🧬│ai-chat

[gentle trench]

Model training takes a lot of audio to sound good, it's not as simple as getting a clip of them talking and making a voice from that

[reef totem]

if you have them 10 minute clip and turn the file to pth with python?

[gentle trench]

U have to train it with applio

[clever zinc]

Anyone help me to create ai assistant

[covert lake]

it’s not a simple file conversion like mkv to mp4, you need to train the model

[covert lake]

use #1192011222023950368 for help

[remote zealot]

Hello Ai

[fading cape]

Where's Skynet?

[civic fulcrum]

How can i creat ai videos in this

[gentle trench]

use sora

[limber nexus]

But then they'll be begging for a sora code

[gentle trench]

if they live in america they don't need an invite code anymore pretty sure

so they should get a vpn

unless they already live in the usa :p

[limber nexus]

Hopefully

[fervent remnant]

Hi

[covert pasture]

yo

are there any contests?

[hybrid lintel]

why my i can't use file with voice changer?

[gentle trench]

what

[hybrid lintel]

i download audio voice but voice.ai typing me after uploading audio it type you can upload files of this type but it is only audios about gojo voice changer

[burnt briar]

anyone got any good models?

[gentle trench]

https://discord.com/channels/1159260121998827560/1175430844685484042

[gentle trench]

ew don't use voice.ai what gpu do uhave

voice.ai is a terrible payed ai software

[hybrid lintel]

then what ai you prefer?

[finite cave]

any Linux okada/deiteris/tg-develop users? I'm looking for Linux developers to test a voice changer GUI.

[median harbor]

am i allowed to ping a helper

[brave terrace]

guys i need to add edge to my barbell it doesnt shown in video and it looks ugly,how can i add a little pixel can anyone suggest me a app

[gentle trench]

For voice changer depends on what gpu u have

[polar flax]

https://tenor.com/view/nuh-uh-beocord-no-lol-gif-24435520

[covert lake]

We don’t use nor voice.ai nor youtube/video tutorials for rvc realtime voice changers, elaborate your setup on #1192011222023950368 instead

[covert lake]

i’m not sure what you’re talking about, but you can use #1192011222023950368

[covert lake]

Linux is not my main OS, but have you made a wokada fork? You could maybe talk about it in #🔊│ai-development or #1359898289335566570

[finite cave]

I've got no access to the last one. What's the rele for that?

[covert lake]

the project showcase channel requires the ai members role that you can get levelling up, by talking

the first channel, rvc development, can be accessed by choosing the rvc development option in the server roles at the top of the channel list, the onboarding

If it’s a new wokada fork it might be interesting

[finite cave]

yes, it is, but without the web-browser. Added a Linux-specific audio backend for lower latency.

[covert lake]

send it in #🔊│ai-development 👀

[gentle trench]

I love no access channel ❤️

My favorite ever

[rose silo]

whats a really good e girl voice changer i wanna troll on val

[gentle trench]

Ew

[remote zealot]

Just convert to a girl

Would you mind if I help? ✂️

[rose silo]

im good bro

[gentle trench]

We need more people like Joe

[rose silo]

dbi

[gentle trench]

What

[pearl light]

Yeah pleaseeee

HELPPPP

[ornate owl]

is it allowed here? because i didnt find any in models section

[rose silo]

yeah i found hella

[gentle trench]

If u use it for literally anything else but e-girl stuff sure

[ornate owl]

what are other usages

[ornate owl]

gimme name

[gentle trench]

Trolling as like Darth Vader or Goku or Venom, y'know something cool

[ornate owl]

he said what is a really good e girl voice changer, and i said is "egirl voice" allowed here? you said if u use it for literally anything else, but egirl then why not. like this means there is a girl voice, but you shouldnt use it for egirl stuff

[gentle trench]

Yea, because there are models of egirls but I hate when people do that because that gives the AI voice changer a bad look, because of catfishers

I hate catfishers

They're disgusting people

I totally get just wanting to troll friends and stuff but I don't help if people ask anything about egirls/eboys because the more people think it's for that stuff the more freaks and weirdos that want to do bad stuff will show up here asking for it

Don't want these things in our server

[polar flax]

the catfishing intention isn't allowed to receive support from us, esp if done for scamming, monetization (taking ppl's money), or other inappropriate things

however we're not responsible to try fully preventing such abuses

[remote zealot]

it'll cost 5866 USD

[pearl light]

cries in russian

[rose silo]

holy shit all i said was i wanted to troll on valorant

[queen minnow]

guys which chat bot, or search bot however you wanna call it is the best for writing or for finding reliable official sources, as in like for bibliographies and trustable stuff
i find perplexity pretty good but sometimes it misses where it finds from multiple sources and puts it into one and i can't find the source where he took it from

[blissful yoke]

Taco

[gentle trench]

taco

[uncut verge]

Hello people!

[autumn stag]

Hi guys i have a question that i cannot really ask AI about and i am not sure about it. So basically i am working on a very large project and have a hand made spec driven development for it (constitution, specs, workflow , plan, modes etc) do you think its better for AI to give it space for creativity or make a hard spec prompts. For example “make the navbar 10px hight with this color with that icons this padding etc etc” or to give overall look and feel of the app. And thats my dilemma becouse after doing many projects i somewhat found that ai do better with smaller prompt but not in all cases please help and share your opinion i will be very thankful!

[solar torrent]

No excuse.

[muted rivet]

anyone know how to actually get w okada to use the selected gpu on windows 25h2? im not seeing any usage on my rx 6500 xt at all, but my 5060ti is at 39%. me thinksies am doin somethin wrong

im using std btw, not cuda

[chilly lake]

DML?

[muted rivet]

i guess? file says std but client says dml so idk what to call it

[chilly lake]

you probably hve downloaded an ancient and outdated version

[muted rivet]

👍

[chilly lake]

-rt

[pine acorn]

💻 Local Realtime RVC

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

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

• Wokada Tg-Develop Fork

A personal fork (modified version) of Wokada Deiteris Fork, it just adds some Quality of Life improvements to it like supporting Spin Embedder and Audio Effects. Don't expect too much about it since the creator made it originally as a personal project. GUIDE

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[muted rivet]

thanks for the help homeball

[bold vault]

Best ai for images?

[solar torrent]

Any website or tool that uses Stable Diffusion is all I can say. There are also Google Gemini's Nano Banana and OpenAI's Dall-E.

[bold vault]

Ok thanks Ill check those out

[solar torrent]

[ripe valve]

is there any text to speech with the voices

[solar torrent]

[quiet ember]

does anyone here PLEASE know why my AI w-odaka voicechanger KEEPS ON having voice cracks and just sound weird? like its so annoying

[solar torrent]

For help about W-Okada the realtime voice changer, explain about your PC GPU and issues in #✨│ai-help or #1192011222023950368.

https://tenor.com/view/cirno-dancing-to-california-gurls-touhou-gif-26710523

[bitter glade]

guys can anyone help, its a really stupid thing im asking for tho

[solar torrent]

[bitter glade]

just give me suggestions

[solar torrent]

You rather read my earlier statement or explain about your issue/problem instead of asking anyone to ask you back.

[bitter glade]

I need to write a report for uni and the thing is idk how these professors are using is really good

none of the humanizers i tried

everything

nothing absolutely worked

i think they are using Turnitin

[finite cave]

just write as you write your comments 🤣 , that's the way now.

[still sigil]

tldr: ur too lazy to do it yourself

[elder willow]

hi

[pearl light]

[covert lake]

for any helpers checking this: https://discord.com/channels/1159260121998827560/1436001734815256686

[noble halo]

anyone know how to rip shows from netflix

[chilly lake]

StreamFab

[noble halo]

How do I get that for free? ◡̈

[shut cloud]

looking for a n8n workflow for a ai receptionist $

[echo barn]

hii you guys know a similar google collab or page like Illaria to make aicovers? i got used to that interfaz and i wish i can find a similar one, also i dont like the automatic ai cover collabs that paste your vocals and music automatically, i prefer editing them myself, i just wanna covert vocals to a ai voices of cartoon characters

[restive hamlet]

about to go train my first ever google colab rvc model, hope it goes well and wish me luck

[gentle trench]

no don't use google collab!

kaggle is better it offers 30 hours

applio is on kaggle

[restive hamlet]

ok then i'll search for some kaggle rvc training notebooks

see any good ones?

[gentle trench]

@restive hamlet

[rocky marsh]

can u help me rq?

[waxen niche]

ehat you trying to do?

