✨│ai-help

[low shard]

Are you using https://pypi.org/project/tts-with-rvc/ ?

Oh nvm

[urban fractal]

lol

[low shard]

Have you also tried with other models

Might be a model issue

[acoustic lion]

I am curious, will AI RVC ever improve to the point it can mimic coughing and laughing (odd sounds that can be made from the human mouth in general) Cause if someone screams with a high pitch the model wouldn't comprehend what the user is trying to accomplish

[urban fractal]

Yea they all sound like robots, or russians. I try to set the voice but it doesnt exist or its the preset built in ones

for ARIANAGRANDE_ES_BY_SZAJEAN.pth
would I set the voice to something? or leave it empty? If i dont set it then shes a russian man

[low shard]

You can't expect much, RVC is meant for STS, all that the python packages does is make an API request to Microsoft edge tts then use that audio as an input for RVC inference

Ms edge tts is multilingual and nice quality but not emotional at all

[urban fractal]

Ohhhh

I getcha. lol, makes a lot of sense now. I'm making a bot to have a convo with and ms sam is trash so was looking to better. The built in voices don't seem to sound like that though if I change to a different en_* voice

[knotty moth]

Apolio is a good alternative to that kind of application

[urban fractal]

Thats a python lib?

[low shard]

Applio does the same thing 😭

It uses Microsoft edge tts API requests

[urban fractal]

Awww

[low shard]

Nope it's an RVC fork (modified version)

[acoustic lion]

should this question be asked somewhere else? unsure if it doesn't belong here

[low shard]

It does RVC have improvements, but you can't expect it to sound emotional in applio either

[low shard]

Only the future can tell us

Unfortunately the original devs of RVC kinda left it to work with gpt so vits, which is a TTS AI

[urban fractal]

Are you sure I don't have to set the voice param to match the voice name in the model file? I just don't know what that name would be

[low shard]

Hopefully applio and our engineers will make improvements but who knows

[acoustic lion]

Ah, cause I do intend to do roleplay with AI Voice but not wanting to sound awkward in certain situations

[low shard]

I mean setting a tts voice which has the same gender and language helps for sure, but it doesn't give emotions

[urban fractal]

ok

but she sure dont sing like in the samples I heard

[low shard]

Microsoft edge tts was mostly meant for just reading text on the browser, Soo yeah

[knotty moth]

it can depend on coverage and quality of the dataset and pretrain used. you can try this https://discord.com/channels/1159260121998827560/1339155300720054316 but dont expect to be always optimal for those edge cases

[low shard]

That's very experimental though

Iirc hifigan was disabled in applio main branch for now not being stable enough

[urban fractal]

Ohh, yeah huge improvement by setting the voice to en-US-AvaNeural

[acoustic lion]

Is there any examples of it being used? Real time scenarios

[knotty moth]

I have tried training some metal singer model

[low shard]

Yeah it does help, tho not in the parts of emotional speech, like not as emotional as 11labs

[low shard]

How did it go ?

[analog obsidian]

we need a new embedder for that, but thats very complex to do at the moment

[knotty moth]

not too optimal but not bad for making cover mixes

[analog obsidian]

seoul finetuned contentvec and was able to make models scream

[low shard]

@urban fractal are you having any other issues ?

[low shard]

Well it's something

[urban fractal]

Nope

[acoustic lion]

That can be considered a revolutionary step in real time voice changer no?

[trim sparrow]

nick super quick question how should i know how much epoch to use ?

[urban fractal]

I'll share a video and code once I get it together

[analog obsidian]

yeah

[low shard]

There isn't a right amount

Use the tensorboard

[analog obsidian]

but rvc does not stand for realtime voice changer

[acoustic lion]

Oh, thought it does 🤔

[trim sparrow]

do you have a tutorial video or doc perchance for that?

[analog obsidian]

Retrieval-based-Voice-Conversion

[knotty moth]

noobies said xeus can work but takes much resources

[acoustic lion]

Ah, what term do we use for Real time then?

RTVC?

[analog obsidian]

wokada

[acoustic lion]

Ah

[low shard]

https://docs.aihub.gg/rvc/resources/training/#tensorboard

[acoustic lion]

Then in terms of Wokada, has there been any noticable improvements lately?

[analog obsidian]

w-okada is merely a gui for realtime inference

[trim sparrow]

ah its just a file for applio

[analog obsidian]

it's just a gui, it can't get any better

we need to improve actual rvc

[low shard]

Wokada is a program that basically focuses on the realtime rvc inference

There are 2 versions
Original made by Wok
And deiteris fork made by deiteris

The deiteris fork initially brought many advancements, which now are also in the latest original wokada, but deiteris fork is suggested still for having more options and less bugs, along with the removal of Beatrice models which were low quality compared to RVC and were experimental

[analog obsidian]

every change the gui may receive will be a optimization or qol change

[low shard]

There was also a go-realtime.bat rvc GUI for realtime inside of rvc mainline (og rvc), but that's really outdated compared to the ones I mentioned

[analog obsidian]

tier list for perfomance atm:

1. deiteris
2. mainline realtime
3. original w-okada

[low shard]

I mean it has improvements in performance

[hallow thistle]

This is the better W-Okada.

[orchid flint]

I got a random question as well. Besides the new vocoders is there any team that is actively working on, for example a RVCv3?

[analog obsidian]

rvc boss feels rvc is already perfect and doesn't need anything new

don't expect something from him

[low shard]

Rvc boss Is focusing on gpt so vits

[orchid flint]

Ok

[analog obsidian]

they tried rvc v3 but didn't liked the results

[orchid flint]

I see so new vocoders are the only way forward for now?

[acoustic lion]

Ah I am certain this is what I have but not sure if I have the latest version, does it update automatically?

[sly furnace]

i installed everything sir , can you tell me the settings

[analog obsidian]

imo the new vocoder thing is nonsense, we need a new embedder

[low shard]

Weird, could you try to open CMD and do env/python.exe -m pip install --upgrade gradio

[lofty lichen]

anyone working on new embedders?

[crude flame]

those are up to you

what you want to sound like

[orchid flint]

What would a new embedded or look like? I’m not familiar with those

[analog obsidian]

seoul was finetuning contentvec last time, but i have no idea how that went

[hallow thistle]

W-Okada doesn't update itself. You must download one and install again.

[analog obsidian]

he's trying to make models laugh in realtime

[sly furnace]

male to female what should i use

[crude flame]

again its up to you

[low shard]

Ban

[lofty lichen]

is KLM 5 RefineGan best for talking female with medium high pitch voice?

[sly furnace]

i am completely new to this

no idea

[analog obsidian]

it would improve: the model ability to laugh, whisper, etc

[trim sparrow]

worked! ty

[crude flame]

just mess with it untill youre happy

[sly furnace]

any recommeneded settings would help

[low shard]

@simple ore btw are you aware of this?

[sly furnace]

😭

[low shard]

It seems that Gradio needs to be updated

[lofty lichen]

is KLM 5 RefineGan best for talking female with medium high pitch voice?

[low shard]

You're welcome

[acoustic lion]

Is it alright if you can give me the right link for it?

