🧬│ai-chat

[chilly lake]

there are no QC any more

[gentle trench]

https://tenor.com/view/garn-47-garn47-car-huh-confused-gif-14701624453881581373

[unique pawn]

what happened to em?

[spare yacht]

Anybody got sora ai codes

[gentle trench]

sora ai code should be a banned word

[chilly lake]

it is fking annoying

[gentle trench]

make it unable to be said like the k + the y + the s

[still sigil]

its allowed if its banter

[inland totem]

hello everyone

[solar torrent]

Here's your Sora, as you wished. https://static.wikitide.net/bluearchivewiki/9/99/Sora.png

[gentle trench]

that's a big forhead

[solar torrent]

That still counts as begging, by the way.

[polar flax]

https://tenor.com/view/sora-ときのそら-ホロライブ-tokino-sora-hololive-gif-5398913613780908156

[solar torrent]

https://i.redd.it/jk7dp6mfc42d1.jpeg

[limber nexus]

i could land a plane one that forhead

[rigid pine]

anyone have a sora 2 code?

[unique pawn]

https://tenor.com/view/star-wars-time-to-leave-clone-trooper-apologize-gif-7667729

[knotty cipher]

hey

[unique pawn]

I need my model maker back mannn, yall can't keep me waiting for over a day, i'm fiending to post rn 😭

[covert lake]

we got 2 still

and we are thinking of removing qcs soon

@dawn temple btw when are we going to change the qc system?

[shy haven]

my voice changer keeps cutting
like im talking then it stutters

[dawn temple]

Soon, I'll start working on the bot next week

[shy haven]

can you help me with my settings

[solar torrent]

For W-Okada, better ask in #✨│ai-help or #1192011222023950368.

[solar torrent]

I no longer have Weights Pro, but you can tell some staffs behind the website might be underpaid at some point, so they put some ads there to generate some revenue.

I've been known that this website had have a few hidden ads before, but not as many as this one and it was less obvious for most people to spot on frontend or client-side.

[gentle trench]

Why would you pay them a subscription?

[solar torrent]

The thing is, I never paid to them. The "free" Weights Premium or Pro was given to me by Bea.

[gentle trench]

Interesting

Didn't know u could get given a subscription

Is it like someone else paid for a month of a subscription or something like that, how discord u can gift nitro

But for weights

[solar torrent]

I helped something a bit on "Weights" website 11 months ago, although the time I got the actual prize was some time in the early of 2025.

[gentle trench]

That's neat

[short birch]

what you thinking about gemini-3.0-pro

[unique pawn]

I use it everyday

It's awesome ngl

[potent cradle]

I just got an 5090, what kind of cool things can I do with it? Besides gaming and pcvr

[gentle trench]

ai voice changer will run basically perfectly

[limber nexus]

you could do so many things on that thing

[sly notch]

помогите настроить

[potent cradle]

I know. But what are some

[limber nexus]

honestly ive never thought about the things you could do with a 5090, but you could, run a ChatGPT level model and have a local no limit chatgpt, you could generate 8K images in only a few seconds, honestly soo much

[hazy current]

@novel scarab

[novel scarab]

who r u ?

[hazy current]

zain nro

[polar flax]

chatgpt model will never go public to host locally, only through cloud API access, and you mean hundred bil param models which need a workstation with more than 96 GB vram

[limber nexus]

ahh, wait yeah youre right, thanks for correcting me

[polar flax]

there are many AI applications that benefit on up to 32 GB vram, though 5090 is technically moddable to 96 GB vram (pretty like RTX pro 6000)

note that only some skilled ppl like paulo gomez (search in youtube) and some chinese ppl can do vram modding & bios tweaking

[hazy current]

u can give it to me
That would be cool because i need to get rid of my 4080 super

[polar flax]

just wait for 5080 super 24 GB, while the 4090 price hasn't significantly dropped

[potent cradle]

Okay, anyone online who does that where i can pay them to mod it?