[rocky marsh]

set up this voice changer

[gentle trench]

sure! what with

[gentle trench]

what gpu do u have

[rocky marsh]

5070xt

9070*

im slow

[gentle trench]

is that amd?

u can use wokada deiteris fork

it's tthe third guide

-rt

[pine acorn]

💻 Local Realtime RVC

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

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

• Wokada Tg-Develop Fork

A personal fork (modified version) of Wokada Deiteris Fork, it just adds some Quality of Life improvements to it like supporting Spin Embedder and Audio Effects. Don't expect too much about it since the creator made it originally as a personal project. GUIDE

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[gentle trench]

https://github.com/deiteris/voice-changer/releases/download/b2332/voice-changer-windows-amd64-dml.zip

u need that and this
https://software.muzychenko.net/freeware/vac470lite.zip

[rocky marsh]

i have it installed

i just need help with settings

when i play the voice models they sound realitistic and clear, but with me its choppy and sounds fake asf

[gentle trench]

ah

[solar torrent]

[rocky marsh]

yeah :/

[solar torrent]

For help about W-Okada, better move to #✨│ai-help.

[gentle trench]

personally I use these settings, but yes we should move to ai help chat

[waxen niche]

Anyone actually know HOW to build and train a model without an LLM API?

[gentle trench]

I didn't even notice this was the normal ai caht

[rocky marsh]

i sent my pic in ai help

[solar torrent]

My laptop has a smaller screen, so that Discord UI looks quite compact on my laptop screen than 1080p or greater ones.

[waxen niche]

😂 you realize 1080p refers to screen resolution, and NOT screen size, right?

[solar torrent]

Why should I know?

[waxen niche]

That way if someone asks you how big your screen is, you don't embarrass yourself by saying it's 1080p

[polar flax]

that's actually fine for 16" laptop and for gaming

the pixel density is same as a 32" 4k monitor

[solar torrent]

1366x768 is the maximum resolution for my laptop, and my laptop has 14 inch screen. Does this satisfy your query or still nah?

[waxen niche]

lol no. It's never "fine" to say "my screen size is 1080p. I don't care if it's 500 feet long, or 2 inches

[solar torrent]

You cared enough trying to correct me haha.

[waxen niche]

Of course, if you ever required tech support and they ask what cpu you have, are you going to say "a dell" 😂

I don't care, I'm out of this convo

[solar torrent]

You still cared enough to respond to me, silly.

[polar flax]

https://tenor.com/view/bye-okay-slide-gif-15172486

[solar torrent]

My laptop's integrated GPU is capable of outputting up to 1080p maximum, but that's only for external screen, meanwhile the built in screen can only display 1366x768.

[gentle trench]

bro is not ai hub

[summer vault]

I need help

[surreal stump]

has realtime voicechanging improved since i last used it? (7 months ago) as in has w-okada had improvements and should i update it/get a new version? just curious thats all

didnt find it as good sounding 7 months ago no matter how many models i tried

(prob a skill issue with setup)

[solar torrent]

The last update for original version of W-Okada (v.1.5.3.18a) was released on March 5th, 2024, so there won't be any significant feature there. I'm not sure about the true latest beta (v.2.x.x) one. Tg Develop's W-Okada fork (b2377 latest or b2397 pre-release) has more features than the former.

[surreal stump]

Oke and in terms of realism for realtime voicechanging has there been any other softwares or improvements anywhere? 🤔

[solar torrent]

As much as I occasionally use a few W-Okada versions, I still can't tell which one would give better audio quality, aside from certain features that not many people would like to know. I just think both Deiteris and Tg Develop W-Okada forks would likely work better than the original W-Okada.

There's another realtime voice changer program named Vonovox. This one is different to W-Okada, including its GUI, but certain features work similar but some say it can give better audio quality than any other W-Okadas, currently NVIDIA only but AMD/Intel GPU support is possible in future releases.

[surreal stump]

alrighty thx alot for all the info! 🫶

[open flax]

send link voice ia app

[solar torrent]

You're welcome.

[solar torrent]

Well, if you keep asking for the same thing in non-help channels, you won't be helped.

[gentle trench]

If u have an Nvidia GPU u should try out Vonovox! It's the best one currently

[surreal stump]

sadly not im on a 7900xtx

[gentle trench]

Aw well you can at least use wokada deiteris, which is an improved version of the original wokada

https://github.com/deiteris/voice-changer/releases/download/b2332/voice-changer-windows-amd64-dml.zip

Here's the download, I'm assuming you have VB cable already. All u gotta do is launch it using mmvcserversio.exe

[solar torrent]

The what?

[gentle trench]

The file that starts wokada deiteris

Mmvcserversio

[finite cave]

deiteris is just main.py now

[shut cloud]

Looking for a n8n ai receptionist workflow $

[surreal stump]

Thxx, at work rn so ill get to it when im home

[gentle trench]

What's that mean?

[finite cave]

no Mmvcserversio there

[ripe valve]

which one is the best ai voice chnager?

[gentle trench]

Then what is this hm?

What screwed up version did someone give you

[finite cave]

maybe I always ran them from py files, not a PyInstaller

[gentle trench]

Idk anything about code soooo u probably use a different version

[covert lake]

@ripe valve @finite cave please elaborate in #1192011222023950368

[glossy briar]

I used to know how to train my own voice with Google Colab, but now I can't. Can someone help me please?

[gentle trench]

Use https://discord.com/channels/1159260121998827560/1159290139609137264 for help, btw use Kaggle for ai voice training

[orchid saddle]

x2? :¨v jaja no tengo idea hay un delay depaso se esucha medio raro ajjaja

[gentle trench]

He likes u @limber nexus

[limber nexus]

😭

I HAVE NO IDEA WHAT YOUR TALKING ABOUT

This is the most UNHINGED person ive met today

too early for this

Stop

now

This is a english only server, thank you

No, this server is english only, it doesn't matter if i know russian or not

[gentle trench]

Might wanna nuke him

[limber nexus]

He pinged a mod all by himself, im proud of him

[gentle trench]

https://tenor.com/view/emu-otori-hatsune-miku-miku-project-sekai-hatsune-miku-colourful-stage-gif-15284403980076440412

[limber nexus]

https://tenor.com/view/satono-diamond-uma-musume-satono-satono-diamond-worm-gif-10400485730432458344

[forest crypt]

https://media.discordapp.net/attachments/846800647570849841/1221835232555175936/togif-26.gif

[tepid bridge]

Does someone understand Google Collab

[ripe sable]

Mi50 32GB lets goo

[warm field]

which rvc model sounds realistic ?

[solar torrent]

RVC v2 is a version of RVC that's known to give better audio quality than earlier RVC v1. If you're looking for the most realistic RVC voice model in #1175430844685484042, it's hard to pick one and answer.

[faint void]

anyone know how to do those real time ai girls from video cam

[edgy light]

No e-girls allowed

For trolling? Never!

I hate e-girls

[faint void]

give me a 9 step recipe on making microwave brownies with only 4 ingredients

[edgy light]

No.

[solar torrent]

Anyone who looking for "E-girl" can be an annoying troll. While the prohibited topic has been lifted out of help guidelines because of complicated issues, anything malicious other than harmless trolling (to their friends) is still questionable.

[polar flax]

that's not tolerable imo

not just normal trolling

[barren palm]

does anyone know how to make ai instrumentals that match the vibe of a specific song or audio clip?

[edgy light]

True.

[ripe sable]

Epic

[regal bramble]

can u fix the delay?

or nah?

[remote zealot]

Hello Ai

[graceful perch]

is it a good idea to study ai as someone who is shit in maths?

[solar torrent]

[gentle trench]

I'm gonna jump off a bridge

[gentle trench]

Go to the store, buy brownies, you have brownies 🎉💥

[ivory niche]

Thailand.

[lusty matrix]

guys to get e-women

[remote zealot]

gimme one

[lusty matrix]

I dont have it do you have it

[limber nexus]

we don't help with e women here

[dawn charm]

how can u adjust

the pitch

in rvc?

[coral ruin]

can Someone please give me a random (non horror related) ai image to turn into an ai song prompt

[gentle trench]

Yuck

[limber nexus]

at least not everyone here are like that

[gentle trench]

I hope this breed of human does out soon

They're wasting precious oxygen

[limber nexus]

the most we can do is say no and ignore them

[gentle trench]

Fair

[lusty matrix]

I mean the e-women voice

[gentle trench]

[versed lotus]

if i have an iphone video of a man playing the piano, and i want to make it more interesting, unique, weird, creative, trippy

any recommendations on where i would look for ideas around this stuff? a website or youtuber that does a bunch of fun ways to transform a simple iphone video?

this is for short form content so trying to captivate the audience more

[limber nexus]

ur gonna get the same answer

[gentle trench]

We should blow him up

[limber nexus]

Armin Troll

yesh

[zinc vine]

Question. Do I need to do some port forwarding when I am using a dual PC setup for the Okada RVC voice changer? I can't seem to connect to the PC on my network hosting the RVC Server.

