✨│ai-help

[analog obsidian]

😭

[simple ore]

lil longer

30-60 seconds?

[prisma grove]

aren't rvc samples supposed to be 10sec or less?

[simple ore]

you dont have a normal speech?

[prisma grove]

no? it's miku

[analog obsidian]

she is not a tts voicebank

she is from a vocal synthetizer meant for singing

[prisma grove]

I'll try to find a human cover of karma

[analog obsidian]

i find those so funny 😭

[prisma grove]

uvr goes brrr

it's taking a while tho

[analog obsidian]

mvsep bsroformer is better

[prisma grove]

I'm using replay, it's only got uvr

[analog obsidian]

[prisma grove]

and it didn't separate the vocal for some reason

[analog obsidian]

anyways, how slower the training is going to be with this lr and fp32?
i got my lowest point in 2 hours with this dataset using the default lr and fp16

[simple ore]

depends on gpu

[analog obsidian]

4060

[simple ore]

can probably even do BF16 with it

[analog obsidian]

u can train bf16 in rvc? wat

[simple ore]

I made a script to use accelerate and a chosen precision

well, just to try

[rare goblet]

Ayo? @simple ore level 18 !!!

[simple ore]

with AMD BF16 is super slow

[analog obsidian]

if u want i can try it but i have 0 coding knowledge so idk if i need to do some magic trick besides moving files

[simple ore]

it is for Applio

[prisma grove]

ok so

[analog obsidian]

so just replacing files right?

[analog obsidian]

if thats the case i can try it

[prisma grove]

jubyphonic cover → uvr (kim vocal2) → miku rvc

[simple ore]

nah, just dropping the file, installing one package, configuring it, then running the training from command line

[prisma grove]

this is the input audio

[analog obsidian]

sounds easy to me

[prisma grove]

she has no problem pronouncing "r" (like in the word "dark")

she is pronouncing "th" (like in "the") as "d" though

[simple ore]

that's because the trained set has no 'the'

so the inded looks for something similar and finds 'de'

[prisma grove]

no

she has many instances of the word "the"

this song (karma) is literally in the dataset

[analog obsidian]

i find funny last time i said applio training was slower for me but after i tried the new version now is faster than mainline

[simple ore]

newer torch, better performance

[prisma grove]

see here

[analog obsidian]

the ui still very slow for me and inference kinda buggy as well, so i decided to use applio for faster training and mainline for inference xD

[serene horizon]

So I started again in Kaggle, but when I go to train, it still says “error” in gui.

But in Kaggle says it’s working and is going through the epochs.

🤷‍♂️

[analog obsidian]

the only thing i want to get know is a stable fm that doesn't overfits like at 15 minutes of training

[analog obsidian]

[prisma grove]

I'm gonna try with a tts audio

the people at de party

it got it right once

to be fair it sounds like that in the original audio

[turbid pagoda]

Its been a while since I've used applio. but the voice model i want to use is not appearing after i copy and download the link does anyone know why?

[simple ore]

downloads have been broken since the website rework

download manually and unzip into logs

[turbid pagoda]

got ya thanks!

[low shard]

weird, @glad zealot do you may know why ?

[prisma grove]

here's a version based on a better tts

[rare goblet]

Ayo? @prisma grove level 6 !!!

[prisma grove]

imma try something in polish to see how that goes

it should do good considering there's miku v4 chinese in there too (the required sounds are pretty much all there)

[rvc folder]/assets/weights

[serene horizon]

Yes, just found it. Thank you.

[prisma grove]

nice, the rz sounds like sh but it's understandable, there's no rz sound in chinese either

the "y" is very good tho

I'm gonna make a 30 min dataset and see how that goes, I'll throw in some more v4 chinese too

actually I just noticed

is this any good?

also what's the deal with pretrain models?

[serene horizon]

Hmm. My 50 epoch training .pth I just did in Kaggle as a test works in inference, but in the resulting file, some lines are really clear/loud, while some lines very muffled/very quiet.

Any ideas what is causing this?

[vagrant carbon]

can someone help me... the voice ai is repeating after everyone in the vc on discord, is there any way to fix this?

[pastel oak]

Your mic is picking up your headphones
Lower volume ; move s.threshold/n.gate to the right, enable sup2

[vagrant carbon]

it worked and its better now, thanks!

[gusty scroll]

Yeah I noticed it too, I did several tests, I don't know why, I think it could be that the notebook needs to be updated. I've already contacted @glad zealot who created it. So hopefully it can be fixed.

[prisma grove]

#🔍│help-w-okada

[brittle wing]

Huggingface illaria rvc gives a error on inference @low shard

[low shard]

look at the top right, there should be an error message for a few seconds

it should say smt like 'GPU task aborted' or smt depending on the error

example:

[brittle wing]

what does that mean ?

[low shard]

it was just an example, you need to show me what error u get at the top right

do you get any?

[brittle wing]

yes

[brittle wing]

maybe i used too much gpu 😭

[low shard]

can you show me what it is?

you can just post a screenshot of the one at the top right

[brittle wing]

im so confused

[low shard]

click convert again

see if it shows

an error

at the top right

of the space

[brittle wing]

i will try later

[low shard]

common error messages at the top right are shown in: https://docs.google.com/document/d/1YbXcLFPaGjhOdG5NFkK3QrucCEpHZBwFUxkeMO8aB18/edit?tab=t.0#heading=h.2ahxfypqn13

[low shard]

i mean the error at the top right for some seconds, not just the one that appears in the UI (interface)

but alr

[brittle wing]

Is there a colab for Illaria RVC? @low shard

[low shard]

No

Ilaria RVC on google colab is broken

and the ilaria rvc zero version is made specifically for ZeroGPU of huggingface space

could you please

just show the error at the top right when u click convert

it should be like:

1. u click convert
2. after sometime u get an error at the top right
3. screenshot and show me that at the top right not the one in the middle of the interface

be careful cus the one at the top right dissapears after like 10 seconds

[brittle wing]

alr

[brisk grove]

What are these information's about dataset and batch size and 32k and else?

[solar parrot]

What happen to the colab that would automaticially do the vocal seperation and take the source straight from a youtube link?

[rare goblet]

Ayo? @solar parrot level 1 !!!

[pastel oak]

dataset is the audio of the person you are going to train a voice model on

batch size is how many samples of audio it trains at once, usually 8 is used assuming your gpu has 8gb vram. some specific use cases like if you got a very short dataset you may use 4, or very long using 16

32k is a sample rat example, reaching up to 32000 hz

[brisk grove]

So they are basically just information about the Voice model?

[pastel oak]

Well if you as a viewer are browsing for voice models, yes

[brisk grove]

thx for info

[simple ore]

batch size has little to do with VRAM

[molten merlin]

Hi guys, I'm new on training AI Voice models, what does the D, DUR, and G stands for on the outputs? (i.e. D_1000.pth, DUR_1000.pth, G_1000.pth)

[brittle wing]

hi i have an issue that i dont have friends

[brittle wing]

Does anyone know if its possible to download the voice models that people use on character.ai for the voice calls? Theres so many made of this one I want on that website but they're not posted anywhere so im just wondering if its possible through f12 or another program to download it to use in rvc

Ive literally scowered the internet and their discord for answers but theres nobody asking online and nobody answering in the server and im surprised with the amount of trained voice models there are on there that nobody is interested in ripping the higher quality voice models on there to use for RVC

[low shard]

UVR5 ?

[mild birch]

does anyone know how to work dis shi?

how do i get my voice sounding like carti

[low shard]

so in realtime for calls?

whats ur pc gpu?

[mild birch]

3080

like to record a song

[low shard]

so like record ur voice, then convert it with a model voice right?

[mild birch]

yeee

[low shard]

As you got a good PC, you can use RVC locally, you can choose between:

• Applio: A fork of RVC with some extra features like Applio TTS, same quality tho
• Mainline: The original RVC

if u want an easier way, there is ilaria rvc zero which is on cloud (remote good pc)

[brave garnet]

Ilaria RVC: CLICK HERE 🤗

Guide on how to use it: CLICK HERE 📝

Don't forget to thank Ilaria if you find it useful! 💖

[low shard]

but id suggest to do it locally as you got a good pc and will have no gpu limit

[mild birch]

imma be honest i have no iidea what all this means what even is a RVC? what is the first step i need to take??

[rare goblet]

Ayo? @mild birch level 1 !!!