[analog obsidian]

our current pretrains are already pretty good

[low shard]

Pretrains help mostly for language tbh

[acoustic lion]

And is there any sorts of logs regarding new updates, when it was ,what it introduces

[analog obsidian]

thats old, use klm 4.9

[orchid flint]

Oh awesome, one of the things I hope in the future is that RVC will be able to handle more synthetic voices since there are characters that I want to make but I am struggling to be able to train.

[knotty moth]

dont touch refinegan pretrains, yet

[trim sparrow]

if i got 17.1 minutes of datasets would a epoch of around 450 be okay?

[hallow thistle]

This is the bitcrush effect. It's basically another distortion effect that reduces the resolution or bandwidth of audio to sound harsh and noisy.

[analog obsidian]

4.9 is a hifigan pretrain

[acoustic lion]

Ah I see, so the core of it is RVC and there'll only be improvements to Wokada if there are more revolutionary steps for it?

[hallow thistle]

What is your PC GPU?

[acoustic lion]

4070 ti

[analog obsidian]

if rvc gets better, w-okada will also get better

[low shard]

-realtime

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[low shard]

1st link

Wokada deiteris fork

[analog obsidian]

w-okada can't get any better by itself

it's just a gui for inference

the only thing that can be improved there is perfomance

[low shard]

Hell yeah let's run Wokada on a raspberry pi 6 🔥

[orchid flint]

Lmao

[low shard]

Wokada working on a NES

Let's make Wokada hackrom

[acoustic lion]

I was thinking more of an intel duo core

[low shard]

Duo? Why not 0.5 🔥

[knotty moth]

there wont be NPU support to work on as well

[acoustic lion]

true true

[hallow thistle]

W-Okada simulator.

[low shard]

Let's make Wokada run on hopes and dreams

[acoustic lion]

NPU?

[orchid flint]

Yes

[hallow thistle]

More like Intel Pentium.

[versed edge]

It's not a headset with a mic attached, the mic and headset are separate

[lofty lichen]

refinegen disabled in applio?

[knotty moth]

the latest intel & AMD laptops

[acoustic lion]

Hmmm, Celeron.

Doesn't sound like a problem to me then, Towers all the way

[orchid flint]

You need to activate it by using this #🔥│model-maker-chat message

[lofty lichen]

no access

[orchid flint]

Dam

Ok

[low shard]

Try increasing in sens a bit

[low shard]

Neural Processing Unit

GPU + CPU for ai laptops

Which is barely usable since no programs support it

[orchid flint]

They can I’ll just grab the link to the file

[low shard]

I mean there are, but pretty few if you compare it to Nvidia GPUs

[acoustic lion]

Kind of funny how an AI is refusing to work with another

[orchid flint]

https://cdn.discordapp.com/attachments/1159290096458149938/1361751274319511742/train.py?ex=6801df73&is=68008df3&hm=107944ce38042b45db19df94c805a4e1639557cb40184c81815d7133841642b3&

Put in rvc/tabs/train.py

[acoustic lion]

But AI laptops sound pretty cool

[low shard]

AI is mostly supported on Nvidia GPUs

You can also do it on AMD and Intel but it's a much more pain

[acoustic lion]

Though looking at the site I was provided, last update was December 2024, so I do have the latest build for it

[low shard]

AMD in most cases needs zluda, a cuda emulator

[orchid flint]

Eh, more like it was just set to false in the train.py file

[pastel oak]

Ok first in 250% and out 400% is a bit excessive, lower in to 100, out ig doesnt matter too much if u want it to be loud but test it at 100-200% range anyway

Try audio: server mode with s.r. 48000 and [windows wasapi] prefix on input and output. This uses better audio device than client. Could help

In advanced
Protocol sio, force fp32 mode on

[crystal dove]

Can someone help me transfer music to the lyrics of a singer? My PC is a mess now and it's urgent

[acoustic lion]

Hmm, but does the RVC inside of Wokada come updated everytime you launch the program?

[analog obsidian]

no

[crystal dove]

pls if anyone can help me dm me

[analog obsidian]

if rvc gets a new important change (like a new embedder) the w-okada dev has to update the gui to use the new embedder

[low shard]

Please elaborate the issue and what's your PC GPU

[acoustic lion]

Ah

[orchid flint]

When was the last time RVC got an update?

[acoustic lion]

Well since it says december 2024, I guess we haven't reached that goal

[analog obsidian]

5 months ago (mainline)
3 days ago (applio)

[orchid flint]

Oh, ok

[gray cobalt]

i deleted the model files but they're still visible and usable for some reason

and trying to delete causes a problem

[analog obsidian]

it's hard to find a fast alternative, all of the cool ones require an absurd amount of vram for both training and inference

[low shard]

Did you delete them from model_dir?

[crystal dove]

no, its like right now my pc is a bomb like i can use it cause my CPU and GPU are in dust due to a blackout last night, I'm on my phone, that's why I'm asking for help

[gray cobalt]

from wokada itself

[trim sparrow]

@low shard when clicking the generate index it does this

[acoustic lion]

Like 4090 amount of vram?

[gray cobalt]

hmm

[simple ore]

i'm tired of people hitting themself over the head, instead of simply downloading the compiled version

[analog obsidian]

yup

even higher

[acoustic lion]

Oof

[trim sparrow]

i just followed the doc my bad 😭

[obtuse onyx]

sry i want to ask, do i need to pay to use voice.ai app?

[analog obsidian]

i think xeus is able to run locally but it ask for too much

[gray cobalt]

i see

[low shard]

He did use the compiled version

[gray cobalt]

im shit at using pcs bruh

[acoustic lion]

Only way I can think of making it work is by hooking 2 4090s in one PC and leaving one to power the demanding embedder

[simple ore]

nope

[crystal dove]

can anyone help or...?

[orchid flint]

4090 only has 24 gigs of VRAM. We would want more like 80 lmao

[acoustic lion]

80?

DAMN

[low shard]

Isn't that the same steps of https://docs.aihub.gg/rvc/local/applio/

[acoustic lion]

You'd be needing an entire NVIDIA A100 to run that

[knotty moth]

4060 is enough to handle voice changer alone, plus streaming

[acoustic lion]

Ah no I know that

my 4070 ti is doing fine work

But we're mainly talking about the demanding embedders

[orchid flint]

Yeah

AI go crazy

[simple ore]

that's correct steps, but that error only happens when you download old source and try to install it.. so it grabs outdated library

[acoustic lion]

But what about those AI TOPs in the 50 series? Could they be made of use to RVC?

[low shard]

what's your pc gpu

[trim sparrow]

1650

[orchid flint]

💀

[low shard]

ohh so ur pc is broken and ur trying to do ai covers on phone?

[acoustic lion]

Oh god

[low shard]

it's fine dw

[crystal dove]

ajam ajam

you know how to do it?