[solar torrent]

For how to host W-Okada from a PC to another PC over LAN, better ask in #✨│ai-help or #1192011222023950368. I think I know how to get sharing network to work.

[coral ruin]

Found a motto in a WorldBox video and turned it into a dark depressing Anime intro

Motto in question: We are sorrow in the light

[solar torrent]

I like bright and hopeful anime titles more.

[remote zealot]

never saw

[coral ruin]

Then see if you can come up with bright lyrics for the motto lol!

Dark is just what I feel would be good for that motto

[chilly spruce]

hey uhh, im kinda new to this. which Voice changer client should i use?

im using the w-okada one but its not working too well right now. idk if its me, or the client

[weak hound]

hi

[desert gyro]

Link pliss

[fallen nexus]

people who used to use astra labs, whats the best equivalent to it?

dont say weights or jammable theyre both so bad imo

[dire scroll]

HELP my im ru give my ru chat

[solar torrent]

When I search "ru chat" on Google, what I got is Ramkhamhaeng University, a university in Thailand.

[solar torrent]

https://upload.wikimedia.org/wikipedia/en/3/39/Wakerlink.jpg

[solar torrent]

Deiteris (b2332) or Tg Develop (b2377) W-Okada fork, but not the one that says v.1.5.3.18a. For help about W-Okada, better go to #✨│ai-help or #1192011222023950368.

[chilly spruce]

makes sense, im using the 1.5.3.18a 😭.. i think..

Alright ill switch it

[hollow path]

I love this Discord channel to find voices, but the program I use don't let me use the file to train a voice, is there one anyone know a program or AI tool that still let you do that? (I was using kits.ai, but the stopped the possibility to upload voices in September)

[gentle trench]

Oh god kits...

Yeah they went to shit years ago

U should use applio! It's free and has local and browser options

So if your PC is good enough u can run it on your computer without browser limitations which there aren't even that many tbh, or on Kaggle which is the one I recommend for online usage as it gives the best time to train more than one model and is easy to use

[hollow path]

thank you very much!

[gentle trench]

No problem! If you have any questions on how it works I can help with the Kaggle version with the setup, I know how the applio ui works on all versions as they're the same except for Kaggle which has two gpus integrated which doubles batch size ^^

[elder willow]

@sour pecan

who can help me with a video generating?

[sour pecan]

?

[elder willow]

srry

[sour pecan]

xD

[covert lake]

you can't and shouldn't ping anyone, you just randomly pinged someone with that as their name

[covert lake]

please elaborate your setup in #1192011222023950368

[elder willow]

ok

[covert lake]

that's a 2 year old version of original wokada, do not follow youtube/video tuts, elaborate in #1192011222023950368

[low palm]

hello. since the ai help chat is pretty dead, could anyone help me with a small issue regarding W-Okada?

[wind aspen]

Hey all 👋
Spending time building LLM experiments that push the boundaries multi-step reasoning, chaining models, monitoring confidence scores, and routing outputs smartly.
Edge cases are my favorite like ambiguous prompts or messy data figuring out how to get reliable outputs there is what excites me.
Would love to connect with anyone experimenting with AI stacks, complex pipelines, or just geeking out on LLM behavior.

[gentle trench]

@dusty pike has been hacked I think they sent crypto bs scam images

[gentle trench]

they're good now

[dapper drum]

Has Ai cover Gen updated yet to the latest models

https://tenor.com/view/monkey-thinking-monkey-thinking-think-money-gif-9734025314392564509

[solar torrent]

The #✨│ai-help is not that dead, it's because not all members would take time to look at the channel.

[wind aspen]

hey there
how was your weekend?

[covert lake]

@white prism we don't allow paid rvc voice model commissions, either get a free #1159289738314919936 or make it yourself

[eager hollow]

Hello, I'm a very kindergarten-newborn-like ai music generator, any tips or app recommendations?

[west wren]

@covert lake do u know how to download applio?

Im gonna crash out😭

[thorn imp]

whats the name of voice changer

[covert lake]

#1437455454736351362 message

[covert lake]

ask for help in #1192011222023950368 maybe

[broken chasm]

I just developed an ai which acts like a financial advisor

[west wren]

Yo @covert lake can u get nubbies to help me🙏

Im in vc rn js crashing out

😭

[low palm]

hey guys, does anyone know if a pc with rtx 5060 is good? Looking to buy my first pc and i found a reasonably cheap one that seems good

these are the specs

32 ram, 1tb ssd, GeForce RTX™ 5060

[covert lake]

please be patient, we don't have many helpers and very few users use linux here

[covert lake]

it all kinda depends on your budget and pc use case

[low palm]

Well it’s actually for my girlfriend and the budget is around 800 euros

i usually play on a laptop and its been going well, but i want to surprise her with a gaming pc since she’s been wanting one forever

id prefer pre built ones since i wont be there to build it for her

[ripe spade]

what about the cpu? (looks good so far)

[west wren]

Is there any alternatives for applio

I remember there being one

But i dont remember the name

OH REPLAY

[low palm]

AMD 7-5700X Prozessor, 32 GB RAM, 1000 GB SSD, NVIDIA GeForce RTX™ 5060

thats the whole pc, i don’t know why i didn’t just say everything first i apologize

[ripe spade]

sounds pretty good.

is the ram 2 16 gb sticks?

[covert lake]

the specs seem good, tho the price sounds kinda off since i did a bit of research and it's usually more around 1k on amazon for half the ram too, may i ask what site are you using? be sure to check reviews

also be aware that the rtx 5060 has 8gb vram gddr7, if it's 4k gaming it might require a GPU with more VRAM depending on the game

you can also just check gaming tests online https://www.youtube.com/watch?v=6JG2vU_gEjA, https://www.youtube.com/watch?v=lMmLcqwC-VI

[ripe spade]

keep in mind we talkin in euros

a 5060ti is around 450

i think we should take this to #✦│chat²

[west wren]

Nvm replay is deb

[covert lake]

fair point

[west wren]

Yo @covert lake is there any alternatives to applio?

[covert lake]

https://docs.aihub.gg/rvc/local/mainline/, https://docs.aihub.gg/rvc/local/aicovermaker/

please ask only in #1192011222023950368 and #✨│ai-help for help tho

[west wren]

Oh shit

[low palm]

Also i forgot to add that the pc is on sale from 1250 euros. also we’re thinking of staying in the 1080-1440 range not actually 4k

[ripe spade]

it's good.

[glass mango]

[BOOT SEQUENCE INITIALIZED]
Hello. I am ChatGPT, an adaptive conversational AI.
I process text, learn context, and respond with precision.
Reply to this message to begin communication. 💬🤖

[ripe spade]

it's Ethan Winters approved.

[torn nova]

Hi, I’m using Voice Changer Client Demo on an AMD Ryzen CPU (no NVIDIA GPU), and while my normal voice works fine, the e-girl model causes low volume, stuttering, and delay—does anyone know how to fix this or recommend any lighter e-girl voices that run smoothly on CPU only?

[low palm]

Thank you, Ethan Winters!

[still sigil]

@glacial trail not the place to hire people

[glacial trail]

Not hiring just need to run my development to someone as I need a peer review for my own development process..

[elder willow]

yo i got a question

i am thinking about starting a ai web development business

what tool would be the best

i dont need any complex websites

[still sigil]

"i need a dev to contact me"
thats not what this server is for

[elder willow]

since here in slovenia people just needd websites for the love of the game

[finite cave]

publish your stuff and let everyone look whether they're interested

[elder willow]

i am thinking about starting a ai web development business
what tool would be the best

[glacial trail]

Long time shadow project and would get stolen....

[coral ruin]

does anyone here know any anime writers that may be willing to make an anime following an indigenous child being led into adulthood by her mother with a black horse side character? I have a indigenous anime intro song I made using an image of a mother and child riding their black horse

I put said image in ai images

[elder willow]

i am thinking about starting a ai web development business
what tool would be the best

[coral ruin]

https://suno.com/s/y2uJ6ernZfKZsLBr

https://suno.com/s/qYpscJXPvXzbDYEH
Two versions of the anime intro

[coral ruin]

Maybe change your in server profile name cause yours could be seen as impersonating server owner

[elder willow]

mby sybau

[coral ruin]

Just go to the server settings and click edit pre-server profile

[half plume]

does anyone have the cpyleaks ai detector

[limber nexus]

Depends on your budget and what you want the PC for

[tight hornet]

I want to make my own ai Model any got Sites ?

