat

[quartz moon]

So I downloaded the file, which file do I run to use it?

[rare goblet]

Ayo? @quartz moon level 3 !!!

[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 stump]

@pseudo flint #✨│ai-help message

[pseudo flint]

wtf is this

[tough fiber]

can someone help me about paperspace

[viscid moss]

@odd shale use paperspace

I hope he can help u

[thin stump]

I linked my explanation. But you're getting a mode collapse because your dataset either has silences or it's dirty

[pseudo flint]

batch size

I made 32 batch size

how do I fiz it

[tough fiber]

i hop he can give answer -*-

also i wrote him on dm

[thin stump]

Oh I see, do a batch size of 8 if the dataset is 30 minutes or below. Batch size of 16 if 40 mins to 1 hour

[pseudo flint]

ok

[tough fiber]

broo im going crazy

mangio rvc fork takes 15gb data and i cant even delete

[viscid moss]

[trail trout]

what you mean with applio cell

him just ruined my life with this suggestion i was rendering fine with mangio wehbgvewhjg

just reopen it you mean?

im not even sure if this one is now doing somethings

[grave marsh]

how do you change your voice after download the zip for the model

[brittle wing]

Is that overtraining

Just asking

[grave marsh]

and is it like for every type of voice call or just for the specific app

[brittle wing]

What are u using

[grave marsh]

https://huggingface.co/sxndypz/rvc-v2-models/resolve/main/saya.zip

[brittle wing]

You out that link in the download model tab

-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

[grave marsh]

I got it downloaded in my pc

[brittle wing]

That's not it

You input the link in one of the colabs' inference tab & it downloads the model so you can use it

[thin stump]

Did you forget everything? No offense but you got all the help here. Idk what you're misunderstanding. The g/total dip is not "overtraining". It's a checkpoint that you should jot down. You can see how it rises up afterwards

[brittle wing]

Some people told me it's overtraining
Also I'm at work I can't concentrate on both the model training and working so it's normal to forget

Some people told me it's overtraining at 55k steps

[grave marsh]

I extracted the zip file into a folder now what?

[rare goblet]

Ayo? @grave marsh level 1 !!!

[brittle wing]

The previous training of the same model before I started from scratch

[brittle wing]

Wait a minute.

[thin stump]

kk, then I would trust them and use the 50k steps

[brittle wing]

THIS

?

[thin stump]

yeah, it's cluttered so you have to guess the step count

[brittle wing]

So the 380 epochs one I had lately probably

In theory I should stop training

Cause it sounds good at 380

[grave marsh]

Sure.

[trail trout]

worker_thread
return await future
File "C:\Users\CilginFlix\Documents\RVC\env\lib\site-packages\anyio_backends_asyncio.py", line 851, in run
result = context.run(func, *args)
File "C:\Users\CilginFlix\Documents\RVC\env\lib\site-packages\gradio\utils.py", line 678, in wrapper
response = f(*args, **kwargs)
File "C:\Users\CilginFlix\Documents\RVC\core.py", line 58, in run_infer_script
infer_pipeline(
File "C:\Users\CilginFlix\Documents\RVC\rvc\infer\infer.py", line 278, in infer_pipeline
get_vc(model_path, 0)
File "C:\Users\CilginFlix\Documents\RVC\rvc\infer\infer.py", line 231, in get_vc
cpt = torch.load(person, map_location="cpu")
File "C:\Users\CilginFlix\Documents\RVC\env\lib\site-packages\torch\serialization.py", line 1028, in load
return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
File "C:\Users\CilginFlix\Documents\RVC\env\lib\site-packages\torch\serialization.py", line 1246, in _legacy_load
magic_number = pickle_module.load(f, **pickle_load_args)
EOFError: Ran out of input

bro i fucking hate this applio

wtf is this

[brittle wing]

https://youtu.be/pdlhk4vVHQk?si=QGDc3WsYtUKTXeCh

Same goes for every colab through just that Applio and Ilaria are only for lead vocals

Honestly ai cover genui is the easiest

[low tangle]

i did 25 epochs last night using rmvpe, should i start over using mangio crepe?

oh wait

sorry misread lol i will continue normally

[tough fiber]

how can i fix this

i accidently deleted so

anyone ? --

[tough fiber]

i think he afk or offline idk he doesnt give answer for 50min

[brittle wing]

can anyone help me install rvc gui

[trail trout]

re run or re install not working, applio is just broken, u have the latest version? i fucking hate applio and the person who suggested me this

[low shard]

rvc gui is outdated, please check our docs https://docs.aihub.wtf

[brittle wing]

oh wait what

rvc is like the one where you can use it as a microphone right

or is that somthing else

[rare goblet]

Ayo? @brittle wing level 1 !!!