[acoustic lion]

Actually are they called AI TOPS

or just TOPS

Honestly not sure what they are or what they do

[low shard]

but the download link is the same as the one in applio docs

[low shard]

ehhhhhhhhh

it has 4gb vram

[acoustic lion]

Yeah doubt 4gb of vram can do anything unfortunately

[trim sparrow]

damn

[acoustic lion]

You can probably get a 1080 ti for cheap

It has plenty of vram

But well, not the fastest

[low shard]

you can inference fine, but about training, you're limited

you can train but at a low batch size which is usually more unstable

like 2 or 4

[trim sparrow]

i had it on 4

ima try again in a sec at 2

[knotty moth]

any RTX are better due to having tensor cores

[acoustic lion]

That's true yeah

[analog obsidian]

main branch applio has checkpointing, the speed lost isnt that much

[acoustic lion]

These tensor cores making me tense up...

[low shard]

wasn't that giving worse quality

[analog obsidian]

it was the inplace thing

im training my models in mainline now

[low shard]

so checkpointing works all fine now?

[low shard]

why

[analog obsidian]

yup

[knotty moth]

6 gb vram is recommended for that

[low shard]

welp

he's cooked

[trim sparrow]

is there any other steps for tenorboard with applio besides opening the .bat?

[analog obsidian]

well, it's the original rvc

[low shard]

@trim sparrow if u want to do training, you could also try cloud, meaning you will use a remote good pc and won't run on your pc

but, it has limited time in free tier

[trim sparrow]

do you know how much time limited?

[low shard]

also cloud notebooks sometimes break easily since the cloud provider update packages and python version for example

[trim sparrow]

whenever i open it i go the link and it says no dashboards for this current data set

[low shard]

4 hours max of gpu daily on colab (u need a google acc)

30 hours weekly for kaggle (needs an acc + phone number verification and harder)

22 hours monthly on lightning.ai (needs an acc + phone number verif)

[low shard]

that's because you need to start training first for a bit

and then refresh it

[trim sparrow]

ah okay you think kaggle would be a good option?

[low shard]

yeah

it's owned by google btw

[trim sparrow]

do you know the website?

[knotty moth]

reach epoch 1 then try refresh it

[low shard]

this server is english only btw

cloud is way suggested for phones

[sly furnace]

@crude flame https://www.youtube.com/watch?v=bXEUto3lMyk
how i can get smooth voice like this ? i am not talking about the model but this looks so good

[trim sparrow]

whicih one do i start

with

[simple ore]

what's inside of your model's folder (logs/model_name) ?

[low shard]

you need to check the tutorial inside of the kaggle

oh right I didn't send you lmao sorry

[trim sparrow]

i actually hjave no clue but it started working when i restarted applio

[low shard]

Train (make) RVC Models on cloud:

1. Prepare the Dataset
2. Setup RVC:
   Choose a cloud way to use RVC,

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (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)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • Applio (UI)
  • Mainline (UI, No guide as of right now)

3. Be sure to know about the tensorboard

Google Colab = Easier but risk of getting disconnected
Kaggle = Harder but way more gpu time
If you are looking for the easiest way and for free, is using https://weights.com/ which ofc uses RVC

RVC Inference (use models) on pre-recorded audio on Cloud

You can use either:

• Weights.com: Easiest Possible Ever Automatic
• Ilaria RVC Zero: Fastest free on cloud
• Applio UI Colab: RVC Fork with some extra features like TTS
• RVC AI Cover Maker UI: Automatically Separates the Vocals and Instrumentals, converts the voice and mixes them back

[trim sparrow]

it tells me to put ngrok token but i dont see the place where i input it in the code as shown in the doc but it says this

[simple ore]

in applio colab you just type it in the field

[trim sparrow]

im on kaggle

[simple ore]

replace "token" with your actual token

[trim sparrow]

im not seeing that at all is crolled through the cells and tried the find thing to just search up for !pip install pyngrok and it didnt show anything nor did !ngrok config

[versed edge]

if i use server mode do i need to download a noise cancellation software for the bg noise since sup2 doesn't work

[trim sparrow]

yeah im not finding that anywhere

[simple ore]

post a screenshot

[trim sparrow]

https://www.kaggle.com/code/hinabl/mainline

is this the right one to copy?

[simple ore]

that's an entirely different colab

[trim sparrow]

is it this one https://docs.aihub.gg/rvc/cloud/applio-kaggle/

[simple ore]

but you can do it here

[trim sparrow]

ah okay thank you

[loud thicket]

I don't think this fits as an AI related question but I'm wondering what role do I need to be able to send a message in a model thread to ask a question there.

[pastel oak]

Need separate yea

[ancient karma]

Does anyone know how to train voices remotely? My GPU isn't good enough to train a model.

[versed edge]

any recommendations

[pastel oak]

Steelseries sonar

[simple ore]

for nvidia - broadcast app, here's how good it is https://x.com/thegunrun/status/1252789873699745792?lang=en

[lofty lichen]

have u trained with refinegan

[orchid flint]

I have not, I likely won’t until or if it gets supported on wokoda

[lofty lichen]

keep getting (The parameters of the pretrain model such as the sample rate or architecture do not match the selected model) when using it

[orchid flint]

Im almost certain that’s because you need to use the pretrains in the discord

Apollo doesn’t have any official pretrains

For refinegan anyways

[lofty lichen]

when i choose

[orchid flint]

I think so

#1327283027776245810 message

[lofty lichen]

im using KLM 5 RefineGan keep getting

(The parameters of the pretrain model such as the sample rate or architecture do not match the selected model)

[simple ore]

as I see @tight ether hasn't tried making another pretrain with the latest code

[lofty lichen]

alr thx

[trim sparrow]

ERROR: Exception in ASGI application
raceback (most recent call last):
File "/kaggle/tmp/.venv/lib/python3.10/site-packages/uvicorn/protocols/http/h11_impl.py", line 403, in run_asgi
result = await app( # type: ignore[func-returns-value]
File "/kaggle/tmp/.venv/lib/python3.10/site-packages/uvicorn/middleware/proxy_headers.py", line 60, in call
return await self.app(scope, receive, send)
ERROR: Exception in ASGI application
File "/kaggle/tmp/.venv/lib/python3.10/site-packages/gradio_client/utils.py", line 973, in _json_schema_to_python_type
f"str, {_json_schema_to_python_type(schema['additionalProperties'], defs)}"
File "/kaggle/tmp/.venv/lib/python3.10/site-packages/gradio_client/utils.py", line 919, in json_schema_to_python_type
type = get_type(schema)
File "/kaggle/tmp/.venv/lib/python3.10/site-packages/gradio_client/utils.py", line 880, in get_type
if "const" in schema:
TypeError: argument of type 'bool' is not iterable
An error occurred launching Gradio: When localhost is not accessible, a shareable link must be created. Please set share=True or check your proxy settings to allow access to localhost.

i get a error when trying to run through kaggle

[low shard]

that's really weird, try to make a new code cell, put uv pip install --upgrade gradio then run the run start cell again

[trim sparrow]

"error: No virtual environment found; run uv venv to create an environment, or pass --system to install into a non-virtual environment
"

[low shard]

sorry I forgot to put ! above that code, what if u add that?