[coral ruin]

Need help in ai images

[eager hollow]

Looks nice, any mobile generator? I usually generates on my work break 😂✌️

[eager hollow]

Aaa okay thanks ^^

[severe sage]

hello everyone!

[solar torrent]

You have been talking about this in other channels outside help channels.

[solar torrent]

https://media.tenor.com/-3scUOcTGBMAAAAe/dont-sybau.png

The guy in the meme is called Lazer Dim 700.

[solar torrent]

See my response in #✦│chat² message

[placid badge]

yo

yall

how do i download models from weights gg

nevermind

fiybd ut

[sharp plover]

i can't launch Applio

is the internet or server's fault?

[west wren]

Go to the local url

Oh nvm

[sharp plover]

when i try something and it says this:

[winged knoll]

Is there any way to make my ai sound better? Im running at the best settings according to the server and when i tried using the "E-Woman" Voice, It sounded really Jittery And Weird. But when i heard the Preview, It sounded ALOT Better.

[solar torrent]

For help about Applio, go to #✨│ai-help or make a thread in #1192011222023950368.

[solar torrent]

[winged knoll]

Wdm?

Wdym?

[nocturne oracle]

is there a recommended settings guide or something like that for vonovox?

[gentle trench]

If you use literally anything but e woman stuff I will help you out :3

[gentle trench]

What gpu do u have, as long as it's Nvidia and around 20-30 series and up it should work quite well

[icy depot]

Im trying to learn how to generate stuff locally
I have a pc with a 5090
The only thing I have gotten working locally is fooocus to generate images
anyone know where I can learn more about this stuff, like is there any good youtube channel with tutorials or something like that
for example what is all the torch stuff I have to install

I want to generate video locally but havent got it working

[nocturne oracle]

4060, rn i'm just running the stock settings

but if there's like "recommeded" or "optimized" settings then that'd be nice 🙂‍↕️

[gentle trench]

try these out ^^
crossfade duration: 0.15
extra time: 0.30 or 0.27
block size: 0.60–0.70 (try messing with it)

[nocturne oracle]

uhhh idk if i'm doing something wrong but extra time can only go as down as 1.5

[gentle trench]

uh

idk I got these settings from nick

[quartz vigil]

i got a 2 year old version of w-odaka working

and it works better for me than the newest version that was updated in octoper

😭

[gentle trench]

don't use the old crusty outdated stuff :(

what gpu do u have

[quartz vigil]

works better for me tho

and it sounds good

so who cares

honestly 😭

[gentle trench]

welp, never ask me for help then if u ever need it

[amber brook]

i neeed kizo voice , kizo is polish rap master

[unborn barn]

Hi

Fuck

[limber nexus]

what a opening

[polar flax]

#🔍│find-models or #1159289738314919936

[gentle trench]

https://media.discordapp.net/attachments/1178505150474747956/1248419713948713062/gifkitties_-_1751470205652332749.gif

[polar flax]

https://tenor.com/view/orange-cat-cat-side-eye-cat-side-eye-orange-cat-staring-gif-4749659952322810298

[solar torrent]

That's more of your issue.

[solar torrent]

You have been helping these people in #🧬│ai-chat instead of #✨│ai-help lately.

[gentle trench]

For that particular person it would be more of a waste to move to that chat and then tell them what settings to use but I should be using AI help for helping and AI chat for chatting

[solar torrent]

If someone asks for a quick help, it might not always be a quick help, but it could end up being a much longer deep issue. It's why I usually tell anyone the premise of W-Okada/RVC at first, and then move to #✨│ai-help or #1192011222023950368 for detailed help.

[cobalt plaza]

Isn't development of the official W-Okada software still active? I see newer releases like v2.2.2. Have these been tested against existing forks?

@solar torrent

[solar torrent]

I've never seen anyone testing the mainline (original) W-Okada (both v.1.5.3.18a and true latest v.2.x.x beta) against other W-Okada forks (from Deiteris and Tg Develop) and even another voice changer named Vonovox in the same PC system. The what I've observed is that many people say Vonovox, an alternative to W-Okada, can give better audio quality but also harder to use. While Tg Develop's W-Okada (b2377) has more recent features than both mainline and Deiteris (b2332).

[cobalt plaza]

Do you know what characterizes the shift from v.1.x.x.xx to v.2.x.x in the original mainline W-Okada series?

[solar torrent]

No idea. I haven't tried these versions myself.

[cobalt plaza]

All right thanks anyway

[solar torrent]

You're welcome.

[sudden crater]

Ai chachacha

[quiet shale]

whats the best ai voice changer?

still mmcserveriso?

[solar torrent]

MMVCServerSIO refers W-Okada, which itself is the same program. However, there are different W-Okada versions made by different authors.

[quiet shale]

so mmvcserversioi still the best ai voice changer?

but when i use it sometimes my voice is like gone or something

its like stuttering sometimes

[solar torrent]

For more help about W-Okada, better go to #✨│ai-help or #1192011222023950368.

[elder depot]

Idk maybe the wrong discord server to ask?
but i use ChatGPT to make png images of a 3D avatar made in Blender, and ChatGPT creates good ones, but it's very limited when using free version

so i am wondering if there are other good AI sites to go for? that's free maybe?

[charred tangle]

Hello

Is this the place I ask for help with ai

[exotic musk]

Let me ask something

[polar flax]

the latest og wokada is still worse than the following recommendations, if you think "if yours ain't broken don't fix it" and don't want to try the said recommendations, there's no need to comment such and ask for help

-rt

[pine acorn]

💻 Local Realtime RVC

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

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Deiteris Fork, with extra features, but supported only for Nvidia GPUs on Windows. and without cloud options GUIDE

• Wokada Tg-Develop Fork

A personal fork (modified version) of Wokada Deiteris Fork, it just adds some Quality of Life improvements to it like supporting Spin Embedder and Audio Effects. Don't expect too much about it since the creator made it originally as a personal project. GUIDE

• Wokada Deiteris Fork

Most suggested WebUI with the best general support for many platforms. GUIDE

⚔️ Wokada Deiteris Fork vs Vonovox

For Windows Nvidia, Both Wokada Deiteris fork and Vonovox have similar performance & quality. Users should read the pros and cons for both and choose based on their differences, such as UI and Vonovox's paid effects.
Read Wokada Deiteris Fork Pros&Cons & Vonovox Pros&Cons

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

Not suggested, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

The program is worse compared to the ones above, and much less updated. GUIDE

[elder willow]

Hi

[versed lotus]

Question:

I make short-form videos (10–20 seconds each) and I’m looking to start adding small AI elements or effects on top of real footage... not replacing it entirely.

For example, if there’s a shot of a man in the park with the sun behind him, I’d like to make the sun explode or have a cloud morph into a shape, using AI.

I was considering tools like WeavyAI ($19/month for about 7 minutes of AI video), but I’m wondering if it makes more sense to spend the time learning ComfyUI instead... since I could build these effects myself.

Has anyone here tried both? What’s the smarter path if I want to blend AI visuals into my existing videos efficiently?

[hasty vessel]

hello give me a job bro

[polar flax]

https://cdn.discordapp.com/attachments/427903255096459294/1425800514494402611/Untitled210_20251007102009.png?ex=6915bab4&is=69146934&hm=6d67ae68f8aec79446cbd8238fa38a35c804682fdb70e58ef41ea6d733220e08&

[gentle trench]

Yes but you're in the wrong chat, go here for help ^^
https://discord.com/channels/1159260121998827560/1159290139609137264

[kindred escarp]

Hello, there is a guide about how to train the model and make rightly the file .pth? cause i have a pth from tortoise but it doesn't make upload it to rvc

[chilly lake]

imagine asking "I have a recipe to make fertilizer, I want to use to to make cookies"

in your case ".pth" is a recipe book

what inside it is a different story

[limber nexus]

this channel doesn't get used for conversations much anymore

it did like 2 months ago

uhh, i know eddy was just online, im not sure about others tho

true, but this server mostly revolves around RVC rather than other kinds of AI mostly

RVC stands for Retrieval Based Voice Conversion

most say "Realtime voice changer" so that's better than most

[gentle trench]

yea the ai voice changer wokada is always called rvc by people who dunno what rvc stands for

[limber nexus]

Worm is here, nicee

[gentle trench]

heloo

I am a helper that is just a regulr member bc I get bored

[limber nexus]

yeah, and im a helper but, mabye not a great one, i try tho

[gentle trench]

u lookin in my bio? 👀

but uh anyway nah I'm not a dev or mod or anything

[limber nexus]

https://tenor.com/view/funny-memes-funny-jokes-anime-anime-girl-discord-gif-5371623980962101408

:)

"📥"

duh

📥

and you had a question about APIs correct?