[potent cradle]

I upgraded from a 4080, overclocked to run like a 4080 ti super, if that was a thing oc

[hazy current]

Bro i upgraded from a 1650 to a 4060 to 4080 super

[polar flax]

even it is still risky for such professional modders

[potent cradle]

1050 mobile > 3070> 4080>5090

[hazy current]

Literally

[polar flax]

also there may be no modders available in your region, maybe only in china

[potent cradle]

Aw man

I saw a 4090 upgrade vram kit thing

You think it will makes its way to the 5090 soonish

[ivory hinge]

Heyo does anyone know if there any ai autotune that can runs live ?

[ivory knot]

buatkan saya gambar ikan

[gentle trench]

https://cdn.discordapp.com/attachments/1159290139609137264/1419738795833819269/Jaxcostume.gif

[potent cradle]

for a 5090 should i use W-Okada or Vonovox for best quaily

[gentle trench]

vonovox

[elder willow]

Hey everyone, Grant here 🖐️hope the week’s been good so far. Studied Law & Finance at LSE, but dropped the theory in 2020 to focus on algo trading… I study market structure and build automated risk systems mainly how ETFs distort volatility and how to design code that absorbs that chaos instead of getting wrecked by it. Excited to see what the builders in here are working on

[solar torrent]

Vonovox can give much better audio quality, as much some other people saying, but its UI is less friendly and more professional than b2335 W-Okada.

[potent cradle]

It looks like it auto adjusts the delay, or can j manually change that

[solar torrent]

The v.1.5.x.x and older versions of mainline W-Okada, and also Deiteris/Tg Develop W-Okada forks, aren't known to work with GeForce RTX 50 series.

[stark jungle]

I'm wondering if Nvidia NeRF has been used by anyone here?

[polar flax]

soon

[frank jasper]

is this really the best voice changer there is?

i have 5090 and 5070

i see stuff like veo 3 wan 2.5 and wonder how voice changers that focus on voices cant get that good

but a image model can? so i just wanna know if this is still the best or not

can run anything really

[still sigil]

because veo 3 and wan 2.5 are made by multibillion dollar companies and the voice changers are made by individuals/small teams

[frank jasper]

ahh ig ur right

does mak sense

[polar flax]

you haven't seen the best and the worst of voice models and image loras you could ever find

[green grove]

can anyone help me with the voice changer?

[frank jasper]

oh ive seen some pretty bad voice models but have i seen the best voice models. no

ive made a few

ive checked a bunch

[frank jasper]

i make loras on every new model that comes out that i think is worth it for me

i just hope they release wan 2.5

i pray

[polar flax]

I'm pretty sure it's not that hard to find bad SD 1.5 and flux loras

though SD 1.5 tends to be worse

[frank jasper]

its not i just go on cicvit but i also make stuff that suites my own needs its not that hard

i use flux only

bc i do realism mostly

sdxl etc just not consistent

so i dont use it much

[polar flax]

RVC seems to still have limitations for average to mediocre quality datasets

but have you tried TTS cloning with chatterbox?

[frank jasper]

i never heard of any of that

[frank jasper]

yea iu jut run clown fish with it for the most part

[polar flax]

#📰│dev-updates message

[frank jasper]

i mean from samples rvc sounds better ngl😭

only if ur datset good though so could be why

[gentle trench]

U can find the worst of the worst if you just on on weights.gg and search through the models directly made on weights

[strong dock]

is there a way to hear it back when you said something because it works but I cant hear it back myself

[gentle trench]

Get voicemod and turn on the heat yourself button

[strong dock]

omd I dont know why I didnt think of that thats rlly obv 😭 ty

[solar torrent]

I'd rather take that as sarcasm and not some practical solution.

[strong dock]

Im not being sarcastic its acc real 😭

[solar torrent]

[gentle trench]

Using voicemod is the easiest and less tech savvy way of being able to hear yourself with the voice changer

[loud wasp]

Hello, are there any voice changers that work on Windows 11 on ARM with Snapdragon CPUs?