[tough fiber]

anyone can help me

bro im dealing hours with this sh*

i really annoyed

[low tangle]

i did everything it said to do on the ai hub guide for resuming training on rvcdisconnected but it decided to start over from the beginning :-/

[odd shale]

Eddy era necesario pingearme mientras estoy fuera de casa?

[viscid moss]

[odd shale]

Pero responde con palabras alusin.

Además que tú todavía nos debes una conversación a mí y Blaise.

[odd shale]

-paperspace

[azure marsh]

Suggestion for @tough fiber

RVC v2 Tutorial for Paperspace

💡 Click on the button below to open the guide.

Credits: LollenApe

[odd shale]

Además que parece que se te olvidó que el comando -paperspace existía, verdad?

Que eficiente eres. 😐

[odd shale]

El gato te comió la lengua?

[broken sky]

is this working or na ?

i dont get any sound dunno if its a driver issue or its an error

[trail trout]

damn if someone once again suggest me to install applio i swear i will insult him

removed that shit

[pastel oak]

it works yea so maybe your input or output is wrong dunno

[broken vigil]

Has anyone encountered this when training model in RVCv2?
UserWarning:
"Grad strides do not match bucket view strides. This may indicate grad was not created according to the gradient layout contract, or that the param's strides changed since DDP was constructed. This is not an error, but may impair performance.
grad.sizes() = [64, 1, 4], strides() = [4, 1, 1]
bucket_view.sizes() = [64, 1, 4], strides() = [4, 4, 1]
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass"
What to do and how do i fix this?

https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/1678 Found that blaise-tk was asking about this problem on RVC GItHub page but seems like nobody have the answer. Maybe you guys know? Also https://github.com/pytorch/pytorch/issues/47163

[rare goblet]

Ayo? @broken vigil level 1 !!!

[pastel oak]

Try inputting
Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass"

as a search result on this server across all channels - theres a few results, maybe one of those has it answered

Or any other part of the message, not sure

[zinc anchor]

At least you know it's not an error

[broken vigil]

Ok i'll try, thanks

[zinc anchor]

Looks like it's not up to you to fix but it should be taken care of by the dev

[broken vigil]

You want to tell that everyone is facing the same problem? Or it's just some specific case

[zinc anchor]

Look at the grad strides "array" (I think it is), which is strides(), and the one for bucket view, which is the other strides(); one has [4, 1, 1] inside of it and the other one has [4, 4, 1], it's not set perfectly because they don't have exactly matching values

I don't really know what I'm talking about in this case, just trying to extrapolate something from that warning

Don't worry about it

[trail trout]

what is the best way to remove harmonics using uvr5? first denoise and after remove instruments or first remove the instruments and after denoise dechorus etc.?

this option is doing absolutely nothing

[broken vigil]

Yeah i know
grad.sizes() = [64, 1, 4], strides() = [4, 1, 1]
bucket_view.sizes() = [64, 1, 4], strides() = [4, 4, 1]
But I'm just wondering if everyone is facing the same issue. And if this only affects performance but not the quality of model then I'm fine

[zinc anchor]

What do you mean, guitar harmonics?

[trail trout]

human harmonics im going to crazy and this time i have little harmonics on song but the export is like shit

[zinc anchor]

Vocal harmonies, background vocals?

[trail trout]

vocal

[zinc anchor]

Try the BVE model

[trail trout]

stand alone?

should i first split and use bve?

[zinc anchor]

Should be in the downloadable models

Haven't used it in a while

Wait a sec

[trail trout]

i did look at it but couldnt also find it

[zinc anchor]

UVR-BVE-4B_SN-44100-1

[trail trout]

oh i have that one

[trail trout]

but should i use stand alone like this

i did combine models everytime working on famous singers but not on this song i have i dont know why

i have to use minimum 2 models

[zinc anchor]

I don't use it in ensemble mode

[rare goblet]

Ayo? @zinc anchor level 41 !!!

[trail trout]

vr architechture?

[zinc anchor]

Yeah

[trail trout]

aggresion should be high?