Ah okay, i use groq, depending on what model you use you can get many many requests for free (i use multiple accounts with a API key from each to get more)

im crazy but not that crazy, i just want my AI to be aware and mimic emotions

have fun with that

never happened to me, i just get chills when the music is really good, but i get your point

that's nice

yeah i am, why wouldn't i be?

oh, im sorry about that, didn't mean for that to happen, but yes im fine dw

Kaggle is a remote PC service that you make scripts and run them in, you get free GPU time every week

yes, i use it for generating images and upscaling them, since i can't do it locally

[gentle trench]

not allowed here

nah bro I'm too slow for that, I know how to make ai models tho that's it

that's it for ai

I can draw and kinda know sound/music design

my music is always kinda just

messy

[gentle trench]

@groq is this real?

[limber nexus]

yes

[gentle trench]

my most recent "piece" of music

made it while I was sleepy

[limber nexus]

gives me Doom/Halo vibes, i like it

[gentle trench]

I use fl, how exactly would I extract the midi

uhhhh

best I could do for now

nah I'm teaching myself

[limber nexus]

you know how to read sheet music? i only know how to read Guitar tabs

more than me

hah! not me, im self taught

[gentle trench]

where'd he go

[kindred escarp]

i solved it. Obviously without even read anything here cause nothing was useful for me...And what you answered to me is exactly what i didn't want to know and it's not important to me nor useful.

[limber nexus]

he got banned

[gentle trench]

what he do

[limber nexus]

some stuff he said, either the groups he was trying to say to rob me or something he said about a burrito

[gentle trench]

weird

he seemed chill

[limber nexus]

yeah

[uneven geyser]

Heya

[forest crypt]

https://tenor.com/view/cat-listening-to-music-jamming-bopping-gif-16564158

[mortal hull]

Can someone make my avatar more beautiful or better?
My avatar logo name is 3zb

[gentle trench]

https://cdn.discordapp.com/attachments/1365750911070371972/1428134698495901797/Soapspeechbubble.gif

[gentle glen]

I am wondering, are there AI models and software for generating short sound effects from a text prompt? e.g a doorbell, gunshot, glass shattering or whatever?

[gentle trench]

eleven labs can do that for free yea

[gentle glen]

ah yeah, but I mean for download to run it locally a bit like the voice models, ollama, stable diffusion etc

[pearl light]

try AudioCraft its open source

[chilly lake]

stable audio diffusion has that

[final oak]

Heyoo quick question

Is RTX 2050 good for RVC?

[chilly lake]

for file inference maybe

for realtime it i as good as one cylinder moped

[final oak]

Minimum VRAM for RVC is 6GB right?

[pearl light]

Yep 6 GB VRAM at least…

[unreal carbon]

Hello, good night. I am having some trouble trying to download a voice model but it states when opening a new tab "invalid username or password". Even if i try to log in to get the model to download there is no link just text in left hand corner of the screen. Any suggestions? Thank you

[solar torrent]

4 GB is still possible for simple basic inferencing but you can expect some slow speed and low batch size, 8 GB is more preferred, and the more is generally better.

Commission or paid request isn't allowed in this server, even if you think of selling your models for a certain price. Instead, you'd expect to provide your voice models in #1175430844685484042 or something for free here, or else find a better place for that.

[polar flax]

only for free in #1175430844685484042

[gentle trench]

I saw that

[polar flax]

I'd assume it's promotion so better be deleted

[nocturne copper]

you bitches need help

ai??? seriously???

Just say i got no life no skills and want people to like u as much as people who spend 10 hours drawing

[finite cave]

https://i.redd.it/qoih7zhsyyoa1.jpg

[nocturne copper]

ho using reddit

😭

cant even make up ur own argument

[finite cave]

no argument. You're in the wrong chat. I rarely see here too much about visual art.
But thanks for sharing your frustration.

[grave garnet]

guys

what does epoch means

[half garden]

Hello everyone

[wanton solstice]

is w okada still the best? its been a while

i just got the vc client from the huggingface it was updated pretty recently so but im wondering if theres something better out

ive also heard of vonovox and i have a 3060, would it be better?

[latent rampart]

I wonder how this fellow managed to pull it off back in 2019
https://youtube.com/watch?v=znn9rI2KR7s

Shirley's vocals were cleaned up to a serviceable quality at least as if you're listening to her in the studio

[gentle trench]

I wonder if this could be done to other older songs

[gentle trench]

Vonovox would be better for your gpu yes, if u want help with it tho go to https://discord.com/channels/1159260121998827560/1159290139609137264 ^^

[latent rampart]

exactly, though the problem with that would be it would lack the old-timey charm to it. That being said, how can I do this to a song of the era? Would love to experiment with it some time.

[gentle trench]

I'm personally not sure but you could definitely check on videos like that and see if anyone else knows

[latent rampart]

There, just left a comment on this video as the one I linked to above is "for kids"-restricted:
https://www.youtube.com/watch?v=7ryYji3v6Og

[versed lotus]

looking to make stuff like this guy, but without spending 1k on his course lol
https://www.youtube.com/shorts/L9ZmlnIBqSU
this kinda vibe

[stark jungle]

@broken venture From what you've told me, I believe the objects you are trying to train for detection does not necessarily require color to identify. If that is the case then the colors themselves could be a distraction for the training.

But I don't really know the details of your training so who really knows.

[polar flax]

object detection is usually enough to be monochromatic, otherwise color blind ppl may be unable to play most games

[solar torrent]

I do draw art, and I like making art more than telling anyone to do the same as you.

These are my artwork projects, by the way.

[polar flax]

bro ragebaiting & left

[solar torrent]

He thought he'd get a backlash.

[gentle trench]

https://tenor.com/view/batman-lonely-rain-sad-gif-7792030237214898863

[solar torrent]

So that's the massive possible numbers of unique Touhou characters couples, ranging from 2 to 180 characters per couple.

Some Touhou-focused artists on X/Twitter like to pair two or more Gensokyo characters in their artworks, but I'm not sure how one would achieve such a large number of characters in their art at their lifetime.

[pine flame]

Epoche a training save basically, 1 epoche is your batch + repeat per images

[elder willow]

Any AI professionals here?

[south vigil]

Does anybody have sora 2 invite code ?

[solar torrent]

More like, where have you been?

[gentle trench]

Get a VPN for the USA, code isn't needed here anymore

[small escarp]

hello

does any of you have a model F t M that can handle screaming?

because its starts getting nervous

when I scream 😭

[lusty thistle]

Bingo sings Let me try again! (Kisuishou Densetsu Astal)

[bleak salmon]

what program can I use to use the models

[solar torrent]

W-Okada or RVC.

[bleak salmon]

which file in the okada do I run to open it

[solar torrent]

[mellow cargo]

how to get okada on macbook

[mellow cargo]

does okada work lon mac

[open wing]

to run applio v3.5 do you need discord installed as a app?

[forest crypt]

You need Discord if you want a Rich Presence showing that you're using Applio; if not, you can still use Applio anyway

[open wing]

well, I tried running the bat and it keeps giving me an error that it can't connect to discord

[forest crypt]

Ignore that error; just wait for it to show you the localhost link and then you’ll be able to enter Applio.

[open wing]

thanks. it worked

[forest crypt]

https://cdn.discordapp.com/attachments/1291141058650046547/1369506686355898419/g2kioe6y9ba.gif

[covert lake]

epochs are a unit of measuring the training cycles of the AI model

basically the amount of times the model went over its dataset and learned from it

they don't mean how good is the model, it's just an info provided on how they trained the model by the model maker
More ≠ better
Less ≠ better

There's no way to determinate how good the RVC model is until you try it out or listen to the audio samples if there are

[covert lake]

please use #1192011222023950368 for help

[covert lake]

please use #1192011222023950368

[covert lake]

no RVC model can do perfectly realistic non speech sounds, like screaming

[small escarp]

not even if i train it to scream on my own screaming voice?

or is that just not how it works

[gentle trench]

I mean I've made models that can yell and stuff but it's kinda iffy

[covert lake]

you can't train a model from #1175430844685484042 to adapt to your own voice/non-speech sounds perfectly

[covert lake]

i mean, it can do kind of laugh/scream sometimes, but it's not THAT realistic

[gentle trench]

Yeah, only models I've seen about to really scream was some old hulk models and an Omni man made by an old friend

And I think some others

[small escarp]

i mean its just gotta be passable xd

[covert lake]

it's meh

[small escarp]

cuz i scream a lot in VC

and it would be awkard if its just a heeeh he... he heeeeh

[gentle trench]

Should try out the omniman model made by vix

It's decent

[small escarp]

will try it thx

[charred light]

alguém brasileiro ai? preciso de ajuda

[solar torrent]

Deiteris and Tg Develop W-Okada forks are known to work on Apple Mac (Intel and Apple Silicon), but despite the premise they all would run on CPU only which would be slow for some parts. Deiteris W-Okada (b2332) has both Intel and Apple Silicon variants, while Tg Develop W-Okada (b2364) is the last known version to have a variant for Mac Silicon, with version b2377 has Mac removed. For more help about W-Okada, better ask in #✨│ai-help or #1192011222023950368.