[trim sparrow]

i added it

it wouldnt work without the !

should i js create a env? "!uv venv .venv"

[low shard]

what if u remove the uv?

i'm doing a quick test rn, installing requirements

[trim sparrow]

i just did the venv .venv thing and it worked but not ngrok is saying theres multiple instances from before n i have no clue how to terminate

[low shard]

either restart session or try to check if you're able to in the ngrok dashboard

[trim sparrow]

TypeError: argument of type 'bool' is not iterable
An error occurred launching Gradio: When localhost is not accessible, a shareable link must be created. Please set share=True or check your proxy settings to allow access to localhost.

this is so annoying

idk i give up on this

[low shard]

I found a fix for applio kaggle
delete the notebook you made

make a new one

add !uv pip install gradio==5.25.2 -q below !uv pip install numpy==1.23.5 -q in the 1st cell

then follow the guide as normal

[low shard]

@simple ore since vidal is not active, is anyone able to even update the kaggle anymore?

[trim sparrow]

im using the google colab one right now it seems to be working is it worth it or should i try notebook again?

[low shard]

it's not as much suggested as kaggle for training, since the gpu time is random daily and max 4 hours

and you don't know how many epochs to train, soo

[simple ore]

use the latest kaggle notebook

[trim sparrow]

i put this under the install one?

[low shard]

it is the latest kaggle notebook, i just tried it myself making a new copy

[simple ore]

@low shard you can just import the notebook directly from github instead of taking 2-month old one

[trim sparrow]

how do i do that?

[low shard]

the github one?

[simple ore]

yes

latest code has !uv pip install gradio==5.23.1

[trim sparrow]

how do i import it from github?

[simple ore]

use import notebook, then

[low shard]

oh we always used https://www.kaggle.com/code/deiant/applio

since it's in https://docs.aihub.gg/rvc/cloud/applio-kaggle/

[trim sparrow]

do you have the github link for it?

[low shard]

@crude flame u gotta update the docs then i guess

[lofty lichen]

training with refinegan slower? or just me

[simple ore]

that one is old, but same fix

[simple ore]

not much, about the same

new code uses fp32, so a bit slower

[low shard]

applio docs need to be updated too, https://docs.applio.org/applio/getting-started/other-alternatives#set-the-environment uses https://www.kaggle.com/code/deiant/applio/notebook

[simple ore]

I assume Vidal owns it, so he needs to update it

[trim sparrow]

how do i put my database inside of Applio?

do i put the dataset inside of the "working>Program_ML"

@simple ore

[simple ore]

i have not used kaggle

[trim sparrow]

thank you so much and also is there any way to see the console so i can see the progress?

nvm got it, thank you again you and nick!

[sly furnace]

which is the best version for ai cover? fork or mainline rvc?

[austere harness]

i need help

my voice model doesn't work on discord for some reason

[low shard]

Vidal owns the site ?

[low shard]

Share a screenshot of wokada and discord settings

[low shard]

Mainline/ applio rvc

Original wokada and Wokada deiteris fork are meant for realtime

[austere harness]

ok wait

[low shard]

!give-media-perms 1h @austere harness

[austere harness]

@low shard here are my settings

theres no audio coming out

[low shard]

Uncheck sup1

What browser are u using

[austere harness]

i use google chrome

[low shard]

Check if it works when u use client

[simple ore]

he owns the notebook the site links to, but it should be just a copy of the repo

[austere harness]

nope doesnt work

[austere harness]

this doesnt work too

[low shard]

Did u set input to microphone and line 1 to output after using client

[austere harness]

yes i did exactly that

no audio

[low shard]

Did u also follow this when u did the 3rd step

[low shard]

Yeah, what I was saying is the applio docs should be updated to using the GitHub import version instead of the Vidal link

[austere harness]

yes i think so

[low shard]

Could you check?

[analog obsidian]

stop the conversion, then disable sup1

[sly furnace]

which is better?

[analog obsidian]

mainline for inference

[low shard]

Between mainline and applio? Quality wouldn't change much

[austere harness]

other voice models work through discord, but this one for some reason just gives no audio i dont know why

[simple ore]

well, i'm not sure about that... the copy is frozen for a specific release

[sly furnace]

whats inference now 😭

[sly furnace]

something will be better right ? thats why they are 2 not one

[analog obsidian]

getting the vocals for ai cover 😲

[low shard]

It could prob just be a model issue then, it's the only single one that doesn't work?

Could u also send the link to the model download link

It's prob just a bad model

[austere harness]

yes all other models i tried work on discord just this one gives no audio through the discord, i set to monitor the voice i can hear the voice model when i set monitor to my speakers

[sly furnace]

so i should do rvc mainline to make model then use appolio for making covers?

[austere harness]

but theres no audio on discord

[analog obsidian]

applio is easier to use

[austere harness]

https://huggingface.co/Coolwowsocoolwow/SAS_counter_terrorist/resolve/main/sas_counter_terrorist_e260_s8320.zip?download=true

yes here it is

[sly furnace]

dont care about ease , need the best one

mainline better?

[low shard]

Both wouldn't change in quality

[sly furnace]

😭

[analog obsidian]

theyre the same thing brah

[low shard]

Mainline and applio quality are the same

[sly furnace]

so whats the difference

[analog obsidian]

mainline is the original rvc

[low shard]

What changes in applio is the user interface being easier and some performance improvements compared to mainline which is the original

[sly furnace]

yea and appolio is the fork

[low shard]

And some features like tts

[austere harness]

doesnt work lol

[analog obsidian]

change to client

[sly furnace]

what do u suggest

appolio ?

[austere harness]

also doesnt work lol

[analog obsidian]

[low shard]

Applio* yeah

[sly furnace]

thank you so much sir

https://applio.org/

this right?

[low shard]

First, what's your PC GPU ?

[sly furnace]

rx6600 , it supports amd right?

i have 1080ti but its being used at another system

[low shard]

You need a special guide for AMD

[sly furnace]

like the onyx ?

[low shard]

Because AMD is less supported in most ai programs than Nvidia

[sly furnace]

no worries i will get help from my friend he has 4070

[low shard]

https://docs.applio.org/applio/getting-started/installation#amd-gpu-support-windows

You need to use Zluda

A CUDA emulator

It's just longer to follow but will work

[sly furnace]

i will follow it but do u think i should use my own machine or nvidia will be better?

[low shard]

Well that RTX 4070 will have better performance

In speed terms

Quality should be the same if you use on ur amd GPU rmvpe

[sly furnace]

yeah i know that mate but thank for you explaining so simply

you are so good!

[simple ore]

4070TI Super is 6-7x faster than 6700xt for Applio

6600 is memory limited

[sly furnace]

dont have that , he has 4070 laptop so its a mobile version so it will actually perform like 4060 desktop version

[simple ore]

yeah, 4070 -30% for laptop

[sly furnace]

yeah i am a techy myself haha

just getting into ai stuff

[austere harness]

@low shard so is there a fix for the voice model not working or should i just get a new model?

[low shard]

i just tried it myself, it works fine

try re-uploading it in wokada

and be sure u extracted the zip

[austere harness]

ok

[trim sparrow]

@low shard which batch size should i use with kaggle?

[low shard]

depends on dataset lenght

[trim sparrow]

its 14 minutes

of singing/regular speaking

[low shard]

8 should do the job

[trim sparrow]

and then save should i put that on 20?

and for the epoch should i start at 250 and then when it stars dropping just stop the training?