[low shard]

rvc = Retrieval-based-Voice-Conversion

Its the program to convert the voice to one of a voice model, its speech to speech

[topaz nimbus]

basically free

[low shard]

locally = on ur pc, u download it and use ur pc gpu

cloud = remote good pc, used mostly for those who dont wanna download it locally or dont got a good pc

[low shard]

Yes its open source, the code is public to see

[topaz nimbus]

or if you are on mobile and don't have a pc

[low shard]

if u dont wanna go through any issue and just need a quick conversion, use this

[low shard]

yea, u can technically do it locally on cpu but its sooo slow, so its better to do it on cloud

[fading steeple]

Goat thank you bro

[mild birch]

thank you bro fr fr

makes alot more sense now

[low shard]

your welcome both of yall

[topaz nimbus]

mobile isn't even possible, the pc cpu and ARM (phone) cpu are so different you can't even do it in the terminal, at least without installing an APK

[low shard]

you are actually talking to the first person who used rvc on cpu lol

https://discord.com/channels/1159260121998827560/1289538710307602554

[topaz nimbus]

Hold on lemme see this

[low shard]

i did it mostly for the memes, but yes u can do it on phone cpu

ofc its NOT SUGGESTED as a phone cpu will be super slow

but its doable lol

[mild birch]

i got a r9 and 3080

im good right

[low shard]

yep ur good to do it locally too

[mild birch]

got it

[topaz nimbus]

wonder if 4gb will work too as my phone is from 2020

[low shard]

Prolly not, a test from someone’s phone from 2018 that has 6gb of ram was able to only inference 10 seconds

the super bare minimum is 6gb of ram

even if 8 or more is suggested

[mild birch]

so any voice in the channel named voice models is a rvc hugging face model?

[low shard]

like 80% of them, of course there are also GPT SO VITS models which are from another Text To Speech program which wont work on RVC

u can see it from the model post tag, rvc models got the rvc one, gpt so vits got the gpt so vits tag

but like 80% of them are rvc, so dw

there are thousands of rvc models

[mild birch]

ok bet thank you

[topaz nimbus]

i don't mind waiting longer

[rare goblet]

Ayo? @topaz nimbus level 1 !!!

[low shard]

i mean as that 2018 phone couldnt work to inference more than 10 secs, if u used more than that it would crash the app (nothing bad happened ofc, the app just closed)

so i really think if u try to do it the app would just crash, i mean u are free to try but 4gb are too low

Id suggest to use cloud, like ilaria rvc zero for inference

[brave garnet]

Ilaria RVC: CLICK HERE 🤗

Guide on how to use it: CLICK HERE 📝

Don't forget to thank Ilaria if you find it useful! 💖

[topaz nimbus]

i might as well just try at least

[low shard]

be careful tho

none of the phones was harmed (i even used it on my own which has 8gb) but ur phone seems too weak

[topaz nimbus]

if a linux ps4 jailbreak comes out for 12.00, well then I'm using rvc on that, besides, i can always try and remove non inference features

To lower the requirements

Might not be possible but worth a try

[slim wadi]

would a 4060 (8gb) be good enough to run rvc well at low delay? or atleast compared to a 1660 graphics card?

[topaz nimbus]

yeah

[low shard]

would actually be interesting tbh, i never seen someone run ai on a linux ps4, but a reminder that ps4’s gpu is AMD, the guide would be kinda different, check https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/blob/main/docs/en/README.en.md

[topaz nimbus]

i mean my version is too new but i could try

[low shard]

removing non inference features wouldnt really lower the requirements, it doesnt check ur pc gpu when u install it, like u could intall it and even if ur pc isnt good for training ud be able to inference as fine if its good enough for that

so u dont need to remove them as as long as u dont use training, running inference would be just fine

[mild birch]

where is the files and versions tap

im a little lost

[topaz nimbus]

which i am planning to do

[low shard]

yea its better than a 1660, for realtime voice changer in calls, I suggest to use wokada fork tho

[slim wadi]

well i use w okada, what is fork?

[low shard]

yea, u might be able to train as its a good amd gpu but goodluck

[low shard]

fork = modified version

The wokada fork has better optimizations

-rt

[azure marsh]

This interaction has expired, use the command /guides realtime if you wish to see it again.

[low shard]

its the first link

I mean ofc u can just use wokada too, the fork could help for performance tho

[low shard]

if u ever be able to do it let me know tho as its interesting

[slim wadi]

thank you so much, was running into performance issues while using it if i had anything else open due to my low memory

[low shard]

btw for the meanwhile u can use cloud (remote good pc) like ilaria rvc aero

i use cloud mostly too

[topaz nimbus]

I'll upload it on youtube if i manage to even get linux running, although I'm not risking using the dev update version update file

[low shard]

your welcome, might also need to check the settins suggested in the guide

[slim wadi]

looking through them rn, literally just found out i have a gtx graphics card

[mild birch]

yo

[low shard]

Lmao that would be funny, would also be cool if u show how u do it so if anyone else wants to do it they can

i honestly have a ps4 too but i dont think im ever jailbreak it as i use it for gaming

[low shard]

oh u have a gtx that explains

[slim wadi]

is it a website? or did i download the wrong version

[topaz nimbus]

i have a ps4 that i used to use for gaming but that i don't use as much for gaming, barely

[rare goblet]

Ayo? @topaz nimbus level 2 !!!

[fading steeple]

?

[low shard]

U mean the guide ? Yea its a site, if u mean the program, u are doing it locally (on ur pc)

[low shard]

like where to download the program locally?

It should be explained in the guides i sent u, there are blue texts that redirect u to the guides

[topaz nimbus]

@low shard do you know what this could mean? run-install.sh: 51: Syntax error: "elif" unexpected root@localhost:~/Applio#

[slim wadi]

so obviously its running on my pc, but is the UI on a website now? it looks the same as it was before now its just in my browser

[low shard]

Oh understood, i still use it for fortnite or multiversus, even if not as much as i used before i started to learn more about coding & ai

[mild birch]

like where can i access files and versions i dont know how you got to that page in ur ss

[low shard]

huh pretty weird, are u using the prebuilt wheel or manually build wheel of applio?

[low shard]

yea its running on ur pc, the ui is just on ur browser running locally

its not cloud (remote good pc), dw

[topaz nimbus]

i tried both, pre-built wheel and manually building wheel

[low shard]

which are u using? Could you please elaborate more?

[mild birch]

like which model? or method? or?

[rare goblet]

Ayo? @mild birch level 2 !!!

[low shard]

u also did sed -i '/sudo/d' run-install.sh Right?

[topaz nimbus]

ima try running that again if i didn't

[low shard]

Are you using Ilaria RVC zero or did you download applio/mainline on ur pc?

[topaz nimbus]

same result

[low shard]

btw for everyone im helping, if i disappear its bc im on school ipad lol

[low shard]

mm, lemme check if applio changed their .sh rq

[mild birch]

illaria rvc zero

[low shard]

you dont need to check the files, first of all did you download the model ?

[low shard]

yea seems like they changed the .sh 2 days ago lol, i need to check a new way as its very different than before sorry for the issue

will update it when i get home

[topaz nimbus]

nah you're good

[topaz nimbus]

ok

[low shard]

for the meanwhile id suggest you to use cloud

easygui on termux should work tho

[mild birch]

yes i downloaded a model that i got ffrom the discord

or was i suppose to download one from the site itself ?

[low shard]

yea u should download it from ilaria rvc zero itself

Check the guide pls https://docs.google.com/document/d/1YbXcLFPaGjhOdG5NFkK3QrucCEpHZBwFUxkeMO8aB18/edit?tab=t.0#heading=h.c5ea25jd74zp

[mild birch]

okay yes i just did that mb '

how do i unzip ?

[topaz nimbus]

is easygui compatible with arm 64 natively or do i need to do alternative steps

if not either way, then I'm fine with the cloud

[low shard]

you dont need to unzip it, its on cloud so it will run on a remote good pc not urs

u can find it in the models list when u refresh the models

[low shard]

yea its compatible, they are both RVC forks

its kinda the same steps as appkio except u have to do one line to upgrade pip, and can optionally upgrade gradio to have a better looking ui even if its not needed

[polar tendon]

Ah ok

[rare goblet]

Ayo? @polar tendon level 2 !!!

[polar tendon]

I'm on my main now

[low shard]

did u try it?

[brittle wing]

-colab

[azure marsh]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• 🆕 FaceFusion UI, by Nick088 Google Colab
• 🆕 FaceFusion NO UI, by Nick088 Google Colab
• 🆕 EasyGUI, by Rejekts Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

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

[haughty schooner]

what does
[Voice Changer] Pipeline is not initalized...
[Voice Changer] Waiting generate pipline.
mean?

[pastel oak]

Initial install failed, this can randomly happen.

Make sure to disable antivirus if you have one

Then, delete the folders "pretrain" and "model_dir", then you can restart the program which will reinstall the failed files

[radiant loom]

how can i change the model name?

[simple ore]

just rename them?

[radiant loom]

how?

when i right click on this, just dont figure the option to rename it

[simple ore]

win11 may be a bit retarded with right click options

[radiant loom]

https://tenor.com/view/kirby-l2p-gif-19544446

nice

[oblique bay]

Hello, I'm currently using a NVIDIA Geforce GTX 750 Ti (Also tried installing CPU version of the voice changer but my cpu is Intel Core i5-8400 at 2.8 GHz and its worse than my graphics card).