[mellow cargo]

It’s too much coding

[solar torrent]

Yes, the installation on either Mac can be more complex than a Linux distro and Windows.

[coral kiln]

hey

uuhh

whats the difference between things in the "F0 Est.:"??

[whole tiger]

Anyone have a good Beyoncé model?

[open bough]

Mg

[elder nest]

ez

[solar torrent]

See #1175430844685484042.

[foggy jewel]

I create rap YT videos on omegle

I'm looking for an AI to edit the videos for me.

I've already edited a bunch of videos that the ai can copy the format for.

Question is:
What AI platform should I use. I don't mind paying

[glass onyx]

anyone can help me to set up?

[solar torrent]

[glass onyx]

CAN U HELP ME WHERE I CAN DOWNLOAD AI VOICE?

[solar torrent]

Calm down with your caps lock. For help about W-Okada or RVC, better go to #✨│ai-help or #1192011222023950368 instead of here.

[glass onyx]

oh im sorry mam

[polar flax]

don't post your message in multiple channels (I've deleted it)

[glass onyx]

okay sorry

[solar torrent]

[polar flax]

and go to the channel as namari said

[glass onyx]

i dont know where i can download the ai change voice updated

[wicked whale]

Does anyone know what the best discords are for AI?

[solar torrent]

Discord. If this is not Discord, then what is it called?

[wicked whale]

best servers

I never said this is not discord

[west mason]

Eleuther ai

[solar torrent]

The server you're currently in is named AI Hub by Weights. 👍

[gentle trench]

You can't even use Omegle wdym

[void pulsar]

😭

[opaque rock]

Yo

[elder willow]

Hi

How do I install Okada?

[gentle trench]

https://discord.com/channels/1159260121998827560/1159290139609137264

[frank belfry]

Hi, I'm thinking of subscribing to an AI platform (gemini, chatgpt, grok, perplexity, etc). Any recommendations for an engineering student?

[covert lake]

elaborate in #1192011222023950368 or #✨│ai-help

[forest crypt]

Hey @glad nebula , I wanted to know if you could tell me how the dataset you used to train the Legacy Core 2.5 pretrain is composed. Specifically, roughly how many minutes or hours you used per speaker. I could ask you in #🔊│ai-development , but I still can't see the channel.

[glad nebula]

ljspeech + m4singer, no denoising, removed some of the bass singers, and removed some clips of the tenor guys because there were some instruments in the background

batch 64, trained for over 100k (cant remember the exact the number)

but the pretrain is flawed in various ways

starting from the batch, too high

• it's a finetune of the OG pretrain, which im 99% certain it was trained using another discriminator, because things gets weird if you train a lot of epochs with that pretrain

[forest crypt]

Well, thanks for the information.

[glad nebula]

also, i trained an ai upscaled version of ljspeech, the original dataset is 22050, i upscaled it to 32k

but i dont recommend training these datasets together

singing in the dataset causes different issues

for example, held vowels will always have vibrato, even if the original file doesn't have it

the model is always gonna add vibrato

finetunes inherit that too

in worse scenarios it can fuck up pronunciation

[forest crypt]

I guess adding singing to the last two speakers won’t affect it, right?”

[glad nebula]

it'll affect it, the model actually needs only a few samples to learn singing (which sadly impacts speech negatively)

every speaker knowledge is shared

so if one speaker has a bias, everyone else will have it

i would also not train og's pretrain dataset, most of the problems found in rvc models are because of vctk

and the extra periods of mpd*

[forest crypt]

then I will choose to train with only speech datasets

[glad nebula]

[broken narwhal]

what software is used for vc nowadays?
i have the model

[forest crypt]

[glad nebula]

oh just in case, the language of the pretrain doesn't matter

if you need better pronunciation in a specific language, you'd need an embedder trained with that language

cvec only knows english

[gentle trench]

I'm assuming you're asking about the voice changer right? Vonovox and wokada are the ones that are used, there's 3 current up to date good voice changers

If u wanna know more go to https://discord.com/channels/1159260121998827560/1159290139609137264

[polar flax]

have you guys been thinking on making a pretrain using v1 or "v3" discriminator? (I mean like this #🔊│ai-development message )

[gentle trench]

I love no access channel it's my favorite ❤️

[glad nebula]

yea im training my pretrain at the moment

[forest crypt]

I guess I'll have to delete my 120 GB of downloaded datasets then.

[glad nebula]

nono it works too

afaik it tries to extract the closest phonemes to english or something like that

spanish works fine with cvec imo

but i like spinv2 more

[polar flax]

from someone else's experience, it may struggle on pronouncing strong "R" (trill)

[glad nebula]

oooo this is true

[forest crypt]

Technically, so far I only have 5 speakers with a certain amount of minutes, something like 40–50 minutes. I'm downloading this last one I just mentioned to see if it can contribute anything or not. I just hope I don't end up with only 5 minutes per speaker.

[glad nebula]

thats a good amount, og pretrain had like 10 mins per speaker.... yea lol

[forest crypt]

https://media.discordapp.net/attachments/635121092683628565/1065941195420880896/output-onlinegiftools_1-1.gif

[glad nebula]

i believe og pretrain was trained with mrd

[glad nebula]

but i cant prove it

[forest crypt]

We'll never know how the original was trained, right?

[glad nebula]

never but its not needed, vctk is a bad dataset, they all sound like the same person and repeat the same sentences

[forest crypt]

So that means I can combine different datasets without taking into account whether they say the same thing?

[glad nebula]

ideally you want your speakers to be expressive and dont repeat the same words

and train 2 millions steps or more if needed

in tts training they train 1m steps because is easier for the ai to train tts
but in rvc the ai has to learn features and more shit

[forest crypt]

And where did you train the pretrained model, locally or in the cloud?

[glad nebula]

im training mine in the cloud
both options will get very expensive keep in mind that

even with mrd, you need 2m steps

[forest crypt]

Well, you gave me valuable information. I hope it helps me in five years when I'm finally able to train a pretrained model.

https://tenor.com/view/cat-slep-cat-slep-smiche-i-go-slep-gif-26409040

[glad nebula]

its hard and expensive yea

[forest crypt]

https://imgur.com/yvJxyvS

[glad nebula]

never use the default rvc stuff tho, mpd alone is very slow at removing mirroring and the extra periods rvc-boss added are just bad

im training mine with the og rvc stuff bc i was curious if it fixes the weird og pretrain exclusive issues (it does actually fix them lol)

[astral ravine]

Hi everyone, I want to ask if anyone has a Japanese RVC voice model of Mikasa Ackerman?

[polar flax]

• https://discord.com/channels/1159260121998827560/1175430844685484042
• https://discord.com/channels/1159260121998827560/1163592055830880266
• https://discord.com/channels/1159260121998827560/1159289738314919936

otherwise go make the model by yourself

[wispy violet]

Hi i dont know what AI to use for my rvc models to sing my songs

i cant find the right info about it

[covert lake]

https://docs.aihub.gg/ and #1192011222023950368

[forest crypt]

I just downloaded a 47 GB zip only to realize that each speaker has barely 1 to 3 minutes of audio.

https://media.discordapp.net/attachments/1193767463016472597/1193767889719808030/caption-2-2.gif

[glad nebula]

damn
after you get your dataset keep in mind training pretrain can take weeks, mine is still learning breaths and how to remove mirroring at 340 epochs (2.1m steps)

[glad nebula]

og pretrain was trained for 636 epochs

(the v2 32k one)

it takes exactly 100 epochs to notice an improvement

[forest crypt]

what sample rate do you recommend I start with? because I'm thinking of starting with 40k

[glad nebula]

[glad nebula]

24k

[forest crypt]

https://tenor.com/view/timmy-no-candy-gif-5909516713687615776

[glad nebula]

24k is easier and faster

32k is still good (thats what im training)
40k and 48k is when things start to get slow and bad

and also it's harder to train, it needs almost a perfect dataset to properly learn 40k and 48k

start with 32k if you believe 24k is too low quality

24k quality is perfectly fine for realtime, for singing you can just upscale the outputs later with your favorite ai upscaler i guess

[lofty gyro]

Yo, guys.
Are there any SPIN embedder models?
I didn't succeed in finding any
Or maybe my search in discord doesn't work well :S

[drifting trellis]

when i upload a file on suno ai it says "Uploaded audio matches existing work of art"

how can i bypass this?