[zinc anchor]

Try 5 first

[trail trout]

the page mvsep is separating good but not vocals, it can take the instruments very clean

this one you mean? this did nothing on my side

i mean i dont know what it does? if i enable it same result

[zinc anchor]

You can use the full vocal track you isolated with MVSep and BVE separates lead and background vocals

It's not perfect but better than nothing

[zinc anchor]

This with splitter mode disabled

[trail trout]

with that model you suggested for example

the beat is leaking

it is not even separated i think

it sounds like it did some eqo balancing but very bad

look the results are similar to this example: https://mvsep.com/result/20240415071052-96823c53a5-n-a.mp3

[zinc anchor]

The difference between vocal split mode on and off: with it off, you use BVE as a standalone model, by inputting the isolated vocal track (no instrumental) and you get lead and background (although they're called "vocals" and "instrumental"), with vocal split mode on, the lead/background separation is done after you have separated vocals and instrumental with a model like Voc_FT or BSRoformer

[trail trout]

i do extract the vocals almost clean but the chorus and harmonics are breaking the output

i tryed the aggression 100 and still have harmonics

even the harmonics are not that much

[zinc anchor]

Wow

Have you tried the settings I showed you

?

[trail trout]

this one is what im trying now

combining with my combination

[zinc anchor]

You don't have to use BSRoformer here, just a model that isolates vocals and instrumental, like Voc_FT for example

[trail trout]

i was not even able to find bsformer model on uvr5 actually so it is not posible for me to test that one :d

[rare goblet]

Ayo? @trail trout level 6 !!!

[trail trout]

this is my combination

[zinc anchor]

These should be the outputs

[zinc anchor]

Yeah I forgot it's only in a beta version for now

[trail trout]

yep looks like again ai shit is not doing my settings

i dont have it like yours

[zinc anchor]

Try using just one model instead of ensemble

I suggest Voc_FT if you don't have a strong Nvidia GPU

[trail trout]

yea but that lead only things i dont have?

even if i try it is not doing it

have the same setting with you

[zinc anchor]

You are using ensemble mode here

Maybe vocal split mode doesn't work with ensembles

[trail trout]

with that i have zero beat leak but okay to know if it is working i do exactly what you do now

[brittle wing]

How do I get some clean lead vocals to inference cause anytime I inference in applio there's some noise after I convert a built in MDX noise

[trail trout]

@zinc anchori dont touch this right?

[zinc anchor]

Yeah you can use default

[zinc anchor]

Wdym, you have noise in your results or in the vocals you want to infer on?

[brittle wing]

From the vocals themselves after I filter them

And I know I Denoised the dataset before training

[zinc anchor]

You said it's MDX noise so you used a MDX model for filtering

Do you use UVR?

[brittle wing]

Yes that's wrong

[zinc anchor]

I think it has an option for denoising outputs

[brittle wing]

I'm using a website

Mvsep X-Minus

[trail trout]

so it did removed the harmonics almost but now let me tell you what it did

[brittle wing]

But they don't do me any favor

[trail trout]

beats was completely gone first

after removed harmonics, the beat came back

.d

[brazen gorge]

Hi ÇılgınFlix, welcome to .fmbot.
To use this bot you first need to add your Last.fm account.

Please use the </login:928593006293037113> command. The bot will then send you a link so you can connect your Last.fm account.

[zinc anchor]

@odd shale You know ways to remove MDX noise?

[brittle wing]

They just make everything worse

[trail trout]

this was what i did btw

[zinc anchor]

You're using the right settings now, so it's up to stuff like the quality of the models or some advanced settings you can play with

[zinc anchor]

Yeah this is what I meant

Any options that don't involve UVR?

[brittle wing]

Uh wait there's an MDX net colab with deniise option available but there's still some music residue left in lead vocals.

[trail trout]

i dont have this option

[brittle wing]

Nice

[zinc anchor]

If you have a decent Nvidia GPU you can download the InstVoc model (there's a version 2 which should be better) which is already better than Voc_FT

[odd shale]

You're on Settings, not Advanced MDX Net Options.

[trail trout]

oh i see

thanks

[brittle wing]

Thanks

[zinc anchor]

Np

Teamwork

[odd shale]

right click, then Advanced MDX Net options

[trail trout]

yea i found it

but it does not lets me select denoise model? is it okay?

[odd shale]

Show your settings on Denoise Output

[odd shale]

Actually it seems like you already have it selected.

Look.

[trail trout]

yea but the noise still persist

it shouldnt on this audio

[zinc anchor]

Sadly BVE is not perfect

Depends on what noise you mean

[odd shale]

Send me your audio via DM to see if i can denoise it using X Minus pro.

[zinc anchor]

Vocal harmonies that are still in the lead vocal track?