[burnt blade]

Hello everyone,
I specialize in smart contract development (Solidity, Rust) and full-stack dApp architecture using technologies like Next.js, Node.js, and PostgreSQL. I’ve built and integrated onchain systems across Ethereum, Solana, and L2 networks including zkSync, Arbitrum, and Optimism.
My toolkit includes Viem, Ethers.js, Wagmi, TailwindCSS, OpenZeppelin, Foundry, Hardhat, and Anchor. I’m also familiar with IPFS, The Graph, LayerZero, and cross-chain messaging protocols.
I’m focused on building secure, scalable, and user-friendly Web3 applications.
If you have a new idea and need a developer, feel free to reach out—I’d be happy to collaborate.
Thanks!

[short prawn]

hi does anyone here have a girl slovak voice model?

[gentle trench]

read the rules'

[short prawn]

who said im going to catfish with it

or troll

[gentle trench]

it's way too common so just making sure

[short prawn]

alr

[simple mulch]

yo wont ai voice changer break my mic?

[gentle trench]

nope

as long as you're using a trusted one like Vonovox, or one of the two wokada forks

[simple mulch]

i'm using w-okada i think

i just want to do some voice chat trolls and i don't wanna break my mic

[gentle trench]

trolling of what kinda exactly

[simple mulch]

idk

so is w-okada safe?

[gentle trench]

it is yes

btw if u need help pls ask here ^^
https://discord.com/channels/1159260121998827560/1159290139609137264

[simple mulch]

ok thanks

[gentle trench]

np

[simple mulch]

idk why i sound robotic

on samples the voices are better :/

[gentle trench]

what voice are u using?

[simple mulch]

idk i tried a bunch of voices they all sound bad and my settings r good

[gentle trench]

it could be your mic maybe, is it picking your voice up good

[simple mulch]

i have a rode nt1

its good

i think

[spark furnace]

s

[gentle trench]

s

[gentle trench]

what's that

oh ur mic sorry

I forgor that was what we were talking about

[simple mulch]

yeah it's pretty good

so idk whats the problem

[gentle trench]

are you able to record a sample of what it sounds like when u use any model?

[simple mulch]

not rn sorry

its just the same voice but robotic

[gentle trench]

btw how close or far awar from your mic

[gentle trench]

that's ok

[simple mulch]

like normal talking

[gentle trench]

I meant like are you physically close to it or

[simple mulch]

pretty close

as always

[gentle trench]

maybe try getting your head a litttle farther back? see of that helps

[simple mulch]

its the same

[gentle trench]

hm

wait can't u just record with snipping tool for the voice changer to show how it sounds

there should be a feature in it to record

[simple mulch]

wait

[simple mulch]

where can i send?

[gentle trench]

u can send here pretty sure but you'd have to ask a mod if u can't send images ect

[simple mulch]

i can't send here

[gentle trench]

there's 6 regular mods online and one old man/j

[simple mulch]

can't i just send it on your dms?

[gentle trench]

sure

[alpine gust]

available red dead redemption ultimate edition account full access price 20$

[gentle trench]

https://tenor.com/view/oh-moderators-gif-25571015

[mental prawn]

wsp

[simple mulch]

me

i can take one if free

[twin spade]

[surreal wing]

i ate a live dolphin

[limber nexus]

this message was sent with a delay of 3 seconds

[surreal wing]

it was?

[limber nexus]

yeah, this one too

[surreal wing]

e

no it wasnt

it was only grey for like half a second

[limber nexus]

yeah it was

[surreal wing]

i can record it 😭

[limber nexus]

on your end

[surreal wing]

ur responding in like 3 seconds

unless u type and react at the speed of light

nuh uh

[limber nexus]

on your end its fast but it took a few seconds to be uploaded to discord

and i just type fast

[surreal wing]

this message was sent with a delay of 6 seconds

[limber nexus]

yeah, no it wasn't, yours was sent with a delay of 3 tho

[surreal wing]

this message was sent with a delay of 16 seconds

[limber nexus]

no it wasnt

[surreal wing]

yeh it was

[limber nexus]

so, you use plugins then?

how else would you know?