The voice changer I'm downloading off of:
(https://huggingface.co/wok000/vcclient000/tree/main)

Is it even possible to make it sound smooth voice with my graphics card? I want to see if this graphics card can even handle voice changer. If any tips or solution please let me know. Thanks

[knotty moth]

https://tenor.com/view/kirby-epic-embed-fail-epic-embed-epic-death-gif-25316408

[pastel oak]

GTX 700 series has no cuda cores so thats one point. overall very outdated gpu

I recommend you try the fork, its optimized better. Here its important you download the amd, intel, cpu version and then still select your GPU. The nvidia version will not work
https://rentry.co/Forkvoicechangerguide#download-amd-intel-and-cpu

[oblique bay]

Got it thanks.

[low shard]

Btw I fixed that, works now (tested by myself)

[gleaming river]

eh idk where to ask so..is it still the best ai song maker? if not, pls feel free to recommend me new ones, thanks in advance
https://colab.research.google.com/github/hinabl/AICoverGen-Colab/blob/main/Hina_Mod_AICoverGen_colab.ipynb#scrollTo=NEglTq6Ya9d0

[low shard]

yes rvc is still the best for ai covers, that's just a fork (modified version)

anyways the quality between of rvc and its fork is always the same, that one is just made to be easier especially for ai cover making

[gleaming river]

oh ic ic, thank you!

[tame mica]

are you

in a zip file

perchance

[low shard]

yw

[brittle wing]

what do i do with config json

i couldnt find dynamike voice model anywhere

[odd monolith]

Is there tutorial how to make your own rvc ai voice?

[brittle wing]

can someone walk me through making an ai voice modell

[fresh cairn]

has anyone got this working on Linux?

[low shard]

You can search rvc ai voice model at:

• #1175430844685484042
• #🔍│find-models
• https://weights.gg/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• https://applio.org/models
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• #1159289738314919936
• #1191429836321849435
• make it yourself with our docs guides https://docs.aihub.wtf/essentials/how-to-make-voice-models/

[low shard]

whats ur pc gpu?

[low shard]

dont worry, u need only the .pth and added.index for usin the rvc models

[low shard]

written guide yes, whats ur pc gpu?

[low shard]

for what? like wokada or rvc? and also whats ur pc gpu

be sure to be using rmvpe as pitch extraction,
and that your dataset is clean, and to also use the tensorboard

If i may ask, what language is it in ?

sure if u want to

Also, are u using the tensorboard ?

yea loss/g/total

[low shard]

btw hina fixed the bug

[mystic crest]

Guys does anyone know how to like clone a rapper voice to urs for example a rapper “vocals” but u turn to that you made it

[low shard]

so, using a made voice model to convert your voice in realtime for calls?

or u mean making the model first

and, whats ur pc gpu?

[mystic crest]

yeah basically

[mystic crest]

idk i got a laptop

[low shard]

You can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab -> GPU

[mystic crest]

i just wanna turn other vocals into mine

[mystic crest]

i think it’s ass

i got a asus

[patent pasture]

Is there any method to get exact pitch value by using some software?

[low shard]

could u tell me the gpu name doing what i told u tho?

just in case to check if u could do it locally (on ur pc) or need to use cloud (remote good pc)

[radiant loom]

XD

[patent pasture]

or we can only listen and do several attemts?

[radiant loom]

yes

ty

[mystic crest]

im not on it rn but its bad.

sometimes when i be recording on fl ts lags

[low shard]

alright then u should use cloud

For Realtime Voice Changing for Calls on Cloud (remote good pc for those who don't have a good one, YOU CANT DO THIS ON MOBILE):

• Google Colabs (4 hours daily of free T4 gpu, easy to use, require only a google account) :
  • How to use Original W-Okada's Voice Changer Google Colab
  • Modified W-Okada's Voice Changer Google Colab
• Kaggles (30 hours weekly of better GPUs, T4x2 & P100, harder to use, requires an account and a phone number)
  • Original W-Okada's Voice Changer Kaggle
  • Modified W-Okada's Voice Changer Kaggle

Its way better to use kaggle btw

[mystic crest]

i don’t even understand this gang like all im tryna do is the simple shit what i jus told u

[rare goblet]

Ayo? @mystic crest level 1 !!!

[low shard]

basically, there's this speech to speech program called RVC (Retrieval-based-Voice-Conversion)

Those models are used in another program called Wokada for realtime voice changing in calls

As your pc is bad, u cant do it locally (directly using it on ur pc), but you will need to use a cloud computing service (basically u are using a remote good pc to run the program for u)

Google Colab and Kaggle are cloud computing services:
Google colab = easier but short gpu time
Kaggle = Harder but WAY more gpu time

[mystic crest]

so what im tryna do

[low shard]

if u really want an ez way, try the first guide i sent 'How to use Original...'

however it wont have much gpu time, but the other links dont have a guide

[serene horizon]

Thanks. Do I need to do a new build?

[rare goblet]

Ayo? @serene horizon level 4 !!!

[brave garnet]

RVC Colabs and Spaces

⠀

Google Colabs

⠀
AICoverGen-WebUI
Useful for making quick covers, by Hina.

AICoverGen-NoWebUI
Useful for making covers, doesn't include a UI, by Ardha, by Eddy, Hina and Gdr.

RVC Disconnected
To train new voice models, by Kit Lemonfoot.

EasyGUI
The OG interface, by Rejects.
⠀

Made with love by Ilaria 💖

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 💖

[low shard]

I think you can just go to the kaggle link, do copy & edit and then do all from the start

[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 💖

[serene horizon]

Thanks.

When using Kaggle is there anything different I need to do to look at the graphed in Tensorboard?

[low shard]

It's all explained in the guide DW, u have to manually sync the graphs and it's always looking at the lowest point of loss/g/total

[brittle wing]

So i'm trying to train my first model based off a spoken (admittedly very short, 12 sec) sample, for use in Okada, but it just comes out like a buzzing sound? I tried 250 epochs but the tensorflow graph was just a single point. Anything obvious i'm doing wrong, considering I've seen models based off single sound effects

other models work fine, i just made a bad one

[low shard]

12 secs are very short, are you sure its even good quality?

And also u need to use the tensorboard (not sure if u just misspelled it)

[brittle wing]

sorry yeah tensorboard. Quality seems good to be, very clean and crisp, no background noise

[violet badger]

Does this still work? It just says error every time I try to convert

https://colab.research.google.com/drive/1mHKTGH5e3SAyDSBss1KtiYRbDdQzwSMs#scrollTo=pz3lRmDHl_JG

[brittle wing]

Found the issue! these steps fixed it, in case anyone else sees this

Applio recently updated their version of PyTorch, so recently trained models may not work (They have static instead of speech) in:

• Hugging Face spaces (like Ilaria RVC)
• W Okada Realtime Voice Changer
• Mangio RVC, Mainline RVC and their versions in Colab
  For this reason, if when making an inference you can only listen to static, you can convert it using this Colab:
  https://colab.research.google.com/github/Eddycrack864/PTH-Fixer-for-RVC/blob/main/PTH_Fixer_for_RVC.ipynb
  Btw Applio has added a converter in its new update, but if you don't use Applio you can use this Colab.

[rare goblet]

Ayo? @brittle wing level 1 !!!

[polar tendon]

I forgot to try it

[serene horizon]

Still says “error,” but is working.

[low shard]

No, Ilaria rvc is broken on Google colab #📰│dev-updates message

Use Ilaria RVC zero

[low shard]

Oh lol

[low shard]

Did u copy & edit from the latest of the mainline Kaggle?

Like not from the same notebook u made yesterday,but make a new one

[serene horizon]

Yes. I deleted the old one completely.

[low shard]

@glad zealot could u check again pls?

[glad zealot]

Ye me and someone else tested it again and it's working properly

Oh wait that's Ilaria RVc

Mainline kaggle seems to work no problem for me tho

[serene horizon]

Maybe it's just me! I'm not sure what I'm doing wrong there.

[glad zealot]

A wait I didn't pin the latest lol

[serene horizon]

I see it now. Thanks.

[glad zealot]

It should be version 3

[serene horizon]

Yes. Do you have any ideas why I keep getting an error when I try and use a custom pretrain?

[glad zealot]

Download or use it for training?

[serene horizon]

I'm adding it like this "!wget https://huggingface.co/Sztef/SingerPreTrained/resolve/main/f0G_SingerPreTrain.pth"

[glad zealot]

What error it says?

[serene horizon]

To use to train.

[snow hazel]

How long till I can use X-Minus again?

[serene horizon]

I'll just check.

[arctic spear]

hey I did the clone repository and install depndencies on the colab page and when I tried to start a server it said there was no such file or directory '/content/voice-changer/server'
/content

Is there somewhere I need to go to download these things?

[glad zealot]

What colab link are you using?