[trail trout]

trying now

[odd shale]

Cilgin was trying to isolate a Turkish song.

It got a hellish amount of layered vocals, backing vocals and harmonies.

[odd shale]

I've just suggested him to use songs that didn't had too much of these

[brittle wing]

Final answer

Overtrain or not

50k?

[open mountain]

Hi I'm new can someone recommend good way to make an ai cover with custom models on mobile

[lucid umbra]

how do i keep my model private if i want to use it

[zinc anchor]

Oh yeah np

I kinda got busy

[odd shale]

Don't upload it anywhere ig.

[lucid umbra]

if i use it in colab will it come up on the database

[rare goblet]

Ayo? @lucid umbra level 1 !!!

[zinc anchor]

A quick way is to use weights.gg

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

[lucid umbra]

on the old colab there was a "enable private model" option

so i dont know

[rotund prawn]

I am using rvc Disconnected

what pretrained model is the best?

OV2 or italla or snowie or original?

[zinc anchor]

I wonder if that old spreadsheet is still up

[zinc anchor]

It depends on what your needs are

If you have a 10-20 minute dataset (or longer) the original is still the best

[rotund prawn]

almost 10 min

[zinc anchor]

Snowie is the russian pretrain and Itaila is the italian pretrain

[rotund prawn]

7 minutes

I see

[zinc anchor]

These and OV2 are mostly good for clean datasets under 5 minutes

[rotund prawn]

I got it

[rotund prawn]

I did it as OV2

THANK YOU

[zinc anchor]

You're welcome

[wooden pasture]

my training doesnt appear on tensor?

[pastel oak]

This usually takes a few minutes. If its been more than few minutes, click on "refresh" on the top right and make sure your voice model actually shows up on the list on the left side

[wooden pasture]

ill give it some time and check

[rare goblet]

Ayo? @wooden pasture level 3 !!!

[pastel oak]

That should be enough time

Refresh on the top right

[wooden pasture]

brother i hate doing this shit

nothings there

😭

whatyeever i do i cant get it to work lol

[pastel oak]

I only once had this issue and I somehow fixed it by deleting the (in this case) Evie1 folder in logs, restarting RVC and the training

Maybe I also deleted the tensorboard file to reinstall it.. but since your voice model is showing up, it shouuuld be working

Wait

One random question

[wooden pasture]

uh oh

[pastel oak]

Can you click on this lol

[wooden pasture]

right

[pastel oak]

just realised the graph is at 0.0 - 1.0

[wooden pasture]

i have some blue lines

[pastel oak]

😂

[wooden pasture]

lol

mate

im so losty

lost

i just wanna make funny ai songs

[pastel oak]

Yes, now everythings working. Now you just wait 👏

[wooden pasture]

w

untuil it hiots a low point?

i cant type i slit my finger open and have a plaster

[pastel oak]

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

[wooden pasture]

okok

thank you for all your help

the opchs are still rly slow tho even with my gpu

i dont get it

[pastel oak]

How long is your dataset, and what batch size did you use?

[wooden pasture]

8 batch

and the audio i used is 17 mins

in 1 audio clip

[pastel oak]

30 seconds for 17 minutes seems fine imo

I forgot what numbers I gave you last time we talked about this

[wooden pasture]

30 seconds per epoch

[pastel oak]

I think I said 1:30 minutes per epoch on my 3060ti for 30 minutes dataset

So your numbers seem to be back to what they should be

[wooden pasture]

ok good

and i set itt do do 1000 epochs

but i might cut it short?

depending on the grtasph

graph

[pastel oak]

Correct

[wooden pasture]

ok

thank you

[low cliff]

why rvc disconnected not working?

[wooden pasture]

this normal?

[low shard]

What error are you getting ?

[pastel oak]

One person has reported that the fix to this was reinstalling their nvidia driver. Honestly with the many GPU issues you've had, it may be worth reinstalling the proper GPU drivers. Or if you never did that before.. now is the time

[brittle wing]

is Mainline compatible with amd gpus?

[pastel oak]

There is an AMD binary release, yes.

[brittle wing]

would that be "RVC1006AMD_Intel.7z"

[pastel oak]

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

Yes

[brittle wing]

thank you

[low cliff]

Im getting a no GPU download and it won't let me download the index flies

[brittle wing]

i was using rvc gui this whole time

[pastel oak]

After that you run the file called go-web-dml. In general everything with DML

[brittle wing]

thank you

[low shard]

wdym no gpu download? did you finish your free daily gpu?