[surreal wing]

no, im just doing th same thing u did to me

trolling

lol

[limber nexus]

no, i can actually see the delay

[surreal wing]

thats y u wer responding in 3 seconds but also saying it takes 3 seconds right

also the message on ur screen doesnt go white until its public

u can visually see when it sends

[limber nexus]

no, the delay from me is typing, and there is a delay of when it gets sent to the cloud for everyone to see

but i can see the delay

[surreal wing]

i can too

its built in

[limber nexus]

discord does not tell you the delay of a message natively

[surreal wing]

when ur message goes from 🩶 to 🤍

:D

glad u learned sumthing new

[limber nexus]

thats when it gets sent to the cloud which then can take a second before its made public

but i can see the delay

cause of the way i have discord set-up

[bronze dragon]

does anyone sound like a real girl here and can test with me during a call?

[limber nexus]

dude, so many people are gonna be lining up with their AI voice changers

[gentle trench]

You're number 69420 asking for this stuff, read the rulesssss

[edgy helm]

Hey can u help me with something

[gentle trench]

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

[limber nexus]

depends

[stark jungle]

Voice models seem to be all the hype here🤔

[bronze tulip]

could anyone help

[stark jungle]

Doesn't seem as interesting as other uses of AI, why is it so popular👀

[atomic lichen]

some one got good e boy voice ai

[main nest]

Because other common uses are dominated by other services and sites, so a lot of people that come here come for the voice models and applications thereof.

[stark jungle]

That is interesting, when I first came here I didn't think it would be a hub for a niche

[drifting sigil]

I don't have a GPU for ai covers, only CPU

[solar torrent]

You cared about the delay? Well, it's because I use my hands to type on keyboard and then send one.

[gentle trench]

die

[solar torrent]

For helps about W-Okada, better ask in #✨│ai-help or #1192011222023950368

[toxic dirge]

YOOO CAN SOMEONE MAKE A GHOST FACE VOICE PLSS

[solar torrent]

To request a voice model, make a thread in #1159289738314919936.

[loud sorrel]

-colab

[pine acorn]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[golden spruce]

whats the best girl model

voice model

u h h

[sleek atlas]

how to change your voice with a neural network not in real time, but so that I can upload a file and change my voice in voice changer ai

[stable thorn]

Hi
Guys, is it safe to use a random sora2 code I've found on the internet?

[graceful flame]

Hi , You can try LunaBloom AI if you want to create social content or even professional vides for business purpose. Content can be created over 50+ languages and custom AI Avatars also available . Its a text to video generator. Heygen is also good for creating professional AI Avatar driven Videos.

[solar torrent]

This question sounds stupid. For voice models, find one in #1175430844685484042.

[still sigil]

why are you replying to messages from 4 months ago

[golden spruce]

recommend one

arent u a helper

do ur job

😐

[hybrid elk]

how to add voice models to w-okoda?

@solar torrent any help?

[solar torrent]

You asked like if I know every model of the entire library.

[hybrid elk]

nvm i figured it out

[queen stone]

-huggingface

[pine acorn]

🤗 Huggingface Spaces

• **RVC⚡ZERO**

Huggingface Space by r3gm

• **Applio**

Huggingface Space by IA Hispano

• FaceFusion

HuggingFace Space by Nick088

[gentle trench]

Rules, read em

[elder willow]

Guys, does anyone know why I try different voice models with RVC but most of them sound like, the mic cutting off for no reason? Im talking very clearly, mic close to mouth. Seems strange to me.

[gentle trench]

what voice changer are u using? did u get it from a yt tutorial?

[elder willow]

no, im using Real Time Voice Changer Client, the official one

[gentle trench]

there is no official one

[elder willow]

I do have a model that works much better than the others somehow, I wonder how

I mean the github Real Time Voice Changer Client one everyone shows

[gentle trench]

there are multiple

w-okada deiteris, w-okada tg fork, and Vonovox

[elder willow]

ooh

[gentle trench]

and the original which is outdated

[elder willow]

i cannot share image but, it says on top v.1.5.3.18a - onnxdirectML - cuda (Im using AMD CPU and GPU)

[gentle trench]

hmm

I believe that's the original but I could be wrong

[elder willow]

1 model does work good the others cut my microphone off, seems weird

[gentle trench]

we should probably move to https://discord.com/channels/1159260121998827560/1159290139609137264