[arctic spear]

https://colab.research.google.com/github/w-okada/voice-changer/blob/master/Realtime_Voice_Changer_on_Colab.ipynb#scrollTo=lLWQuUd7WW9U

[glad zealot]

You can try using the pretrain downloader

It's the 3rd cell iirc

[glad zealot]

-colab

[azure marsh]

Suggestions for @arctic spear

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• 🆕 FaceFusion UI, by Nick088 Google Colab
• 🆕 FaceFusion NO UI, by Nick088 Google Colab
• 🆕 EasyGUI, by Rejekts Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

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

[glad zealot]

Use the modified voice changer

In this list

-realtime

[azure marsh]

This interaction has expired, use the command /guides realtime if you wish to see it again.

[arctic spear]

thank you

[serene horizon]

He's working now I updated to your new build.

[glad zealot]

Noice

[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 💖

[snow hazel]

How long till I can use X-Minus again?

Cause I keep getting this

[knotty moth]

enter the pretrain path as filename only (since it is in root directory as you see in imjoy) and it should work

[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 💖

[slim wadi]

can somneone link me the download for the virtual audio cable?

[polar tendon]

Because the files are already there on huggingface unlike in google colab

[polar tendon]

I think you mean this: https://vac.muzychenko.net/en/

[unique stratus]

is there any trick to generate output that is less likely to have voice cracks ? It looks like all the Gura models have the same issue with this one song. I don't know about the settings like Voice conversion options and Audio mixing options so I would really appreciate it if anyone can help me out in this part. Thank you

[polar tendon]

Turn down protect

If it's at 0.5 then turn it down

[unique stratus]

Currently it's at 0.33 for me (by default), I'm using the Hina mod one on Colab, should I turn it down to 0.2, 0.1 or 0 ?

[polar tendon]

Play around, see what what works

[polar tendon]

Ok so apparently my memory sucks and it's to avoid voiceless stuff and breath sounds

[unique stratus]

I was wondering why turning it down didn't help the crack at all

[rare goblet]

Ayo? @unique stratus level 2 !!!

[polar tendon]

If there's like heavy autotune in the song then there's i think no way to remove them

[unique stratus]

In case you wanna inspect the song it's named Legacy, a DMC 5 soundtrack

There's a part after 1:05 that always crack

Can I send the audio directly here

[polar tendon]

Sure

[unique stratus]

The rest of the song is fine but that ONE part near 1:07 keeps cracking no matter what Gura model I try

[unique stratus]

heres the original

[polar tendon]

Can i have the link to the model? I would spend my time at least trying to get raid of the voice crack

[rare goblet]

Ayo? @polar tendon level 3 !!!

[polar tendon]

Like the model used

In the audio files

[unique stratus]

https://huggingface.co/Gigrig/GigrigRVC/resolve/main/GawrGura_RVC_v2_Itaila_e200_s92600.zip

I actually tried various different ones actually, but all of them end up having the same crack in that one part

[polar tendon]

I'ma try different ways and see if it gets rid of any voice cracks

[unique stratus]

Thank you 🙏

[knotty moth]

ah, the voice cracks might be caused by backing vocals and harmonies, go separate it first

[unique stratus]

Thanks, I'll try again asap when this is finished

[polar tendon]

oh wait yeah i didn't even notice

[unique stratus]

Btw Sorry for my ignorance, I was doubting if there was an issue with the models, turns out it's the input's issue 🙏

[polar tendon]

It's fine, although i would recommend you inference the vocal files with something like RVC Zero instead of using the AICoverGen as that's faster

And then merge them back into one audio file

Until you're only removing the back vocals and not the instrumental too

[azure marsh]

💻 (Local) Realtime Voice Changer

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

[polar tendon]

I don't think we were talking about that

[unique stratus]

Sorry, does RVC Zero have a Colab notebook ? My laptop has too weak specs to run local models

[rare goblet]

Ayo? @unique stratus level 3 !!!

[fresh cairn]

RX 7900 GRE

[polar tendon]

It's a huggingface space: https://huggingface.co/spaces/r3gm/rvc_zero
It's almost the same concept as colab

[unique stratus]

Sorry, what do I need to put in the Model and Index sections ?

[polar tendon]

Oh you can also use URL-to-model, you paste in the model URL

[finite prawn]

preprocessing gets stuck here. tried restarting applio and my laptop but still stuck here

just realized what it could be, actually.... gonna test that theory... yeah nvm its not what i thought it was. has nothing to do with the model name replacing an already existing name

[polar tendon]

Might be a python problem, idk never worked with it locally

[knotty moth]

u should be able to do under ROCm, unfortunately I dont have hardware for that

[finite prawn]

oh i found out what it was. im too damn impatient lmfao. the progress bar wasn't moving but it actually WAS preprocessing. just took a while presumably because i have a huge dataset

[fresh cairn]

Okok I will try in morning then

[polar tendon]

Damn how much GPU RAM do you have

[finite prawn]

8gb

[polar tendon]

The google colab GPU RAM would run out, and it's 15 GB

[finite prawn]

wait they let you use 15gb vram over on the cloud side? that's so generous lol

[rare goblet]

Ayo? @finite prawn level 4 !!!

[knotty moth]

make sure you are using a folder containing wav files exported through audacity (if it doesnt work)

original rvc should work tho

[polar tendon]

Yeah but it's not unlimited use if you don't pay

[finite prawn]

already am using audacity exported .wav's

[low shard]

Use mvsep or cloud UVR

[polar tendon]

Or if the PC is powerful enough, run it locally

[low shard]

Huh? Even if there are files by other people u can still upload your model and audio

[knotty moth]

mvsep has some good exclusive models not being available for download

[snow hazel]

X-Minus

[polar tendon]

I mean UVR, if you only need vocal and instrumental extraction then UVR locally would work too

[low shard]

Well yeah, you should be able to, what are u looking for? Wokada or rvc

[fresh cairn]

The one that can be used for discord and such

[low shard]

So realtime voice changer for calls on discord, yeah wokada, there is a Linux part in the Wokada fork: https://rentry.co/forkvoicechangerguide

[knotty moth]

iirc try check deiteris repo for installing for ROCm

[low shard]

Weren't you talking about Ilaria rvc?

[fresh cairn]

What's the difference between them exactly?

[polar tendon]

No

[low shard]

I'm telling you alternatives,both of those got UVR models anyways

[low shard]

Between Wokada and it's fork? Fork means modified version, this one is more optimized in performance

[knotty moth]

there are also Kim's and unwa's melband roformer downloadable for UVR

[polar tendon]

Sorry got two replies confused

[rare goblet]

Ayo? @polar tendon level 4 !!!

[snow hazel]

Do they both have ways to seperate harmonies?

[polar tendon]

mvsep should be able to remove harmonies

[knotty moth]

melband roformer karaoke

[low shard]

Oh lol dw

[snow hazel]

On which one?

[rare goblet]

Ayo? @snow hazel level 11 !!!

[snow hazel]

?

Oh on mvsep

Which vocal type?

[finite prawn]

should it be considered normal for training with an hour-long dataset to take around an hour for every epoch? I have an RTX 4070 with 8GB VRAM, and all of the models I've trained before had pretty fast epochs (but they did have 15-20 minute datasets)

oops

my forgetful ass forgot i had the same issue and fixed it a long time ago

disable cache dataset in GPU

[knotty moth]

4070 laptop, not 12 gb one? should be enough for batch size 8

[finite prawn]

my settings were fine except for 'cache dataset in GPU' being enabled. encountered this problem before but i just forgot the solution

just disabled it and retrained and now it's running smoothly

[knotty moth]

never seen anyone recommend cache dataset

[finite prawn]

i just made an educated guess with my lack of computer knowledge that enabling it would be beneficial

[snow hazel]

Yeah I tried but it didn't do a thing.

[knotty moth]

skill issue, try UVR BVE or v2 in uvronline.app

[brittle wing]

@low shard I stil get the error in the illara rvc, i did not see the Gpu task aborted message so what is going on!?

[low shard]

Did you see any other message?

[brittle wing]

no

[low shard]

Don't you see anything at the top right for some seconds?

[brittle wing]

i only got the error messsage

[brittle wing]

no

[brittle wing]

it was just a white screen at the top left from my memory

[low shard]

Could you please screen record showing me u press convert?

[brittle wing]

yes

[low shard]

Yea pls do that so it's easier for me to understand the exact issue u are having

My guess would be that it's the GPU quota but you would get a "retry in.." with the time at the top right

So that's why I'm asking much for the error, cus you have to do different things based on the error you get

[brittle wing]

here

i waited 24 hours for the gpu quota

@low shard

[brittle wing]

i sent it to you

[low shard]

Before clicking convert, be sure it shows the waveform of the audio you upload which means the audio is successfully uploaded

[brittle wing]

ok

[low shard]

(the waveform is that thing shown in the image, it seems like you convert before it fully loads the audio)

Try waiting first for that to appear, and then click convert

[brittle wing]

alr

[brittle wing]

it still gave me a error message

[brittle wing]

now what?

[low shard]

Could you please send me the audio file and voice model download link in dms so I can try myself?