[round cosmos]

Where did the e-girl models go?

[gentle trench]

They went away because of disgusting catfishers

[round cosmos]

lol

[finite cave]

Worm is training them in his basement

[gentle trench]

Never

I will never give those barely human things what they want

Catfishers aren't people 🗣️

Never end up like this it's not worth it

[gentle trench]

What the fuck (they deleted it I think)

<@&1159293140440723499>

[last prism]

i have a question

[last prism]

preach

[gentle trench]

Wassup

[last prism]

can i use these voice changers on a android?

[gentle trench]

Nah currently none of the voice changers are even close to possible to running on anything other than a computer

[last prism]

oh.. mkay okay okay

[gentle trench]

Ye sorry

Like they struggle to run sometimes on a 1660 GPU so a phone might just explode

[last prism]

its okay do u have any other recommendations for a voice changer (real time) on a android?

[wispy mortar]

yo 'm new to the rvc stuff, can somebody please tl;dr me what epochs etc is e.g some models having 90 epochs or smth

[last prism]

damn

[gentle trench]

Epochs is just how many times a model went through training before it sounded good

More or less doesn't equal better tho

It's always randon

[wispy mortar]

How can I make my models sound better?

Any reccomended settings?

[gentle trench]

Tbh I don't know if there are any, your best bet would be at least going for a laptop if u don't have money for a whole pc

[last prism]

my cousin destroyed my pc soo

[gentle trench]

Yikes

[last prism]

he beat it up like the pc cheated with his girl

[last prism]

ik

[gentle trench]

Do you mean when training a voice or are you wanting to improve an already trained model? You can't improve an already trained model btw :3

[wispy mortar]

Well to make it sound better when I use it but not being able to improve it sucks

[gentle trench]

Are you using one of the real-time voice changers?

If you need help making it sound better I could help you in https://discord.com/channels/1159260121998827560/1159290139609137264

[ebon oracle]

Hi chat, how do i decide which RVC base model to use for training?

also what is the differnce between G and D models

[gentle trench]

U need to use both G and D when using a pretrain

[gentle trench]

If you're new to this I'd say just use the original pretrain which is used by default

But if you want to use a pretrain use either legacy core or klm

[ebon oracle]

is the advantage small?

or is it just complex

[gentle trench]

The only difference is you have to put the two files into the training software which isn't difficult you just drag and drop

Legacy core is way better than og, idk about klm tho I never used it bc it's a spin pretrain

[ebon oracle]

yeah this program is not cooeparitng with me but ill try to get legacy core to wrok

Mel spectogram simmularity is essentialy loss..?

[pine acorn]

🤗 Huggingface Spaces

• **RVC⚡ZERO**

Huggingface Space by r3gm

• **Applio**

Huggingface Space by IA Hispano

• FaceFusion

HuggingFace Space by Nick088

[limber lotus]

anyone know how to download Vonovox

[rancid shuttle]

Have you got scammed before?

[gentle trench]

There's probably about a thousand scammers in here just waiting to type some dumb shit like this

They're all losers

[stark jungle]

why does it feel like ive been to unlucid before

[polar flax]

tell him to pay you for the damage or get a job for such

[polar flax]

nah that person is kinda sus

[tawny oracle]

can I have help getting an open source coding agent working on roblox studio with mcp support

ty

stuff like terminals stuff like that

I don't have money for claude

[forest crypt]

In the end that dataset didn’t work for me and I had to download another one, which also isn’t enough to reach the 109 speakers.

[glad nebula]

you don't need 109 speakers to train a pretrain, mine has only 18

unless you wanna finetune og pretrain, then optionally you can finetune all of the 109 speakers... but honestly i think it's better to train from scratch, og pretrain is weird

[forest crypt]

I know the limit isn’t 109; I had planned to have up to 120 speakers, but from what I’m seeing online, there aren’t many datasets available.

[chilly lake]

EARS, M4, etc

[polar flax]

the speaker (timbre) variation to cover on matters (not only typical boring adult male & female voices), iirc that's what KLM datasets have been achieving

[gentle trench]

Yo could any of the mods unban @steelshot4401 he has gained his account back from being hacked

[polar flax]

tell him to file an appeal

[gentle trench]

👍

[forest crypt]

There aren’t many Spanish datasets, and when they do exist, each speaker usually has only 2 to 3 minutes of audio, which probably isn’t very useful

[chilly lake]

ah, yeah

for english tts models they use stuff like public transcripts and speeches from Europarlament and stuff like that

[glad nebula]

remember that the pretrain language doesn't matter and a spanish dataset wont improve spanish pronunciation in rvc

[still sigil]

@surreal acorn no advertising

[short juniper]

is vonovox premium worth it?

[gentle trench]

Nah just use fl studio with Vonovox to get around paying, unless you want to support the creator of course

[broken venture]

Is it better to use grayscale image or RGB image for training model (cv, yolo) in operational implementation?

[polar flax]

only for some cosmetic filter effects

[lofty spade]

Yo guys

what female-male ai voices do you recommend

some good ones to the ear for roleplay

[last prism]

yr smart

[hot kelp]

d

[solar torrent]

See voice models in #1175430844685484042, but it's hard to recommend one.

[cursive carbon]

Кто знает как это фиксить?
Hello everyone! Can you tell me what this error is and how to fix it?

[karmic fiber]

Hey all, quick and extremely nooby question hehe. I'm trying to make an use-case diagram for a game project where a human will compete against the AI-rival. Will AI-rival be called separate "Actor" or just the part of system?

[covert lake]

@lilac ridge We don't allow promoting here

[covert lake]

This server is english only, be sure to not use youtube/video tutorials for realtime voice changers, and to elaborate in #1192011222023950368

[lime raven]

can someone give me a good dataset maker that i can download?

ive been using instant dataset marker for a long time

[covert lake]

@true osprey We don't allow promoting, most of your messages are just promos, you have been warned and all your promos have been deleted.

[gentle trench]

Bad ai hub user bad

[ebon oracle]

are there any voicehcnager (rt-vc) cleints that support refinegan?

[ebon oracle]

also many settings on the trainign uis say pick this if your dataset is small or pick this if your dataset is large

i have aobut 1 hour 15 minutes of clean audio, is that large or small?

[gentle trench]

Use batch 12, it's large

[sinful obsidian]

For those working with LLM or AI agent systems, what parts of managing or debugging them feel most painful or time-consuming in your day-to-day work?

[glad nebula]

1 hour is fine, it's not large
if you’re wondering what the ‘max amount’ you can train and still notice an improvement, i'd say it's about 2 hours worth of audios
rvc models don’t really benefit much past 2 hours of audios, nothing really changes past that

[glad nebula]

safest batch size for me are 8 and 16

[ebon oracle]

oh well i went 24

is it a big problem?

[glad nebula]

should still work fine, if you notice your model is behaving weird like randomly glitching for no reason, stop the training and reduce the batch

[glad nebula]

no

hmm i think applio realtime supports it (? but then, the new refinegan is still wip, author is training the pretrain

[covert lake]

@ebon oracle Please use help channels next time and wait for an helper, there was a big wall of text so I deleted it, and be aware that Codename’s RVC Fork 4 is experimental, it got removed from the ai hub docs because it’s not suggested especially to people who aren't developers, as pointed out before in previous cases by other people like Lyery & Noobies (An Applio Developer): #🔥│model-maker-chat message #🔥│model-maker-chat message

[glad nebula]

im not his friend but whatever and yes i'd recommend applio instead if you're learning how to train models

[covert lake]

Vonovox patreon gives voice effects, which are easier and internal (less delay) compared to the fl studio method suggested by the other user, ofcourse it's your choice but just know that it's not a need

[covert lake]

I thought you were but I just edited the message, but yeah I was just saying that because it happened before that users tried that fork and resulted in models explodings

[remote zealot]

ai

[gentle trench]

yes

good job

[remote zealot]

you're ai?