[austere harness]

ok it works now

[low shard]

5 or 10 should be finhe

[low shard]

put epochs to 500, it will train from 0 to 500

and u check the tensorboard how it goes

[low shard]

you're welcome

[trim sparrow]

alright bet thank you so muc for all the help

should i mess with any of the advanced settings or just not touch those?

[low shard]

u should be fine as long as ur using rmvpe

[trim sparrow]

alright brotha ill lyk how it goes thanks again!

[charred canopy]

Hey guys I need to make an AI cover using Yoda and I don't know how to do it in any sense, I have used jammable and a lot of other services and it's not very good, does anyone have any ideas on what I can try?

[low shard]

what's your pc gpu?

all that those sites use is in reality the FOSS (Free and Open Source Software) named RVC (Retrieval-based-voice-conversion)

you can run it locally on your pc

[low shard]

yw

[jolly tangle]

Ur gpu?

[low shard]

I got an rtx 4060 ti 16gb

why?

[sly furnace]

how much echopos and how many minutes of voice recording i need

for best cover

[jolly tangle]

Just wonderiing

[low shard]

wbu

[trim sparrow]

is this normal?

[analog obsidian]

this is the tedious part of rvc, there's no way to tell

its all random

thats for epochs

[sly furnace]

there would be a part where more echopos makes little to no difference

[jolly tangle]

Im training with 8 minutes. The female can talk

So

[sly furnace]

i want to make ai cover

[analog obsidian]

yeah too many epochs doesn't mean its better

[sly furnace]

so need talking + singing clips right?

and how long i need it to be

[analog obsidian]

if you want to do a singing cover, a singing model its better

[sly furnace]

i can sing any bullshit ?

[sly furnace]

i know shit

in singing

[analog obsidian]

dont expect a speech model sing godly lol

30 minutes is good for realism

[sly furnace]

i made it in past

[analog obsidian]

but anything less than that works

[sly furnace]

it wasnt so good

so i want to make another one

someone made it for me

[analog obsidian]

are you recording your own voice? then sing, use your whole range

[sly furnace]

singing + talking or just singing bullshit ?

[analog obsidian]

just singing

[sly furnace]

okay

bullshit works ?

[trim sparrow]

@low shard mb for all the pings is this bad for the model or normal?

[analog obsidian]

anything yeah

rvc clones the expression of the dataset keep in mind that

[sly furnace]

wdym ?

the pre train ?

[analog obsidian]

if u sound tired the model will sound tired

like that

[sly furnace]

oh

the energy

[analog obsidian]

yuh

[low shard]

it's normal

[analog obsidian]

you can sing anything

[low shard]

also it just started

[sly furnace]

cant i make it mix?

[low shard]

overtraining doesn't happen instantly

[sly furnace]

low and highs

like i am drunk

[analog obsidian]

no point in mixing talking if the model is only gonna be used for singing

[sly furnace]

and sober

at same time

💀

[analog obsidian]

lmao anything works

as long its singing

[trim sparrow]

it can be good to get the tone of voice or no? i did that with mine

[sly furnace]

and if i want two languages

i should make different models for them right?

[trim sparrow]

how can i know when to stop it ?

[analog obsidian]

the model will randomly use the speech samples during a song

instead of fully using the singing samples

[trim sparrow]

welp i guess we will see how it plays out 😭

[analog obsidian]

its all random so i cant even predict that

[low shard]

check https://docs.aihub.gg/rvc/resources/training/#usage-guide

[low shard]

AI is gambling 🔥

[analog obsidian]

it would be better yeah

havent tried that

[sly furnace]

30 minutes is max?

anything over it doesnt matter?

[analog obsidian]

1 hour max

after 1 hour, theres no improvement

[sly furnace]

alright lets go 1 hour of bullshit

and how many epochs

[analog obsidian]

try 500 epochs

and hear all of them until you find the one that sounds more natural

[sly furnace]

💀

[analog obsidian]

graphs don't help in choosing the epoch, they're meant to tell you how well the model is doing during training

[sly furnace]

how much time it takes usually?

to train 500

[analog obsidian]

1 hour set bs 8
hmmm
around 4 hours?

[sly furnace]

u mean 100 200 300 400 500 ?

[analog obsidian]

maybe 5 or 6

idk

[sly furnace]

bs means?

[analog obsidian]

1 day

batch size

its slow asf

[sly furnace]

batch size means each time it takes backup right?

[analog obsidian]

batch size is how many samples at the time rvc will learn

[sly furnace]

yea i meant that only

same thing

[analog obsidian]

too high its bad
Too low also its bad

[sly furnace]

u meant that right?

u said 8 i will do 8

i will spam ping u when i have issues ty so much buddy

[analog obsidian]

i dont answer pings

[sly furnace]

so i have to send cops to ur place?

can i dox u

https://tenor.com/view/paradox-px-px-bratislava-paradox-bratislava-black-and-white-gif-24697278

[analog obsidian]

https://tenor.com/view/noahnoah-gif-15569947813902812404

[trim sparrow]

😭

[sly furnace]

i dm u ?

[low shard]

yeah it's completely normal and good

it's the loss value

[trim sparrow]

alright ts got me stressing 😭

[sly furnace]

i can see you

[analog obsidian]

just ask here bro there are like 10 people that can answer ur questions

[low shard]

loss goes down = less losing shortly explained

[low shard]

[sly furnace]

but i like u 👉 👈

[low shard]

if u want the easiest way ever possible, weights.com exists too

[analog obsidian]

[low shard]

both of yall are banned 🔥

[sly furnace]

but i am a cute panda

[analog obsidian]

training models aint hard
just tedious

[sly furnace]

i know but i want to impress my 3rd gf with my ai cover

so wanna make it better than previous ones in past

[analog obsidian]

record yourself, after dat denoise your audio, then run this in audacity

[sly furnace]

i will require it when i do in practical

[analog obsidian]

and use the simple slicing in applio, 3s and 0.3s of overlap

[sly furnace]

i will lost it till then

so i have to disturb u again

i will ping u again tomorrow

should go to bed

Good morning and good night

btw jokes appart u guys really helped a lot , it didnt helped anything but still u helped haha i appreciate it alot thanks a lot for your time buddies

❤️

[analog obsidian]

good night

[crude flame]

i retruth this

[trim sparrow]

@low shard still normal or no

[charred canopy]

I have a nividia rtx 3050

Problem is I need to do it quickly, I don't want to have to start training anything from scratch

[trim sparrow]

😭

[low shard]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• Cloud (remote good pc, easier and faster than ur PC but it's limited):
  • Ilaria RVC Zero: fastest and simplest that you can get for free
  • Weights.com: Partnered with AI Hub, lets u do them easily but u may be in a queue
  • Applio UI Colab: max 4 hours daily, not granted, of GPU
  • RVC-AI-Cover-Maker-U Colab: Automatically separates the vocals and instrumentals, converts the voice and mix all together back

Easiest possible (automatically separates vocals & instrumentals) : weights.com & rvc-ai-cover-maker-ui
easiest cloud: Ilaria rvc zero
easiest local: Applio

if u just want a quick thing as easy as possible then use weights.com

[charred canopy]

Thank you so much

[low shard]

yes

[low shard]

you're welcome and let me know

[trim sparrow]

should i stop it at 16:50 i started it at 16:00 or around that

[rancid sparrow]

can sm1 send me okada voice changer thru google drive its so complicated to download

[toxic nacelle]

hello i need help with the ai voicechanger

[low shard]

wdym? elaborate the issue more

[toxic nacelle]

huh

[low shard]

what's your pc gpu?

what's the issue? and what guide link did u use?