[brittle wing]

i give up

here

@low shard

[low shard]

yea its a problem of the voice model

[brittle wing]

how do i fix it?

[low shard]

it has the trained index, instead of the added index

[low shard]

did u make this voice model or found it?

[brittle wing]

i made the model myself

[low shard]

do you remember having an added index? you zipped the wrong file

[mild coral]

where is gpu

[brittle wing]

is it added?

[low shard]

oh idk why but in ur pc is called 'graphic processor' which is the gpu, that seems an integrated graphics gpu which is bad, meaning you won't be able to train voice models locally (on ur pc)

As you dont got a good PC, its better you use cloud (remote good pc) for training an RVC Voice Model:

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
  • RVCDISCONNECTED (no ui)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio by Vidal (UI)
  • Applio by Shirou (UI, no guide as of right now)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • Mainline (UI, No guide as of right now)
  • Applio (UI, No guide as of right now)

Google Colab = Easy but low gpu time
Kaaggle = Harder but MUCH Gpu time

[low shard]

you should have a file named like added_IVF169_Flat_nprobe_1_sahed_v2.index or similar to that

[brittle wing]

oh

[low shard]

where did you train?

[brittle wing]

google colab rvc v2 disconnected

[low shard]

go to https://drive.google.com/drive/u/0/my-drive (google drive), check for an rvcDisconnected folder, check for a folder with the same name of ur model (which should be 'sahed' from what i seen), and inside that u should find the added index

let me know if u find it

[brittle wing]

got it working now

[brittle wing]

its the wrong index file

[low shard]

yea u used a wrong one, but in drive u should find the added index

an example

u should download that, then zip it with the .pth and the model should be working now

[spiral notch]

Hi, I'm re-adjusting to using RVC after it's migration from Google, how do I change the octave the voice is singing in?

I realized I put a male voice on a female singer and he sounds.... Bad. Lol.

[low shard]

it's migration from Google
RVC is still on Google Colab, it's just that the other cloud computing service, Kaggle, offers better gpu times

Could u tell me what u are using?

[spiral notch]

I'm using the downloadable version off github, I was under the impression google took RVC down so it migrated lol!

[low shard]

Google Colab & Kaggle technically are not allowed, because it uses a WEB UI (except if u buy google colab pro)

But people just obfuscated/encrypted the code so its still usable all fine

What's ur pc gpu?

[serene horizon]

Is there a place where I can do inference convertions online like training in Kaggle?

[low shard]

you should be able to also inference in kaggle iirc

however, for inference the fastest one is ilaria rvc zero

[brave garnet]

Ilaria RVC: CLICK HERE 🤗

Guide on how to use it: CLICK HERE 📝

Don't forget to thank Ilaria if you find it useful! 💖

[cloud lance]

implementing RVC for an app. if anyone knows optimal block_time settings please lmk

[spiral notch]

My dad built the thing so I had to ask him, an AMD RX6600

[rare goblet]

Ayo? @spiral notch level 1 !!!

[low shard]

Btw You can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab -> GPU
(telling u just in case u ever forget lol)

However, for the Mainline RVC (which is prolly what u used) needs linux for AMD GPUs, for windows its better u use Applio (an rvc fork, modified version, the quality doesnt change, only the UI which is the interface): https://docs.applio.org/getting-started/installation#amd-gpu-support-windows

[spiral notch]

I mean, it works fine for me, and I've had no problem using RVC as I currently have it, I just dunno where in the WebUI the option to pitch up and down is, cuz rn my singer is rly high pitched.

[noble dawn]

Yooo

Anyone here expert at training in local ai

?

My stuff be not sounding good when I have a Nevida gpu 3060 Ti

[low shard]

Oh alr sorry,

You have applio or mainline?

[spiral notch]

I remember downloading it from a github link on docs.aihub.wtf but that's down at the moment I think so I can't see which download I used specifically.

filename is RVC1006Nvidia

[elfin dome]

hi guys ,. i just have now mmvc_client_v0.4.3_x86_64_win and this one need a ONNX models
how can i get ONNX models
thx u

[low shard]

Alr, that's mainline, i dont do it locally but should be this one

[spiral notch]

Ahhh ty!!!

[low shard]

yw

[stark wadi]

Have a question. Certain words sound like they are speaking in like a japanese accent. Would a pretrained starting model with english singers help with this? I'm using RVC mainline and the pretrained weight that comes with it and training off that. I assume that because this was made by a non native English speaker that some pronunciation issues are to be expected, because the initial training data was not English. Is there anything I can do to rectify this?

[simple ore]

if you use an index, then the voice features shift towards the voice model, if you dont use the index, they stay as in the original audio

[stark wadi]

The index is english. But some words, like hard G sounds trigger it.

I meant, the training data is english I meant.

[arctic spear]

Hey, I noticed that when I'm playing a game like Cs2, the voice changer glitches out more. Is there a way I can stop this problem?

[rare goblet]

Ayo? @arctic spear level 1 !!!

[simple ore]

limit fps?

[raw bridge]

hello

im trying to do a text-to-speech program with a cloned voice

[sterile sparrow]

can anyone help?

its not working for me

at all

[low shard]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

• Don't simply mention your issue, like "my rvc is not working".
• Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
• The more context, the better.

__**BE POLITE.**__ <:matsuripray:1159685390156967936>

• Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
• It's okay if you're frustrated, but don't take it into this server.
• Don't DM without prior consent.

__**BE PRODUCTIVE.**__ 🤝

• Don't ask for every little instruction. Put your own effort & test things by yourself.
• Don't ask to ask.
• Check if your answer is a Google search away/on our guides website.

If you've changed your mind about your message, feel free to delete/edit it.

[stiff moat]

How do I use the model i made? I remember there was a local website for it and now I can't find it

[low shard]

Whats ur pc GPU?

And do u mean using it for pre-recorded audios or realtime for calls

[stiff moat]

1650

[stiff moat]

im trying to use the model i made for an ai cover

although i think i've found the page i was looking for

thank you for replying though

[rare goblet]

Ayo? @stiff moat level 1 !!!

[low shard]

you would be able to do inference (using models) only locally (on ur pc)

but wouldn't be very fast, i would honestly recommend cloud (remote good pc)

[low shard]

what are u using btw ?

[stiff moat]

the collab

[low shard]

applio google colab?

[stiff moat]

yeah

i think so

[low shard]

I'd suggest you Ilaria RVC Zero, it's faster as its using ZeroGPU (A100, faster than Google Colab's T4)

[brave garnet]

Ilaria RVC: CLICK HERE 🤗

Guide on how to use it: CLICK HERE 📝

Don't forget to thank Ilaria if you find it useful! 💖

[stiff moat]

thank you

exactly what i was looking for

[low shard]

yw

[brittle wing]

I can't use Ilaria RVC mainline, I keep getting errors

-hf

[azure marsh]

<:huggingface:1179800228946268270> Hugginface Spaces

• UVR5 UI, by Eddy and Ilaria Huggingface Spaces
• Ilaria RVC Zero, by thestingerx Huggingface Spaces
• RVC⚡ZERO, by r3gm Huggingface Spaces
• Applio, by IA Hispano Huggingface Spaces
• 🆕 FaceFusion UI, by Nick088 Huggingface Spaces

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

[humble valve]

what is the best one to use for chatting i want it to use for a rpg character

[low shard]

Tell me what error do you get at the top right

It might be u finished the quota

[brittle wing]

I closed runtime

But

I can't convert

[low shard]

Oh right you meant the Google colab version

[brittle wing]

That's the problem

[low shard]

Yeah it's broken

[brittle wing]

Ouch 💔

[low shard]

#📰│dev-updates message

[brave garnet]

Ilaria RVC: CLICK HERE 🤗

Guide on how to use it: CLICK HERE 📝

Don't forget to thank Ilaria if you find it useful! 💖

[low shard]

Use Ilaria RVC zero instead

It's a ZeroGPU HuggingFace Space which has faster GPU than Google colab

[brittle wing]

I already inferenced with applio but I wanted to check how the results will be because it's slightly different with every Colab

[knotty moth]

it's not bad for inference, but u can use ilaria rvc for faster one

[brittle wing]

I'm more comfortable with Colabs but ok

[low shard]

The results shouldn't be different, all RVC forks have the same quality, may be some have different default settings

[brittle wing]

Sometimes the output sounds different

[low shard]

But the ZeroGPU space has an A100 GPU, it's literally WAY more faster than Google colab T4 GPU

[brittle wing]

In some more realistic

Idk

[low shard]

Tho RVC and all it's forks never changed anything in terms of quality

Are you really really sure? Might be a placebo

The only difference is basically the UI, interface

[brittle wing]

Okay where is the newest ai cover generator Colab it's too slow

[brittle wing]

It's not when I use harmonify

[knotty moth]

zerogpu one is easier tbh

[low shard]

You sure you don't wanna use Ilaria RVC zero?