[brittle wing]

how do i get perm to send images? i need to show the error im getting but cant send the image 😭

[pastel oak]

Spam a few messages until you get Level 1 on the server

[lucid umbra]

is it still a thing?

[brittle wing]

ooo

ok

[zinc anchor]

Not that I know of

It was linked on old Colab notebooks from last year

[lucid umbra]

ohhh

so if i use my model it wont come up

nowhere

[pastel oak]

How was your day good person

[zinc anchor]

Probably not, definitely not if you don't have the option to make your model private

[brittle wing]

good

what about u

[zinc anchor]

Weights.gg made it obsolete

[pastel oak]

I am terribly sick but living somehow, thank you

[brittle wing]

how

many

messages

do

i

need

[pastel oak]

Perhaps it gives more points if you send the messages in some time gaps, so I will on purpose type for a longer time to waste your time staring at the "Shad is typing" screen.. and now you can type something back to me in one message, a longer message pls

[brittle wing]

the way i got timed out for 5 minutes and warned for spamming 😭 anyways does longer messages gives more points? how many points do i need to level up, hope im getting it after this message, i dont know what to type anymore help

[rare goblet]

Ayo? @brittle wing level 1 !!!

[brittle wing]

OOOOOOO

[zinc anchor]

What you're using is an old notebook

-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

[zinc anchor]

Check these out

[brittle wing]

ooh ty

which one is the best?

[zinc anchor]

For inference I recommend Ilaria RVC

For training RVC Disconnected

[brittle wing]

okk ty!!

[zinc anchor]

AICoverGen is also good for AI covers

[dusty rampart]

old ass colab

use the updated ones

[spark citrus]

im just doing this to remember later

ignore

[low cliff]

rvc disconnected is still not working

[rustic trellis]

how do i get my ai recordings to stop having the weird effect on them:

[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

[brittle wing]

---

ModuleNotFoundError Traceback (most recent call last)
<ipython-input-4-ee1b7656f787> in <cell line: 4>()
2 #@markdown Link the URL path to the model (Mega, Drive, etc.) and start the code
3
----> 4 from mega import Mega
5 import os
6 import shutil

ModuleNotFoundError: No module named 'mega'

---

NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

To view examples of installing some common dependencies, click the
"Open Examples" button below.

How to fix

[low shard]

Which colab are you using

[brittle wing]

https://colab.research.google.com/drive/1Gj6UTf2gicndUW_tVheVhTXIIYpFTYc7?usp=sharing

[rare goblet]

Ayo? @brittle wing level 1 !!!

[low shard]

Outdated colab

Use the ones inside the cloud folder inside the rvc folder of our docs https://docs.aihub.wtf

[brittle wing]

send a link ?

[low shard]

There are different colabs based on your needs, please check inside the docs it would be better as theres written the lros and cons

[brittle wing]

guys

why is it when i try to make a cover it says this?

even though i have the song vocals and model

how do i fix

[rustic trellis]

@helpers

[thin stump]

The model that you're using is low quality. It's artifacting

[rustic trellis]

how do i fix?

or should i ask how to prevent artifacting?

[thin stump]

Is it yours or someone elses? Because you can't fix their models if you don't have their dataset

[rustic trellis]

it is mine, yes

[rare goblet]

Ayo? @rustic trellis level 2 !!!

[rustic trellis]

should i retrain it?

[thin stump]

I would read the guide here then retrain https://docs.aihub.wtf/rvc/resources/datasets/#datasets\

[rustic trellis]

im going through my dataset of the accapellas i have and one seems to have the bass clipping it a little bit and it sort of sounds like the bad vocals, should i delete that and retrain?

[thin stump]

Absolutely, delete the parts that sound bad but make sure it doesn't create a popping noise when you do that

[rustic trellis]

that gives me only about 3 and a half minutes of training, is that ok?

[thin stump]

I would say use the original pretrain since OV2 isn't suitable for singers

[rustic trellis]

what is ov2?

[thin stump]

It's for short datasets that is 6 minutes or less.

[rustic trellis]

oh ok

so i should do atleast 6 minutes to sound decent

[rustic trellis]

also how many epochs do you reccomend for about 6 minutes of dataset

[bronze maple]

What is the solution to remove back vocals, behind an acapella?

I was told UVR Karoke, but for me it doesn't really work

[rustic trellis]

use x-minus

[bronze maple]

I'm not talking about the instrumental, I'm talking about the choirs behind a voice

[brittle wing]

Uvr Kareoke is the best option, though you can try using mdxnet 468?