[elder willow]

okey ^^

[polar flax]

the og version is garbage for AMD gpus

[covert lake]

that's like an over year old version of original wokada

[gentle trench]

dw I gave him deiteris fork, I had a friend who used the tg fork one and it's confusing and he said buggy

they have amd

[covert lake]

I had a friend who used the tg fork one and it's confusing and he said buggy
It's not buggy at all, it's only "confusing" because it's a newer personalizable interface, every bug that's in the wokada tg-develop fork is also in deiteris fork lol, like crashing when you edit the model's name via the UI

[gentle trench]

I never got to ask him about it before he deleted it but he said it was buggy, idk what bugs he encountered tho

[covert lake]

smh adapt to new UIs

he can ask me for help in #1192011222023950368 if he wants to try it

but i checked the commits and i assure you that deiteris has the same exact bugs

the only difference between wokada deiteris fork and tg-develop are:

• UI
• spin model support
• voice effects

[gentle trench]

he dissapeared from everything again

like batman

so he isn't here lol

[covert lake]

damn don't eat ur friend 😔

[gentle trench]

I didn't eat him

[empty hinge]

Hi everyone

Wow my name is colorful

[gentle trench]

it's because you're super awesome

[sterile crown]

Anyone got a stackwopo voice

[solar torrent]

Not quite found in #1175430844685484042. To get one, you either open a request in #1159289738314919936 or train one by yourself.

[dapper drum]

What is the best way to get the sound effects from the instrumental?

[solar torrent]

[solar torrent]

When you talk a bit in this server, your name will turn green/blue.

[atomic lichen]

some one has corpse ai voice

?

some one

[quaint fiber]

¿?

[sour mortar]

Hi everyone
I’ve been working on a small AI project related to agriculture.
I trained a DenseNet121 model on Google Colab using my own dataset,
but the accuracy is only around 43% so far.

I’m not an AI expert, but I’d love to learn how to improve the results
any suggestions or best practices for improving model performance when the dataset is small or not perfectly balanced?

Thanks in advance for any tips 🙏

[simple mulch]

pls give me girl voice model

i want to do something important

for schoolwork 😉

[gentle trench]

https://tenor.com/view/stomp-gif-17874349

[gentle trench]

read the rules

[simple mulch]

it's for school

i have to prove that ai is realistic

so please give

[gentle trench]

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

[simple mulch]

local worm

you man or woman

[still sigil]

[simple mulch]

so pls

can i get help

for school

[still sigil]

no

[simple mulch]

plss

[still sigil]

no

[simple mulch]

why

[gentle trench]

https://cdn.discordapp.com/attachments/1395804733004058775/1422643870025515118/Moral_blink_guy.gif

[stone roost]

bro gt 1030 is lagging the voices

[lavish goblet]

Welcome

[keen plaza]

hi everyone. i'm training mouse trajectory AI with a simple PyTorch LSTM model and a mousemove events dataset collected from 65000 real visitors for passing akamai & datadome bot checking. the loss is keeping ~0.000024 after 12 epochs. but the generated trajectory still looks very akaward. can anyone tell me how to improve my model?

[green grove]

could anyone help me??

[polar flax]

the way the mouse event data was collected seems questionable, do the "real visitors" actually consent with it? otherwise it sounds like privacy invasion to them

[keen plaza]

i think mousemove data itself is not related to privacy without webpage context.

besides, i only recording relative positions and on myself website.

[polar flax]

how about when to start and stop the mousemove recording?

[keen plaza]

what's more, many legal marketing services like Microsoft Clarity and Sentry records mousemove events and all DOM changes with absolute positions without prompt, and let webmaster replay them exactly for each sessions.

[polar flax]

do you mean realtive mouse position to the window position (yea exactly means focusing on the window context)?

[keen plaza]

relative mouse position to the starting point. i only need these to train mouse trajectory model.

[polar flax]

I think it doesn't matter, the relative mouse position to the window could be done without concerning "privacy invasion" as long there are no other data that may be sensitive

[keen plaza]

mousemove event won't be triggered if your cursor not inside the webpage.