[toxic nacelle]

amd 6700xt

[toxic nacelle]

laggy and slow in VR

[low shard]

share a screenshot of ur wokada

!give-media-perms 1h @toxic nacelle

[trim sparrow]

should i stop now?

[low shard]

not yet

[toxic nacelle]

1 sec

[trim sparrow]

alright

[toxic nacelle]

@low shard

[low shard]

lemme guess

u used a youtube tutorial?

because those settings are completely wrong and ur using the original wokada

[toxic nacelle]

yeah, noone want to help me around

[low shard]

uninstall everything you got off it

uninstall also vb audio cable, it gives random issues on windows

-realtime

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[low shard]

read the 1st link, wokada deiteris fork

[toxic nacelle]

few mins

may take a while

[low shard]

Alr

[toxic nacelle]

ok ummm i'm no an english speaker soo what should i do

i am mate

[trim sparrow]

@low shard should i keep it going or stop it

[toxic nacelle]

@lofty lichenyou relize i can see deleted text

[low shard]

Welp it's in English because that's the most spoken language internationally

[toxic nacelle]

uunderstandable

[low shard]

It's better you try to read it, use a translator, and let us know for any issues

[toxic nacelle]

yeah but idk where is the download

[low shard]

https://rentry.co/forkvoicechangerguide#download-amd-intel-and-cpu-on-windows

[toxic nacelle]

alrightly

[trim sparrow]

time to end it?

@low shard

[low shard]

Wait still a bit

[trim sparrow]

isnt it spking down like

by a ton?

[trim sparrow]

im pretty sure its at epoch 289

[low shard]

Yes, that's the good thing

This isn't a graph about quality

It's about the machine learning model loss in the training

Read how the tensorboard works in the docs

[toxic nacelle]

@low shardis this right?

[low shard]

GPU: ur amd GPU

Extra: 2.0
Chunk: since youre going to use it in a game, you will need to check the perf value at top left when you click start, then put a little higher value as chunk

Did you do the step 3 and download vac lite?

[toxic nacelle]

not yet

[low shard]

You need to

[toxic nacelle]

1 sec sir

@low sharddone

how how do i set it up

[simple ore]

you're looking at a wrong chart

look at 3 g losses - fm, mel, total

[toxic nacelle]

@low shardok fixed but kinda slow

[trim sparrow]

is it this one?

do i keep it going or should i stop it

(im at epoch 435)

@simple ore

@low shard

[coral estuary]

Hey, I actually just started a chill little server where we share useful tools, automation tricks, and side hustle ideas.
If you're into that, I can send you the link!

[elder coral]

I need help! My applio is still fucked up and i can't make ai covers or TTs voices

I have a good GPU It's Nvidia 3050 Geforce Rtx too

I din't get any errors

[elder coral]

after installing

errors

:sob

[leaden crag]

Hi! what are the differences between client and server? Other than echo/supp1-2? [Okada]

[simple ore]

once again your options are 1) use compiled version that does not require 'installing' step, 2) env\python -m pip install gradio==5.23.1

[elder coral]

I already have that

[simple ore]

then open command prompt and run env\python -m pip show gradio

[elder coral]

in the applio folder

I tried that yesterday

and it needs me a bigger pc

[simple ore]

i mean now and post a screenshot of the result

[elder coral]

is this an

[simple ore]

now run the command I gave you

[elder coral]

i already

[simple ore]

and upgrade it to 5.23.1

[simple ore]

dont bs me, i see thru your lies

[elder coral]

ok now its updating

A new release of pip is available what does that mean?

[simple ore]

now run applio

[elder coral]

ok will work?

[trim sparrow]

noobies if my voice is kinda crackly near higher pitch would it be a good idea to add a dataset of a little of similar artists that have proper pitch or would that mess it up?

[elder coral]

yes it finally worked

[trim sparrow]

im adding falsetto & some higher freq/pitched vocals to the dataset to see if it fixes the problem also took around 60% of the talking out to make sure it doesnt interfere but just there for tone 22 minutes long 650 epochs lets see how this goes i guess

[simple ore]

you either need a singing pretrain or you need to add singing data

[trim sparrow]

ima try this and see how it goes if its still not to my liking ill find a pretrain but i thought i read somewhere its bad to work on a existing pretrain

[pliant pier]

hey so like, is this thing supposed to use 56% of my cpu? it wasnt doing it yesterday

also Im using okada's changer

[urban fractal]

Hey guys quick question with tts-with-rvc
Is there a way to speed up the TTS voice?

[simple ore]

depends on the tts you're using

[urban fractal]

I'm just on windows

[simple ore]

applio uses edge tts, there's a speed parameter

[urban fractal]

tts = TTS_RVC(
model_path="cartman.pth",
index_path="cartman.index",
voice="en-US-EricNeural",
f0_method="rmvpe"
)

[simple ore]

where did you get it?

I assume it is the same edge tts based on the voice

[urban fractal]

https://github.com/Atm4x/tts-with-rvc/tree/main

Yea MS crap so there should be a way to control the speed

I dug into the library but didnt see where its getting the tts audio from and how it generates it

Oh I think I found the speech function in there

[simple ore]

async def tts_communicate(text,
tmp_directory=None,
voice="ru-RU-DmitryNeural",
tts_add_rate=0,
tts_add_volume=0,
tts_add_pitch=0):

rate should be non 0

+5 if you want 5% faster

[urban fractal]

Got this error though
path = tts(text=ai_response, pitch=0, index_rate=0.85, tts_add_rate=5, tts_volume=2, is_half=True)
TypeError: TTS_RVC.call() got an unexpected keyword argument 'tts_add_rate'

[simple ore]

it is tts_rate

[urban fractal]

oh different function there though, but I see the same option here: https://github.com/Atm4x/tts-with-rvc/blob/main/tts_with_rvc/inference.py
I tried tts_rate too and it made no change but no error

[simple ore]

add a debug before `communicate = tts.Communicate(``

and print the value

anyway, that project uses ancient rvc code

[daring sable]

yo

it isnt

like working for

me like

its not talking

fix?

[simple ore]

why neither selected

[daring sable]

is

that y

ty

it sounds robotic

as shit

where can i fix that

sounds robotic

[hallow thistle]

Send your full screenshot.

[hallow thistle]

Set chunk number to around 46 ms for less audio delay. Set "F0 Det." to regular rmvpe. Although you have set extra number correctly, but the audio still sounds robotbic, it can be the voice model itself. Try another one.