ZeroGPU Is faster

And easier to setup ye

[brittle wing]

https://colab.research.google.com/drive/1X8YR4Ruv7zzY8YAMPTfC7hkxqT_d4Q5d here is harmonify

[brittle wing]

Nah I want

[low shard]

That was an old fork from @finite galleon iirc

[brittle wing]

And it still works

I use applio tho

It's the fastest for me

[brittle wing]

Someone said jammable gives more realistic results

[low shard]

Jammable is the new name of voicify, one of those scam sites that use RVC and paywall it

They pay people to promote their shit, used to do it alot on tiktok

It's just a scam, just like kits.ai

They ALL use RVC

[brittle wing]

Wait

Is it the same as any other

[low shard]

Yes...

Just like kits.ai

They pay people to promote their shit and make others use their site, that's prolly why u heard that lol

[brittle wing]

Then why did that one person tell me it's the best I mean I've inferenced the same stuff with applio and the one from jammable has better sound

[low shard]

Jammable is just pay-walled RVC tho, I don't really think there is any difference in the output

[brittle wing]

https://vm.tiktok.com/ZGdeXuXxS/ this is made with jammable.

[low shard]

If you really think there is, might be the audio wasn't cleaned well and in jammable they just cleaned it a bit more, other than that there is no other way as it's just RVC

[low shard]

Yup it's just RVC lol

There's tons of those ai covers site

At the end they all use RVC

[brittle wing]

But it sounds more realistic than what I get

Like I can't get such realistic results I even use the exact same models

[low shard]

I don't think you can import external RVC models to jammable, so that's prolly just based on the RVC model quality that you have used in applio

The reason why u can't import is not bc it's different, it's just like kits.ai

[brittle wing]

Idk

The sound is more realistic

[knotty moth]

it's just their models, prob comparable to the one in #1191429836321849435

[low shard]

Bc it used 2 different RVC models

[low shard]

Still RVC tho

[brittle wing]

The person I literally messaged and talked to literally has credited the models and creators!

[low shard]

U ofc need to get a good quality RVC model for having good quality conversion

[brittle wing]

I use the best ones

[low shard]

It's literally just RVC 😭

You are just comparing 2 different RVC models

[brittle wing]

Five of them are literally the same I use

See?

And the cover is still better than what I can convert

[knotty moth]

yea I'm thinking it's literally another kits.ai

[brittle wing]

They're the same ones I have

[low shard]

It is, those guys used to steal my fucking covers to promo their voicify shit

They used to steal covers on TikTok and put the voicify logo on it too

Bc people get paid to promo that shit

[knotty moth]

who is this bot @brittle wing ?

[low shard]

In the jammables site it doesn't show that u can import external RVC modela

[brittle wing]

So they have their own BTS models here?

[low shard]

Yeah, as u see on the jammables site, it doesn't let u import models, u can only train them on their own site for money lol

[brittle wing]

Then why did this person literally credit the creators with the exact same models I use

Do the developers steal models from here and profit off them for money?

[low shard]

Not really sure about that

But the models can't be imported any other way as I just checked

[brittle wing]

My two old models are probably here hahaha

They look like they got them from weights...

[low shard]

I remember there were some models stolen when they were called voicify so prolly yea

[brittle wing]

The models aren't that good but

[rare goblet]

Ayo? @brittle wing level 7 !!!

[brittle wing]

How do they sound realistic here

[low shard]

You need to properly clean the vocals

[brittle wing]

Which ones for inference or training

Cause there's no longer need for that Melband karaoke doesn't generate noise

[low shard]

Both, u need to have a good quality model trained using the tensorboard ofc

[brittle wing]

Nah I'm tired of training models my second one was good but I overtrained

I'm just lazy

[low shard]

Then rip

[brittle wing]

What-

But Melband karaoke doesn't generate noise when removing the Reverb it's not MDX or UVR architecture
Roformer is noise free

[low shard]

i mean that ofc you need a good quality model to generate good results, i seen u said u overtrained

[brittle wing]

I already got one that sounds the same

But isn't overtrained

[knotty moth]

"overtraining" based on tensorboard graph?

[brittle wing]

Don't ask me too much questions I'm not stupid I know how to train models

Well it was last year the dataset was shorter than what is supposed to be 750 epochs I probably looked at the tensorboard or I didn't I don't remember

Is RMVPE+ better than RMVPE? Harmonify uses that too

Ilaria gives better results than applio

Clearer vocals

But tends to decrease vocals' volume

[knotty moth]

it may depend on the model's peak and dynamic range, other than the volume envelope option

[brittle wing]

How do I fix that

Volume envelope should be default

[plush compass]

Is it okay to use Hubert while using a pre-trained model?

[rare oriole]

Google drive doesnt seem to like RVC

crap i cant send an image

but it basically said I executed code that's "not allowed" in the free tier

[cloud finch]

Whenever i do applio i get this

When i convert

it gives me nothing

and gives that error

[dusty plume]

Can someone when i try to convert vocals it gives me errors

[carmine hearth]

-howtoask

[azure marsh]

This command has been changed to !howtoask

[carmine hearth]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

• Don't simply mention your issue, like "my rvc is not working".
• Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
• The more context, the better.

__**BE POLITE.**__ <:matsuripray:1159685390156967936>

• Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
• It's okay if you're frustrated, but don't take it into this server.
• Don't DM without prior consent.

__**BE PRODUCTIVE.**__ 🤝

• Don't ask for every little instruction. Put your own effort & test things by yourself.
• Don't ask to ask.
• Check if your answer is a Google search away/on our guides website.

If you've changed your mind about your message, feel free to delete/edit it.

[dusty plume]

does anyone know how to terminate the .bat file after your done

[stone lintel]

anyone know whats causing a weird g/fm graph on tensorboard?
my other graphs look fine but this one bouncing around i dont think its a good sign

[sudden lintel]

i have no idea how to download the RVC can someone help me ?

[still phoenix]

what to do?

[brave garnet]

RVC Colabs and Spaces

⠀

Google Colabs

⠀
AICoverGen-WebUI
Useful for making quick covers, by Hina.

AICoverGen-NoWebUI
Useful for making covers, doesn't include a UI, by Ardha, by Eddy, Hina and Gdr.

RVC Disconnected
To train new voice models, by Kit Lemonfoot.

EasyGUI
The OG interface, by Rejects.
⠀

Made with love by Ilaria 💖

[glacial pollen]

FM graph will oscillate

as that's the most important characteristics-type metric for models

also, pro tip. Don't use " ignore outliers " or whatever was it called

instead, keep the graph as it is + smoothing at 0.2 to 0.6

Ignoring outliers serves practically no purpose in case of rvc

[glacial pollen]

To have it more stable in general, it's either you have to take care of samples' not being too diverse / too dirty or just increase the batch size

I've talked about it too many times so won't go full in-detail but, smaller batch_size = more noisy and non-stable graphs whereas higher batch_size gives you more stable and " flat like " graphs

[glacial pollen]

It is the same " under the hood", just the functionality differs a bit.
The only difference from what I know is that rmvpe+ allows you for range clamping or something of that sort.
Whenever you can, use the + version

[odd shale]

Haiii Cody.

[low shard]

Different types of RMVPE Pitch Extraction:

• rmvpe: A Robust Model for Vocal Pitch Estimation in Polyphonic Music, the normal version of the best pitch extraction, it's robust and not sensible to noise
• rmvpe+: has a pitch threshold, it limits the max and minimum pitch possible, basically deleting f0 values below and above certain thresholds
• rmvpe-gpu: Training ONLY, uses your gpu for the feature extraction process, using more gpu so making training faster
• rmvpe-onnx: Wokada ONLY, its a must for AMD Users who use ONNX models

Its basically the same as rmvpe except it uses a min and max pitch

[fleet fiber]

Hey, I'm the person u talking about lmao. I didn't steal models, i imported them on Jammable and used them and gave credits too lol

[rare goblet]

Ayo? @fleet fiber level 1 !!!

[brittle wing]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

• Don't simply mention your issue, like "my rvc is not working".
• Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
• The more context, the better.

__**BE POLITE.**__ <:matsuripray:1159685390156967936>

• Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
• It's okay if you're frustrated, but don't take it into this server.
• Don't DM without prior consent.

__**BE PRODUCTIVE.**__ 🤝

• Don't ask for every little instruction. Put your own effort & test things by yourself.
• Don't ask to ask.
• Check if your answer is a Google search away/on our guides website.

If you've changed your mind about your message, feel free to delete/edit it.

[fleet fiber]

Yes, u literally can import any ai model out there and put it on Jammable!!

[low shard]

since when ?

[fleet fiber]

Idk since when, i started using just like four months ago

[low shard]

its shit using that

its one of those scam ai sites

just like kits.ai

just use RVC, that's what every of those sites use, its Open Source

[fleet fiber]

It's not, go and try it urself? i tried kits.ai before and it didn't work that well but Jammable is so good like? the results i get are way better that using any other site

[fleet fiber]

applio is also good but it takes time

[low shard]

seems to let u upload models, but why would u pay for something u can do for free??