I think that’s what it’s called

[bronze maple]

how do I go about perfecting the result?
I separate the acapella from the instrumental directly with karaoke?

[wooden pasture]

im training and it doesnt say stop

[bronze maple]

where is he ? mdx 468 I can't find it nor in the UVR download list

[brittle wing]

under mdxnet there should be some models with like, 3 numbers in the title, in the downloads area

[brittle wing]

Yeah

[bronze maple]

that's it ?

[brittle wing]

Oops no, hold on haha

Can you show me the options?

[cobalt pilot]

When I try to upload a model on huggingface it just gets stuck to hashing or at the end of the upload, what do I do?

[brittle wing]

Is your connection good? I believe that may be a huggingface issue. Try again a little later maybe?

[cobalt pilot]

My connection is pretty good and stable but alr, I'll try again after a while, thanks

[brittle wing]

Np!

[bronze maple]

This is also an option for extraction, what do you put in?

[brittle wing]

guys

[restive violet]

where can I find my ai cover file after its generated in covergen

?????

[ornate wasp]

there should be a folder called song output

sorry its song_output

[restive violet]

its in drive?

[ornate wasp]

itll be inside of the main aicovergen folder

which will be on the drive you installed it at

[restive violet]

what does this mean

oh wait

OMG I DID IT

[ornate wasp]

thats the name of the folder fo rthe song

for the song*

[restive violet]

And how do I download the file

[ornate wasp]

wait are you running it on collab?

[restive violet]

oh wait I already did it thank youuu

[ornate wasp]

you are very welcome! happy I could help, feel free to ping if you need anything else

anyone get rvc working with tensordock/jupyter notebook

burning all the gpu time I paid for just trying to get it running

[wooden pasture]

still happening after re installing everything and updating drivers

does this do the same as stopping the training

[ornate wasp]

itll stop rvc entirely

which would also stop the training

[wooden pasture]

ok ty

and then just do the next steps after?

[ornate wasp]

yeah I would do that

[rare goblet]

Ayo? @ornate wasp level 3 !!!

[lament blade]

the huggingface ones don't work, something about an invalid response

[wooden pasture]

when i stop my training there isnt the file in here that i need to move. any help? @ me or dm me

[worn prawn]

here i download the real time voice changer

[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

[zinc anchor]

Yeah I'm not sure all of them work

Usually duplicating the space is useful

[crude dome]

Hi there. I have a question. How to record something from rvc? I'm not wanna to make cover, just to record something from rvc

[zinc anchor]

You can record your audio and then use inference on it

[crude dome]

Ok, and how to do that? I even don't know how to record my regular voice. I'm total noob Xd

[zinc anchor]

Are you on your PC?

[crude dome]

Yea

[forest token]

Hello everyone! I noticed this amazing project and i wanted to try it!
But for some reason i am not able to upload or select any voice!
I'm running RVC on Google Collab Pro with T4 GPU and High Ram but i get the error:
Frequent errors occur. Please check if the model of the framework being targeted is loaded.
and the upload process freezes on 94.5%

any idea?

[light root]

In the ai gen hugging face space should I upload the zip file or index file

[zinc anchor]

See if you have a default app like Voice Recording that can do it

Otherwise you can use something like Audacity to record audio from your mic and then export the result

[crude dome]

K

[crude dome]

Like what?

[zinc anchor]

I just typed "Voice Recorder" in the Windows search bar on the bottom left

Looks like it's a default app

[crude dome]

Well it seems I really don't have it

So I think the only choice here is Audacity

[zinc anchor]

You can also use one of those online recorders, or you should have the option to record audio from mic on a Colab notebook like Ilaria RVC

[crude dome]

Got it

[low shard]

What google colab u using?

[forest token]

The one provided on github

[low shard]

github? Wdym send

we have our colabs on our docs https://docs.aihub.wtf

your prolly using an outdated one

our colabs are inside the rvc folder inside the cloud folder inside the docs

[forest token]

Hmm yeah, perhaps. I will check it in the morning! Thanks man!

[low shard]

yw

[zinc anchor]

No, the only option is to sing it yourself

I remember seeing a de-crowd model on x-minus

Maybe you can experiment with models from MVSep

Bandit Plus?

There's also one on MVSep

MVSep Crowd Removal

[pastel oak]

You can also try Crowd HQ 1 by Aufra33

on UVR

[tough fiber]

yesterday i trained my data on 32k sample now it says this when i try to continue training

and yea i choose 32k but idk

brooo wtff

which one is true i tried both

[rare goblet]

Ayo? @tough fiber level 6 !!!