[rare charm]

Hello everyone I am starting a project, what i need is a co-host for my YouTube gaming streaming, i already created a working model, but i have some problems, it uses eleven labs api, so ot cost me a lot, i would like to you guys to give me ideas on what voice clone open source good enough to do a Brazilian kratos Voice that i already have 5 minutes of isolate voice to use on the model, also I am using open ai api, o would like to do this project entirely open source, can i have some directions in what to use for my idea?
I have:
I9
4090
64 gb ram

[daring sable]

at 46\

it goes red

the perf

[hallow thistle]

If you use W-Okada with a game or something, increase the chunk number until perf number turns green.

[urban fractal]

I feel dumb. I was setting it to like 3 or 5. Setting it to 50 sure does the trick. LOL Thanks

[simple ore]

20 as in 20% faster

it does not do more than 50 I think

[urban fractal]

good to know

[trim sparrow]

what is this?

[hallow thistle]

This is the UI of fork W-Okada the realtime voice changer.

[trim sparrow]

do you have a link perchance

[hallow thistle]

What is your PC GPU?

[wanton valley]

https://tenor.com/view/stare-christian-bale-handsome-gif-13379950

This might be the end if mainline-colab doesnt get fixed

[hallow thistle]

That will never get fixed. There's another better one to use.

[regal pivot]

What's the best way to do a duet locally? Right now I'm using ultimate-rvc, which I'm not sure if I can do it there, so I assume I need to separate the vocals another way, but I'd much prefer locally. Specifiacally male and female vocals.

[knotty moth]

unfortunately you can't download this great model

[regal pivot]

I tried MVSEP and it didn't really do that good of a job. I'm trying UVR5 and it just refuses to separate the male backing, messing around with different models to hopefully find a system that works.

[trim sparrow]

how do i continue training a model

i wanna add a new dataset to it but i dont wanna re render the old dataset that was already processed

[simple ore]

you can continue with the same dataset

[trim sparrow]

wdym?

[simple ore]

I would not advise trying to use a finetuned model as a pretrain for a larger model

the base has to be generalized enough

and by training a finetune model you're cutting off some of the parts

[trim sparrow]

no im just adding more singing/rhytem to it

[simple ore]

just start over

[trim sparrow]

ts took like 5 hours😭

[simple ore]

happens

[trim sparrow]

so i cant just add the new audio to the same dataset and process it or will that basically just restart it or

[simple ore]

as I just explained

imagine you have a block of marble and you're carving a statue out of it

a pretrain roughly cut shape of a human being, then you finetune it by cutting it into a shape of a bald man with a beard

and now you want to turn that into a statue of a woman with big tits holding a vase

did I make a point?

[trim sparrow]

sorry im not 100% with the vocab for a lot of the ai stuff even if lets say i created the model from scratch and its just singing/humming etc and i just wanna add more would that still fall in the criteria of that

yeah just not 100% on the vocab for that stuff but i get ur point just not sure of what a pretrain or finetune is

but either way ill just start over im probably gonna go to bed anyway and let it sit

[simple ore]

I gave you totally not AI-based explanation

pretrain is the weights you're using as a base

finetuning is using a small dataset to finalize the voice model

[trim sparrow]

ahhh okay

n quick question for the dataset does it have to be in just one file for all the audio or can it be multiple?

[simple ore]

it can be many files in the same folder

ideally they need to be same quality recording-wise

[past barn]

can someone walk me through setting up realtime voice changer using these ai models

[trim sparrow]

alrighty thank you also do you have any clue on how to delete old models its taking up storage for kaggle😭

[cobalt vine]

can someone tell me what happened

[pastel oak]

Colabs dont work

Youll need Kaggle

[trim sparrow]

use kaggle it worked good for me

[pastel oak]

https://rentry.co/ForkVoiceChangerGuide#online-alternatives-colabkaggle

[trim sparrow]

should i stick with fps16 or fps32 @simple ore

[cobalt vine]

damn i just paid, few days ago for it

[simple ore]

full error

so bad token?

or whatever it says

[hallow thistle]

Make sure you censor your ngrok token before sending screenshot, otherwise someone would snatch your token for their things. What is your PC GPU? Because W-Okada can work locally without having hard time finding ngrok token for Colab/Kaggle.

[thorn pollen]

Hii is there an ai which can like fake mouth movement on a video where the persons mouth is closed

[simple ore]

too late for that 🙂

[cobalt vine]

i did refresg

[simple ore]

fp16 is not stable, more often than not the model just explodes

you can see that by a bunch of triangles shown on the charts

[trim sparrow]

thank god i just put it at 32 and got in bed😭

[hallow thistle]

So what is your PC GPU?

[cobalt vine]

gtx 1660

realtime collab cant use ?

[hallow thistle]

With that GPU, it should work. Just won't be that fast like one in Colab/Kaggle. https://rentry.co/ForkVoiceChangerGuide#download-nvidia-on-windows

[cobalt vine]

it is but it sometimes very laggind and dc when i play fiveM

that why i need collab

[hallow thistle]

That's pretty much it. Most of W-Okada Colab notebooks are broken. Even using fork W-Okada Colab can get your account to be terminated, especially when you have free tier.

[cobalt vine]

but i buy premium a few days ago

its not free

[hallow thistle]

No excuse. If you have bought Colab Pro or compute units, make sure you keep it set to T4 GPU instead of A100 or L4, because these two GPUs eat more compute units than T4 one.

[cobalt vine]

yup i have it

[hallow thistle]

Kaggle gives an option to use two of T4 GPUs for free 30 hours a week. But if you have bought Colab Pro or pay as you go, then use it to your advantage I guess.

[cobalt vine]

So does RealtimeVC mean it can't be used anymore?

[hallow thistle]

Don't worry about it. You can still use fork W-Okada. But make sure you got a right ngrok token if you wanna run it on Colab or Kaggle.

[cobalt vine]

got it, thank you so much

[hallow thistle]

What's up?

[lofty lichen]

@steel forge warning while sleeping?

[steel forge]

What

[lofty lichen]

i got a warning case 92rc2Ub

while sleeping

[knotty moth]

there are many other cheaper GPU rentals

[lofty lichen]

Reason: Be respectful to other server members and do not post NSFW content

lmao

[hallow thistle]

!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.

[lofty lichen]

while sleeping

[steel forge]

You were warned for a gif you sent 9 hours ago of an animated walking penis 😭

Just don't send that gif again

[lofty lichen]

lmao then remove the gifs from the server

[steel forge]

No

[lofty lichen]

yip

[hallow thistle]

Sending what you got warned by mod into public chat won't fix your issue.

[steel forge]

Ok fine i guess

[knotty moth]

just dont

[lofty lichen]

anyone can search it on the gif at the side of the chat

then tell owner to remove gif access

[steel forge]

Its the fact that it was a penis

Yes that's the truth

[lofty lichen]

its not considered nsfw by discord since its there

[steel forge]

By tenor*

[knotty moth]

that gif should also be reported in the site itself since I'm sure it'd violate its TOS

[steel forge]

And it doesn't really matter, we make our own rules in our own server. Our staff saw that as too NSFW, so they removed it.

So if you can avoid sending things that might be on the line, wed appreciate it. We have kids here

[lofty lichen]

not my problem

[steel forge]

Ok

[cobalt vine]

@hallow thistle still the same

[hallow thistle]

Don't be a dick here. Anyone can make a mistake. The warning should make you aware of what you're doing. If you think you were unfairly treated, you can write down your appeal in AI Hub's website and let what other mods think about you.