[low shard]

they aren't way better 😭

its literally just rvc

[fleet fiber]

cuz i have the money and its way easier 😭

[fleet fiber]

Ik, they are but the results are better, the quality is better

[low shard]

.. because jammable just runs RVC in cloud, its just like running RVC on a cloud computing service, jammable isn't running them on ur device

[low shard]

paying for something that is a literal scam and its available for free is crazy

rvc is literally easy to use

[fleet fiber]

it's not a scam tho, why u keep saying that? If the quality was bad and the results were bad then i wouldn't pay for it

[low shard]

you just contraddicted urself literally
I said the quality isn't better, and in the same sentence u said that you know but they are better 😭

[fleet fiber]

Ik, i use it too

[low shard]

because it is.. if i sell something that its free for paid, hows that not a scam

[fleet fiber]

I was replying to u saying that they're all rvc not this one

[low shard]

saying that they're all rvc not this one
Jammables is RVC too

[fleet fiber]

when the quality is better + easy to use + no time spent then it's not a scam to me

[rare goblet]

Ayo? @fleet fiber level 2 !!!

[fleet fiber]

That's literally what i said

[low shard]

first of all quality ain't better, about easy to use that's kinda a skill issue ngl

[low shard]

u said in the same sentence that "they're all rvc not this one"

[fleet fiber]

Again, the quality is better. ur just a hater atp and ik how to use rvc, i just don't want to waste my time for an hour or more making a 3min ai cover when i can make it in a minute or less

[fleet fiber]

Not this one meant that i wasn't replying to that message omg

[low shard]

the quality can't be better if they are both using the same program, its like saying an rvc fork has better quality than the other when they all have just the UI different

And about time, that depends on the device it's running on, ofc jammable is just cloud rvc

And it takes less than a minute for Ilaria RVC Zero running on ZeroGPU (A100)

[fleet fiber]

I tried applio, kits.ai and other sites but imo, the results i got from Jammable were better, it doesn't matter if it's the same program, the results were way cleaner and better to me. I use the same models, i use on applio but it's not the same results at all. on applio, the voices are not clean and they crack but on Jammable, the voices are smooth and clean for me

Also, on applio, the site always stops working so I have to start all over again, that's why i switched to Jammable

[odd shale]

Are you talking about applio colab?

[low shard]

it doesn't matter if it's the same program, the results were way cleaner and better to me.
That's literally placebo effect, its like saying the same calculator works better on ur device than mine lol

[low shard]

Also, on applio, the site always stops working so I have to start all over again
Pheraps you mean applio colab? that's not a bug, Google Colab is a cloud computing service, it cant give free gpus forever so ofc u can't let it run 24/7

Ilaria RVC Zero is already set up, and uses ZeroGPU from huggingface (an A100 is way faster than google colab's T4 GPU)

[fleet fiber]

Now, jr just talking lmao. Talkin about some placebo effect, c'mon now 😭 ik for sure that it's not a placebo effect, i used both many times over the year so yeah, the results on Jammable are better

[fleet fiber]

Yes

[low shard]

its the same program and you literally know it urself..

[fleet fiber]

Ik that, I'm talking about when it's gives those errors mid converting

[low shard]

its like talking to a wall atp

[odd shale]

Ofc that occurs because you exceeded the usage limit. That's how colab works.

Including the fact Google got very strict with RVC colabs.

[fleet fiber]

Never used Ilaria before, I'll try it and see

[low shard]

i think he means having to install it everytime ofc as u start a new session

[fleet fiber]

Yet, the results are better for me

[odd shale]

Yeah, i've understood that.

That was what i'm talking about.

[fleet fiber]

Yeah, ik that

[low shard]

that's errors that can happen in case u did smt wrong, that's really vague as you don't specify what error u get

[fleet fiber]

Yes, exactly and also when it stops working just three minutes after using it and i have to do it all over again

[fleet fiber]

I'm mostly talking about when ur downloading the model or converting the cover

[low shard]

yea you didn't still specify what error message u get so

[low shard]

for me
Yea for you, cus its the same program

its the same logic as running DOOM on 2 same laptops in terms of hardware with the same OS, the results won't change

[fleet fiber]

Yeah, i don't rlly remember what messages i used to get but the point is, i still had to start all over

[fleet fiber]

yes, they do change omg but u won't know that, unless, u try it urself

[low shard]

i cant really help without u telling whats the error message tho, it depends on what step u did wrong or if u downloaded a model with unnecessary files or if u finished gpu quota, or etc

[brittle wing]

I'm literally the girl you chat on Instagram with, I didn't blame you I thought the website creatore steal RVC models and put them on the site without giving the creators credit and get money off of that

[low shard]

they do change omg
If you run on the same exact copy of a game, on a same exact pc, how do u expect to have different results 🤦‍♂️

[quasi dagger]

Do you import the same model files you use in RVC (.pth files) on jammable?

[fleet fiber]

omg, ur getting on my nerves 😭😭 they do change, the results are better, the quality is better. I have nothing else to add cuz atp, go and try for urself

[fleet fiber]

Ik, i just wanted to clarify that i don't steal since the convo was mostly about me, ur good dw

[low shard]

you are literally using a logic in which doing 2+2 on 2 PCs would have different results 😭

[fleet fiber]

yes, and the results are better there too 😭

[fleet fiber]

OMG 😭😭 JUST ACCEPT THE FACT THAT IT'S BETTER FOR ME AND MOVE ON 😭

[brittle wing]

Bro stop it's literally a good site

[low shard]

its factual not better than rvc tho as its the literal same program

[fleet fiber]

same or not, again it's better in the sense of quality

[rare goblet]

Ayo? @fleet fiber level 3 !!!

[fleet fiber]

Level three of what huh

[odd shale]

It's not. (Mate, you're wasting your money there)

[low shard]

it is the same.. u can literally import the RVC models

[brittle wing]

Bro from the cover clip I sent you yesterday they're literally better

[low shard]

if it wasn't the same u wouldn't be able to do that, it would end up like trying to load a GPT So Vits Model in RVC, which wouldn't work

[fleet fiber]

I'm not talking about the models, ik if the models are bad then the results will be bad, I'm talking about the quality here

[fleet fiber]

Exactly, that clip u sent was mine and i made it on Jammable

[low shard]

the quality literally relys on the model and input audio you give it to inference with

There is no literally other way the quality can change on a same program

[brittle wing]

I know

[brittle wing]

Are you serious there is a difference

[fleet fiber]

okay, i agree with this but the quality also depends on the site ur using bcz tell me why i get different results with the different site? if they're all the same then why the quality and results are bad on other sites but good on another?

[brittle wing]

I said that applio and harmonify outputs have a little difference in sound

[low shard]

but the quality also depends on the site ur using
It doesn't, what the site is doing is just rvc cloud computing, its running RVC on remote PCs, letting u use the service (similarly to colab)

[brittle wing]

Bro you aren't deaf come on

[fleet fiber]

it's literally so different on each site😭 like why i get bad results on applio with the same models, same input but it's better on Jammable with still the same models and input 😭

Now, you're not trying to listen at all

ur just trying to prove ur point when it's literally wrong

[brittle wing]

@fleet fiber 💜

[low shard]

can say the same to both of yall

i literally explained yall how those sites work

its not like they have magic servers that make it run with better quality

[brittle wing]

And... that's why it's paid cause it's better

[low shard]

because its paid it doesnt mean its better..

its the same program

which is literally Open Source

[fleet fiber]

it's still better

[brittle wing]

Itsss the same program okay but the sound is more realistic

Idk what kind of coding they set here

[low shard]

its like saying iZip is better than Winrar just because you are forced to buy an iZip license

[brittle wing]

It's not th same

[fleet fiber]

Atp, just go and try it urself then cancel the subscription after that cuz u won't believe it unless u try it

[low shard]

its the same in terms of training and inference, if it wasn't it wouldn't be able to run the rvc models u imported

[quasi dagger]

Maybe different settings. In Applio you can configure how much index to use, volume envelope, etc and that also contribute to the final output but still rvc

[brittle wing]

Also Mel-roformer karaoke model on X-Minus sounds better than the one on MVSEP.

That's an example

[brittle wing]

Okay but how are the output more high quality and realistic

Explain this

[fleet fiber]

ik that it's still rvc and yes, bcz the settings there are different and even on Jammable i can also control the volume + extract the bv, instrumental on the same website and i can even add autotune and effects so maybe that's why the quality is better idk

[quasi dagger]

There are different versions of mel reformer so that's true but RVC is still the same

[brittle wing]

I don't think it's the same

[low shard]

this argument feels like trying to explain iOS users why sideloading is good but they think every file is a virus 😭

[odd shale]

Actually it is.

[low shard]

it is the same, there's no difference in quality between mainline, applio, ilaria rvc, etc

[brittle wing]

Have you tried them all

[brittle wing]

Have you tried jammable sir?