[tough fiber]

new error lol

[pastel oak]

Maybe a drums remover could work if it sees the clapping as drums?

[pastel oak]

did you try a different model name, also toggle between v2 to v1 and back to v2 and then choose sample rate

then preprocess and all that stuff again and continue training with the G and D path

[tough fiber]

if i write there different model name it will continue

last epoch is 230 so it will start from 230

even i write there new model name

[pastel oak]

Yes. As long as you have the G and D path to continue, the name doesnt matter

[tough fiber]

okey lets try

but now

theres no sample error

[tough fiber]

im just getting this

[pastel oak]

I dont know that error off the top of my head

Maybe restart rvc and try again

If not, look for the error code on github or wait for someone else to help

[tough fiber]

tried it same -*-

[tough fiber]

yup ill try this one i guess

[tough fiber]

anyone help

[slow sentinel]

anyone know how to use the mergelab?
I want to merge voice files from different PyTorch Types

[proper shale]

ya mean like, different AIs? dunno if thats possible

[agile osprey]

sm1 help :

when my mic is on if the other person speaks they hear their voice in my voice model kinda echoing idk why

is it a problem with my mic

or do i need to fix sm settings

[proper shale]

enable sup2 + increase s thresh

and make sure you're using a headset or anything similar, so there's no mic bleed

[agile osprey]

okay

ahem sad i cant send images

[proper shale]

if you get lvl 1 you will be able

[agile osprey]

ah okay

my s.threshold is 0.001

echo and sup2 is on

[proper shale]

did that solve anything?

[agile osprey]

i gotta ask sm1 to speak while my mic is on

[rare goblet]

Ayo? @agile osprey level 1 !!!

[agile osprey]

this gud ?

[proper shale]

yeah

btw you're using the AMD version and ya didn't export the model to onnx

it should be extremely slow as it is

[agile osprey]

uhh what

i dont understand

[agile osprey]

the delay is like 5 sec

is that slow

[proper shale]

yeah

definitely

[agile osprey]

oh

💀

which one do i install then

[proper shale]

https://rentry.co/VoiceChangerGuide#uploading-models-amdintel-gpus-and-fixing-its-laggy-issue

read this part

and ignore the rickroll istg 💀

[agile osprey]

and the echo thing is still on

[agile osprey]

😭

[proper shale]

that might be the headphones... mic bleed 😭

[agile osprey]

ah thought so

my bluetooth earphones got a shittier mic so cant even use em

do i switch to laptop mic

[proper shale]

if the laptop mic doesn't have mic bleed issues... sure

id say try out everything at your reach rn

[agile osprey]

how do i change frm pytorchrvc to onnxrvc

[proper shale]

again, this section explains it

you just upload the .onnx file as the model and you're done

[agile osprey]

i dont hv one tho ahem

its just pth and index

[pastel oak]

• upload pth file
• click on big "export to onnx" button when you select the voice model
• it generates onnx file
• you overwrite the voice model slot with onnx file

[pastel oak]

on this pic on the top right is the export to onnx button in case you cant find it

[agile osprey]

ok bet

[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

[agile osprey]

ok the echo prob cuz of my mic

[rare goblet]

Ayo? @agile osprey level 2 !!!

[agile osprey]

gotta get a new headset or sm

[pastel oak]

I havent fully read up on your conversation, but is the issue that the mic picks up your headset audio thats being played due to it being very close to your headphones or was it something else

[agile osprey]

that was likee one of the speculation

idk why it happening

when the other person speaks

they can hear what they spoke in the voice model

[pastel oak]

do you mean the person speakings voice gets transformed into the voice changer and repeats?

[agile osprey]

yea

only for them

i cant hear it

mj said can be mike issue

mic

[pastel oak]

I mean I guess you could try:

• Put earplugs into your ear, and put your headset on your head aswell
• Put your earplugs as your main audio device, put the mic from the headset as your microphone

It is not ideal or comfortable but the earplugs would probably be quiet enough for your mic not to get picked up lmao

[tough fiber]

i need help how can i fix this

[pastel oak]

Or if its really just mic bleeding then add a noise gate / threshold to your mic or something through steelseries sonar or voicemeter

[agile osprey]

im actually using earphones 😭

and very cheap ones at tht

ill just switch

and try

[agile osprey]

okay

[pastel oak]

https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/129

google translate the last answer on the bottom

Or the entire convo

[tough fiber]

i saw this but i didnt understand what he means

with "mono"

[pastel oak]

convert audio from stereo to mono

[tough fiber]

it means data or

[proper shale]

aka. only one audio channel

take ur dataset & do that in audacity

[pastel oak]

If I may add - he is saying that this error first occured when he tried to continue training the model, which begs the question why it worked the first time but not when continuing.
Dont think it would have any major impacts on the voice model if he converts it to mono now and continues ,right?