[cobalt vine]

what the issues @hallow thistle ?

[steel forge]

You need to scroll to the side to see the rest of the error

"Failed to accept"

[knotty moth]

voice changer colabs tend to have issues, another option is this
https://www.kaggle.com/code/suneku/voice-changer-public

[hallow thistle]

Although you can still use Colab for other AI programs like Applio or UVR5.

[cobalt vine]

many of my numbers can't be used on kaggle but ok i'll try

[hallow thistle]

I've never had any problem using W-Okada on Kaggle. The error can happen when you skipped some important steps like where to start the notebook and the settings.

[long oyster]

is this setting voice good enought guys?

[hallow thistle]

Use Virtual Audio Cable lite instead. VB-Cable gives random issues to Windows users, as many people complained to me.

[steel forge]

Well

If your current cable works

[hallow thistle]

[steel forge]

That's fine

[steel forge]

Yea looks about right for everything

I'm so used to seeing nightmare settings so yours are nice to see

[hallow thistle]

If you have any problem using VB-Cable, you can try switch to VAC next time.

[pastel oak]

Force fp32 mode on in advanced for improved quality

[past barn]

yo

i need help

my voice changer is super laggy when i run it

how do i fix that

im on AMD

[low shard]

Server can also have less delay using wasapi or asio

[low shard]

Share a screenshot of ur wokada and what u are using it in

[past barn]

i literally cant share a screenshot

https://prnt.sc/do6N0dD481x2

[past barn]

there u go

[sharp crescent]

Is it posible to tailor AI Voice to your own, like if someone is making AI model for me, they can also tailor it to my voice so that the sound is better, more fuild, less choppy ?

[umbral acorn]

Hai. Can anyone show me how to train model in colab? I just came back after a year so i don’t remember anything

[past barn]

@low shard

[low shard]

!give-media-perms 1h @past barn

[low shard]

Holy shit you're using an ancient version

That's prehistorical

Lemme guess, you used YouTube tutorials?

All video tutorials are outdated asf

[low shard]

Did u first try to check if ur PC GPU is good enough

[umbral acorn]

Its not good thats why i wanna use colab

[sharp crescent]

Where can you find the latest voice changer build ?

[silk pike]

You can check out this guide 🙂
https://docs.applio.org/applio/getting-started/other-alternatives

[umbral acorn]

Alright thank u

[sharp crescent]

Anyone got info since i only know of youtube ones ?

[jolly tangle]

Github

[low shard]

-realtime

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[low shard]

1st link

Wokada deiteris fork

@sharp crescent also what is your PC GPU name?

[sharp crescent]

rtx 3060ti

[low shard]

Wokada doesn't support it yet

[low shard]

Nice, get the windows Nvidia one

[jolly tangle]

When will it?

[low shard]

Who knows

RefineGAN is experimental

It is disabled even in applio main branch

Since it's not that stable yet

[sharp crescent]

Thank you.

[pastel oak]

Someone made a voice model of himself and merged that model into the model he wanted by small percentages to have the tonuation of his in it. I tested it myself but didnt do much, choppyness is usually result of bad model since realtime is way more demanding to be clean

[silk pike]

Also, the colab notebook gives you only 4 hours of free runtime daily, but if you want a longer runtime, then you can use Kaggle notebook, they offer up to 30 hours of free GPU/TPU time per week

[golden walrus]

i guess it used system in and out

wah ?

it's in sound setting

[outer fog]

anyone know why my voice is cutting out while talking

[knotty moth]

same as for discord

[outer fog]

my voice only cuts off while using echo feature on noise, anyone know why?

[low shard]

Max 4 hours, but it can be different daily

One day could be 4 hours

Another could be 1

Another could be 3

It's not granted time

Kaggle has granted time instead

[low shard]

Share a screenshot of ur wokada

!give-media-perms 1h @outer fog

What game are u playing

[unreal linden]

Hey, I'm trying to play the app in server mode but I can't see my headphone in windows WASAPI

[stone lynx]

I want to use my gpu but it shows me there is only cpu option to choose

how can i use my gpu instead cpu

[atomic kindle]

how to fix mic background noises?

is it possible to kinda silent them?

because my mic is ig broken or something

the ai voice detects these noises as my voice and there is like a permanent moaning

and my voice has a kind of an echo without even activating echo

[low shard]

did u follow correctly the wasapi guide?

[low shard]

check sup2

[low shard]

elaborate your issue:

• what's your pc gpu
• what do u want to do
• a screenshot of the program

!give-media-perms 1h @stone lynx

[stone lynx]

my pc gpu is 6700 xt
i want to use my gpu instead of cpu

[low shard]

holy shiet

that's an anciet version, an old version of original wokada

over a year old

never follow youtube tuts for wokada

they are all old

also vb audio cable gives random issues on windows

[low shard]

uninstall all you got off youtube

-realtime

[patent trellis]

💻 Local Realtime RVC

Guides for Programs that use RVC Models in Realtime for Calls/Games

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

ONLY the latest alpha comes close to the Deiteris Fork performance, older versions in youtube tuts are way worse. GUIDE

• **RVC GUI Mainline Realtime**

Unavailable, the guide is outdated and the program is worse compared to the ones above, and much less updated

[low shard]

read the 1st link, wokada deiteris fork

[stone lynx]

i downloaded but i cant find the run http

[low shard]

there's none

you gotta run the .exe

it's a different version of the program, the deiteris fork

please be sure to not skip any steps of the guide

[stone lynx]

wait it says

It says it will open in the browser.*

so i cant use it as program anymore right?

[low shard]

it is still a program, it runs locally on your hardware

both programs always used a web user interface made in javascript and typescript

the only difference is that original wokada made it's own browser window to open the local hosted url, which iirc is removed in deiteris fork for sometimes cuasing performance issues

[knotty moth]

https://rentry.co/forkvoicechangerguide#why-does-it-run-in-a-browser-and-not-its-own-window

[low shard]

i just updated the guide to tell users to check that part around the parts where it says it will open in the browser

[pearl zealot]

How can I completely uninstall mangio rvc from my pc?

[low shard]

i think u already got helped there #✦│chat message

[pearl zealot]

yes :))

[stone lynx]

now it works ty

which one is better

to use

[low shard]

Use rmvpe onnx for best quality

Fcpe is good only for being faster but shittier quality

[stoic oasis]

Hello
Is there a voice changer where I could modify my voice just partially? Something like voice model changing voice by 20% let’s say. Aim to mix, modify, not transform completely. I was checking w-okada and some other software, either they don’t have such features or I don’t know how to use them. I’m experimenting.

[stone lynx]

hello again

When I speak, I hear a small voice repeating the last word I said.Is there a way to fix this

[low shard]

I don't think so, maybe @pastel oak got an idea though

[low shard]

Share a screenshot of ur wokada

[pastel oak]

Dont think so
Alternative i can think of is making a model of your own voice then merging it with one you want with percentages

[gloomy mortar]

guys i have a question

when using the index

i dont get this problem without index

but when i try index i get this

error