[fleet fiber]

well if that's what u think bcz i only tried applio so yeah, Jammable was better for me idk about the other sites

[brittle wing]

Well currently imo the best thing is Ilaria with 0 gpu

[odd shale]

Why i would when i got kaggle, local and applio?

[brittle wing]

That says enough
Yes personally I wouldn't pay money too but the result is realistic.On each their own

[odd shale]

On my humble opinion, there is no such thing as a realistic model/output.

[fleet fiber]

then u wouldn't know, all that talking was for nothing

[glacial pollen]

That's what I said, rmvpe with range clamping, in this case pitch, using hz values

¯_(ツ)_/¯

[brittle wing]

Well said

[brittle wing]

That says enough, that's why you can't tell the difference.

[odd shale]

They 4 are the same mate.

There is no difference.

[glacial pollen]

I personally see realistic models not as something you can't differ from real voice as that's sadly not achievable yet, at least not in standard scenario, but rather a model of which output makes you want to replay 1 or 2 times more as you're wondering if it's for sure AI or not

btw, here, catch 🥬

[analog obsidian]

I still can’t understand fm
So is this supposed to go down like all of metrics or is normal that it always goes up? Noobies told me is supposed to not go up

[glacial pollen]

All metrics are supposed to go down

[odd shale]

Thank you Cody..

[glacial pollen]

just that FM from what I've noticed, at least in those rvc builds I tested including mine, usually is reversed

( the graph )

yet the value on it's own should go down

[analog obsidian]

so is like a visual bug?

[glacial pollen]

either a bug or just a quirk devs left in the code

tho I myself couldn't spot what causes it last time I checked the code ( might be I didn't notice it but ye

[analog obsidian]

uhhh so how we can tell if fm is fine?

[glacial pollen]

just look at the value itself

rather than visual graph

if it's reversed in ur case, as in, up = down ( based on what values tell you )

then ideally it should be going up ( virtually, down )

[analog obsidian]

Hmm what if is going down instead?

then is bad?

[glacial pollen]

and the values increase?

as in, fm loss

[analog obsidian]

They decreased for me

[glacial pollen]

then if values decrease and graph is descending, it means it's all good

and just how it should be

values down + graph down = good ( normal behavior
values up + graph up = bad ( normal behavior

values down + graph up = common bug / quirk of rvc but it just still = good
values up + graph down = common bug / quirk of rvc = bad

[analog obsidian]

You said higher batch sizes helps fm to be stable, when i do that visually the graphs goes up very fast at the start of the training
Yesterday i tried a small batch size and it went down with a few fluctuations but it did not go up forever

why that happen?

[odd shale]

🐢: observes you both

[glacial pollen]

higher batch size contributes to more stable models because more data " at once " ( too complex to explain in easy words so imma skip it ) is used

however, it can lead to too flat converging or just overfitting quicker / more often

there's also that thing if your dataset is too diverse

for instance, weird balance of speech / singing / screaming

and you use high batch size, model might confuse the stuff during inferencing so it's best to, in that case, go for lower batch or balance the ratio

[analog obsidian]

hmm interesting i see

[glacial pollen]

as for lower batch sizes

[odd shale]

What if i use a clean 23 min dataset with 8 on batch?

[glacial pollen]

it's either your set is small ( yet balance is kept in the set ) so that's what it is
or your set's balance is kinda weird or broken, then it should help

[glacial pollen]

then it's a matter of how diverse is it

in terms of characteristics

i.e.: screaming, shouting, whispering etc

[odd shale]

If it's just speech with a bit of laughing?

[glacial pollen]

if it's rather uniform and balanced
high batch size should be good ( for ur case leo ) but sometimes you can decrease it to improve results / generalization

[analog obsidian]

Oh yeah this was a 8 min dataset so thats why it helped

[odd shale]

Nuevos trucos aprendidos jaja.

Gracias Cody.

[glacial pollen]

alternatively

if one wants to maintain the batch size but not tinker too much with the set

just add more " normal samples " and keep the " diverse / outlier ones / unusual ones " in 2-5% whereas 95-98% is your standard ones

[analog obsidian]

too diverse can cause fm to be unstable? The more you know

[glacial pollen]

yes because FM is feature matching

matching of features

if there's too high diversity, model is confused a lil

on what is the " right " features' distribution

more or less, simplified

Because, well, I talked about it a lot as it's crucial

Diversity is fine and sometimes welcomed yet uniformness and so, a bunch of data that can keep it's identity is the most important

[analog obsidian]

also other thing that happen to me is that rmvpe fm graph always went up regardless of batch size but when i switched to mangio crepe hop 32 and batch 2 made the graph not going up visually but instead looked like the one i sent

the end result was fine like u would expect to sound like

[red kayak]

lower batch sizes usually have lower loss values

[brittle wing]

I used to have two water turtles as a kid and turtles were my favorite animal

[glacial pollen]

Anything that contributes to " learning " stable features representation can do it

it isn't always the set on it's own

can be just

well, voice's complexity and it's nature

[brittle wing]

The sound is different you haven't even tried jammable but to each their own.I won't argue

[glacial pollen]

Now pardon me folks, gotta go back to sorting samples.
Ripped all data from danmachi game lol, over 30k samples to sort 💀
take care!

[red kayak]

share a raw jammable file

the way it comes out the website

[brittle wing]

I haven't paid for jammable

[red kayak]

then how would u know if it comes out any different owo

[brittle wing]

But the person who uses jammable has messaged me and I compared both

[rare goblet]

Ayo? @brittle wing level 8 !!!

[red kayak]

okay, may we have a listen too?

[brittle wing]

Hm I gave to ask this person for a sample first

[red kayak]

didnt he already give u a sample so you could compare the 2

[brittle wing]

The cover they made is realistic and made with jammable, proves it

But it has reverb and music

[red kayak]

yeah, thats the thing, you can manipulate the fidelity of the audio post processing

[analog obsidian]

Thanks for explaining me those stuff, they can be very confusing 😭

[glacial pollen]

@analog obsidian Sure thing! If you need more technical info, please search for my msgs on this server. I do leave explanations or stuff here n there, incld. metrics etc.

[brittle wing]

Okay how do I manipulate to make it more realistic, any tips?

I'm waiting for the person to reply

[red kayak]

im not a audio engineer myself so i dont know how to master audio and polish it

[brittle wing]

Mastering?

Ohhhhh

It's clear

[red kayak]

mixing and mastering are what audio engineers are for

the clean up vocals and then enchance their clarit

[brittle wing]

Bandlab has a mastering feature

[red kayak]

yeah u can mix and master with bandlab

[brittle wing]

Yeah but I applied the exact same effects on the exact same output and song and model yet their sounds more realistic

[glacial pollen]

What do you wanna achieve?

gimme a brief tl;dr

[red kayak]

yeah cuz it also takes skill to train a decent model

both models were trained on different datasets

[brittle wing]

I literally used the same model they use also I use the best models available

[red kayak]

yes but jammble doesnt share the model weights they have, so the model u 2 had couldve been worse than the one on jammable

[brittle wing]

You can literally upload models there

It's the same exact model

Again

[analog obsidian]

Bro if they were two different things u wouldn’t be able to use your rvc model there

[red kayak]

then with that in mind we have cracked the case; they use post processing to make their outputs sound better

simple as that

[brittle wing]

What post processing?

Can I manually do that

[glacial pollen]

How can Litsa know what was the exact post processing chain they used

[red kayak]

well they could've used some sort of de esser or eq to smooth out some frequencies and make them more natural and yes u can manually do that you just have to learn how to. Now i don't know what exactly they run the audio files through since i dont have the code they use at hand so i cant tell u anything thats accurate to that :p

[red kayak]

im a wizard now :p

[analog obsidian]

Hes a Magic dog

[brittle wing]

It's some kind of audio sfx

[brittle wing]

They probably do

[glacial pollen]

de-essing, eq, compression, saturation, there's tons of what can be done

[red kayak]

yeah what codename said

[brittle wing]

Wolf

[analog obsidian]

Hes a husky dog

[glacial pollen]

Might be you could try to reverse the impulse maybe?

[brittle wing]

Is it possible with bandlab

[glacial pollen]

Read on irs / convolvers

[red kayak]

i dont work for the company so i dont have their code at hand

[analog obsidian]

Magical husky dog

[glacial pollen]

get the exact same input they used and make sure it's the same model

get an irs response out of it effectively replicating the effects' chain

and apply it over ur model's output
tho, I won't teach you how it's done as that's not my specialization, just proposing an idea ~

[brittle wing]

Ohhh I couldn't tell from the drawing

[brittle wing]

Understandable

[red kayak]

these sort of website usually do that to attract and make it sound more appealing to new users. The raw outputs are usually the same though, post processing can make a huge difference.

just keep that in mind

[analog obsidian]

Eqing usually makes rvc models to sound realistic

U can decrease the volume of the artifacts