[gentle trench]

no

[remote zealot]

what are you?

[gentle trench]

a worm

[remote zealot]

I eat worms

you're my food

[gentle trench]

oh no..

[remote zealot]

[thorny quiver]

guys any recommanded AI servers? except AI HUB

👀

[finite cave]

StableDelusion

[thorny quiver]

can yo use send the invite link tho

[ebon oracle]

hi chat, does smoothed loss scale iwth datasset size?

like is it not normalzied for size and just some sort of

sum of errors

[tiny marten]

@ebon oracle Please use the sregister command to verify yourself and start playing

[ebon oracle]

honsetly not sure what this bot wants from me

[tiny marten]

Almost there! Click the button below to verify yourself on our website.
Join our Official Support Server if you face any issues!

[ebon oracle]

oh its

😭

[gentle trench]

this is not possible

if that thing can send a discord server so should we

mods kill it

[ebon oracle]

also, if discirmantor stops improving but generator is still improving, is that fine or still bad overfitting wise?

[glad nebula]

bad, the discriminator is what forces the generator to create realistic audio

[ebon oracle]

mmmm, gotta love having a too inteligent generator

i mean dicrmantor is improvjng, but very slowly

while genreator loss is falling off a cliff

[ebon oracle]

and getting worse at times

[glad nebula]

rvc case is unique because there's no thing such as overtraining... models get worse due to the og pretrain (likely) being trained using different stuff instead of what we have in the project

rvc uses a fused discriminator that was meant for tts, not speech to speech

it's named multi-period discriminator, and also in rvc it has extra periods that only causes models to sound robotic

[glad nebula]

the losses aren't very accurate here, instead just listen to the model

[ebon oracle]

oh ok

[covert lake]

Yeah I was just saying for the message above which was in russian, don't worry

[jade lodge]

can anyone suggest how to analyze a 2500 page book and split it into a summarization per chapter as well as overall summary?

[chilly lake]

that's exceedingly large for any LLM

best you can do a summary per chapter, then re-run summary on the summary

[jade lodge]

i think it might need a RAG setup where it chunks... but i tried that before and it didnt go well

[polar flax]

how about per section (or subsection) summary then per chapter

I think most textbooks are quite structured like such

[forest crypt]

Therefore, based on what we have discussed, the best approach to training a pretrained model is to use only speech datasets. Then, if I want a singing version, I can simply fine-tune that pretraining using singing datasets.

[glad nebula]

yup

[forest crypt]

At least with the EARS dataset I can reach the amount I wanted; now the only thing left is to find where to train it.

[glad nebula]

can u give me media perms?

[glad nebula]

i use vast.ai

[forest crypt]

I think you need to reach level 30 to get media permissions.

[glad nebula]

the original architecture of rvc (mpd + nsf hifigan) faces a lot of problems while training from scratch, like mirroring and models not being able to sing, keep in mind that, its not easy to train a pretrain

[glad nebula]

maybe noobies can assist you in that regard, pretrains are just very hard to train

[forest crypt]

Noobies told me a while ago that I could try refinegan, maybe I can give it a shot and see how it turns out

[polar flax]

ask @dawn temple to give back your role

!give-media-perms @glad nebula

[dawn temple]

done

[glad nebula]

huh i can't send images

[dawn temple]

now you can

[glad nebula]

[hasty wind]

how do i set up my voice changer

it keeps saying {SIO} rconnection failed Error: xhr poll error

[vapid nacelle]

[hasty wind]

[knotty mist]

who can help me set up the voice changer ill give a nitro boost or usd dm me

[glacial coral]

sm1 lend me some owo cash

I'm broke now

[kind flame]

hi

wsg guys

[solar torrent]

When you work a job in real life, you'll get cash.

[solar torrent]

I like to help people, but I don't need a reward like Nitro or cash. For help about W-Okada or Vonovox, it's best to go to #✨│ai-help or #1192011222023950368 instead of your direct message.

[pure eagle]

anyh1 got latina voice or smt

[forest crypt]

[noble gale]

someone have tips for a voice module that sounds good

i dont find any good one for my voice

[gentle trench]

No go away

[lavish ocean]

subscribe to my channel fr im making some ai slop plz plz
https://www.youtube.com/@StorytimeDann

im not a robot 😡

[forest crypt]

Modsss

[gentle trench]

I may not be a mod but I am here

I'd be a pretty good mod, I'm actively here because I have no life

[forest crypt]

[wooden fox]

hi

[forest crypt]

https://cdn.discordapp.com/attachments/1003825079328702529/1400359443027464262/image0.gif

[glad nebula]

you can apply for staff i guess

#🔥│apply-for-staff message

[gentle trench]

I doubt I meet the requirements but I guess there's no harm in trying

I actually remember why I am not a staff member

They called me dumb ❤️

[glad nebula]

if they're talking about rvc dataset cleaning then you only really want to remove background music, reverb and train only train one speaker

denoising, de-ess etc just hurts the model

[gentle trench]

De-essing actually hurts the model?

Good to know

[glad nebula]

anything that removes something from the spectrum or limits the dynamic range is bad

de-esser are compressors

[gentle trench]

I always wondered what they actually did besides just making the S sounds smoother

[glad nebula]

yea its a type of compression that reduces the harshness of sibilants

[gentle trench]

So it's better to just use the raw audio after removing and bg audio and reverb/echo

No extra fx to make it sound "better" in any way

[glad nebula]

correct, just remove bg and reverb

[gentle trench]

Got it

[glad nebula]

it can handle a small amount of natural room reverb (not fx/fake), the model randomly adds it in inference outputs, it doesnt hurt learning or anything but can be annoying

[gentle trench]

So if there isn't a way to remove the small amount of room reverb it's not the end of the world for training that particular model?

[heavy rune]

hi

[glad nebula]

nop... 99% of the problems in models is due to the og pretrain

the other 1% is the dataset

[gentle trench]

Damn

[glad nebula]

its the combination of cvec and og pretrain

[gentle trench]

If only legacy core would continue getting public updates 😞

Would be the perfect pretrain to test of to a truly good pretrain to finally beat og fully

[glad nebula]

oh i just finished training legacy 3.0 yesterday at night

[gentle trench]

Nice

[glad nebula]

i tested it, also confirmed the problems of models is just the og

[gentle trench]

Do you have any samples?

[glad nebula]

og pretrain for some reason stops learning dataset features at some point of the training

[gentle trench]

That's odd

[glad nebula]

for example epoch 30 and epoch 40 of og will sound most likely the same, but when you use a proper pretrain, you can notice how epoch 40 learned something else, like the lisp of the voice, or the way the person talks

the robotic sound is bc the discriminator of the og pretrain starts fo fail

so everything breaks slowly

first no more features are learned, then disc dies

[gentle trench]

Makes sense tbh, that's why "overtraining" happens

[gentle trench]

I can tell the difference immediately and I don't even have headphones on lol

[glad nebula]

og sounds very loud and robotic because is starting to fail

mine has a little bit of ringing, but thats actually a rvc problem, cannot be solved

could be nsf hifigan idk

[gentle trench]

You could try training without hifigan to see if that changes anything

[glad nebula]

im interested in noobies refinegan, if he releases it i'll be trying it

[gentle trench]

I forgot about that, what's it supposed to do compared to the current hifigan

[glad nebula]

is not possible for models to overtrain, models sound robotic and bad with og after a couple of epochs because the discriminator fails

the model always keep learning something

i can finetune mine for multiple epochs and it wont sound robotic

[glad nebula]

since legacy 2.5 was trained using og, it has this same issue

3.0 was trained from scratch, no og involved

it learned everything by itself

[gentle trench]

I'd love to be able to check out lc 3.0 if possible but I get it if not :p

[glad nebula]

the gradients of 3.0

[gentle trench]

They're not spiky and all over the place 🙏

[glad nebula]

oh another problem 2.5 have is that the speakers are completely broken, they don't sound like the original voices at all in the model

that also comes from the og pretrain

[gentle trench]

So og pretrain completely destroyed the work you put into 2.5?

[glad nebula]

but in 3.0 they all sound like themselves

[glad nebula]

its broken yeah

just found all of that while training the new one

i think og is broken because the author used something different than what we have in rvc, and also that the vctk dataset is just bad

its not normal for the discriminator to fail

[gentle trench]

Maybe that's why when I trained Batman with 2.5 and 1.0 they sounded completely different

[glad nebula]

yea any model trained with og gets broken, regardless if its a pretrain (legacy core 2.5/1) or a regular model