[tough fiber]

is it work i gues let me try

[pastel oak]

Try it ye

[proper shale]

no not online audio converters nooo

[tough fiber]

yesterday i traned 240 epoch with no issue

today when i try it gives this

its weird

i didnt change anything

[tough fiber]

well i dotn have audacity -*-

[pastel oak]

Audacity is for free

[proper shale]

just install it

[tough fiber]

okey but i dont know how to use

i didnt use before

[tough fiber]

btw it seems already mono

:/

[proper shale]

might be rvc screwing up

frankly, idk

[tough fiber]

wait i think i dint im uploading data rn

ill try

[proud elbow]

decent headphones like Hyperx cloud 2 shouldn't leak to the mic. if needed, turn down the headset volume and set the mic threshold.

[slow sentinel]

damn unlucky some RVCs that are hell good are in different types in the merge lab and I cant merge em 😦

[rare goblet]

Ayo? @slow sentinel level 1 !!!

[tough fiber]

well i choose 32k sample and it says this

[finite galleon]

Maybe your audio files is 40k?

[tough fiber]

:)

[zinc anchor]

Damn

[pastel oak]

rvc just uno reversed on you

[finite galleon]

Do you have multiple files

in your dataset

[tough fiber]

nah

[finite galleon]

Whats the sample rate of your dataset?

[river sedge]

hey!

I get this error when I attempt to train my dataset in colab:

Traceback (most recent call last):
  File "/usr/lib/python3.10/multiprocessing/process.py", line 314, in _bootstrap
    self.run()
  File "/usr/lib/python3.10/multiprocessing/process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
  File "/content/Mangio-RVC-Fork/train_nsf_sim_cache_sid_load_pretrain.py", line 254, in run
    train_and_evaluate(
  File "/content/Mangio-RVC-Fork/train_nsf_sim_cache_sid_load_pretrain.py", line 384, in train_and_evaluate
    wave = commons.slice_segments(wave, ids_slice * hps.data.hop_length,
  File "/content/Mangio-RVC-Fork/lib/infer_pack/commons.py", line 49, in slice_segments
    ret[i] = x[i, :, idx_str:idx_end]
RuntimeError: The expanded size of the tensor (12800) must match the existing size (0) at non-singleton dimension 1.  Target sizes: [1, 12800].  Tensor sizes: [0]

Any clue what to do here or what the cause might be?

[pastel oak]

What GPU do you have?

Is it the start of training your voice model, or continuing training?

[river sedge]

GPU is RTX 3070
It was at the beginning of training the voice model

[tough fiber]

when i trained first was choose 32k

yesterday it trained to 230 epoch and i close my pc

today when i continiue training it gives errors as u see

[proper shale]

Oof

[finite galleon]

Yeah ik but like whats the sample rate information of your datasets audio

It needs to be the same as the one your choosing on the webui

[tough fiber]

i dont know actually

how can i check

[finite galleon]

Audacity

But i think you can check just the info on the file manager

If you are trying to train on 32k, convert your datasets audio to 32000 on an audio convertor

If 40k, 40000

[zinc anchor]

Oh yeah, absolutely

[tough fiber]

oh okay

btw should i delete these things

for data space

[pastel oak]

I looked on github and the server but couldnt find a conclusive answer to the problem

I'll mention everything I found and link you whatever I found, but in all honesty, I got no clue

https://discuss.pytorch.org/t/runtimeerror-the-expanded-size-of-the-tensor-must-match-the-existing-size/105523

Also

• GPU might have ran out of memory. Make sure you dont overload your GPU with anything
• something may have gone wrong on processing or feature extraction
• Make sure you dont have too much varying dataset files. Like if one is 20mb big and one is suddenly just 100kb or something. This can cause an issue
• Make sure you dont have save epoch set to a small number like 1, could be
• Try a different voice model name
• Make sure you selected sample rates and all of that properly. For example, click on v1, then back to v2, then select sample rate (sometimes it has a visual bug but who knows, maybe theres something more)

[zinc anchor]

Damn Shad

[finite galleon]

If you dont plan to use 48k you can prob delete the files with 48k

[river sedge]

Thank you! I will see what I can do

[pastel oak]

Dont delete