[polar flax]

well I bet it could be why it doesn't seem to be accurate

again the mouse position to window could be more accurate (without being "invasive"), though you should consider some other factors like window size and the webpage's responsive UI

[keen plaza]

as i know, behavior analysis system of most antibot service only care about how realistic of mouse trajectory itself, not the click positions. so i think relative positions are enough.

[polar flax]

how about the event when it starts recording the trajectory

note that the users don't manually trigger it and the mouse pos when clicking a button could vary

[keen plaza]

nothing happened. just like the Microsoft Clarity. I put things like I will record and upload anonymized behavior data for security research in the privacy policy of my site.

[polar flax]

does it involve a little computer vision or purely mousemove data?

[keen plaza]

when a click event happen, the trajectory will be marked as end then start record the next trajectory.

[keen plaza]

no computer vision. purely mousemove data.

[polar flax]

what kind of the click event exactly?

[keen plaza]

actually it is the mouseup event. it will be triggered when you click anywhere inside the webpage.

[polar flax]

again where they clicked prettily varies, so that's why it is inaccurate

[keen plaza]

i don't think the antibot service will render all DOMs on their backend and know positions of all buttons. the intention of my mouse trajectory model is only prove "there is an actual human browsing your website" to the antibot service so my data crawler won't be blocked

[polar flax]

re-reading your topic, I dont think people could move the mouse and click while being blind, so that's why it is not feasible without requiring such data that may be invasive

[keen plaza]

i think no matter what is the intention of people clicking the mouse, it ends up the same mouseup event. there should be no statistical difference.

after reversing enginerring akamai bot manager code, it uploads all mouse and keyboard and window and sensors events to the server. but I think making a mouse trajectory model is enough.

[polar flax]

different position results in different mousemove, that's the point you're ignoring

[keen plaza]

i read many papers about mouse synthesizer and no one mentions screen position will cause statistical difference of mouse trajectory. but i can't confirm that.

and the akamai only collect pageX/pageY (positions relative to 0,0 of the webpage) as well, not screenX and screenY

[polar flax]

my point is the (pageX, pageY), sorry not have to be (posXtowindow, posYtowindow), never meant to be screen one

and I thought what you were talking is (dX, dY)

which could certainly vary for different actual positions

[solar torrent]

I'd expect NVIDIA GeForce GTX 1050 Ti as minimum possible, which has 4GB VRAM, for W-Okada and even RVC. This one is GeForce GT 1030, has 2GB of VRAM, so you can't do much about it.

[polar flax]

and still you need reading the webpage rendering size and how it positions the responsive UI elements

[polar flax]

using (dX, dY) is like navigating the trip path without a compass

[solar torrent]

You can go for any anime or fictional character voice model. It doesn't have always be the "girl" voice model as you'd expect.

[ancient pebble]

when I talk with the model it sounds fine, but when I record or talk with it in a call it sounds super fake and like choppy

[solar torrent]

For W-Okada the realtime voice changer, go to #✨│ai-help or #1192011222023950368 and explain about your issues there.

[ancient pebble]

okay

[solar torrent]

See #1175430844685484042.

[paper vector]

any of yall got a sora code

[solar torrent]

A character named Sora is in Blue Archive game.

[gentle trench]

here ya go

[limber nexus]

pretty sure anyone that did have one has used it already

[polar flax]

your point is basically like a macro but it tries to reproduce variation of the users' mouse movement for that purpose. it'd make sense to use (pageX, pageY) instead of (dX, dY).

it'd probably require software-level hook (or OS-level), though I'm not sure if websites could have detection as invasive as some sophisticated anticheat softwares.

https://tenor.com/view/robot-gif-24016987

[solar torrent]

For those who live outside North America and keep asking me for OpenAI Sora 2 invites, this is it, their blog stated that the Sora 2 is available only for Apple iOS in United States and Canada through invites.

[rich badger]

-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

[azure relic]

Has anyone found a way to get bypass certain sora ai filters consistently without having to phrase things like Yoda?

[elder willow]

yes

[simple mulch]

give me

[solar torrent]

#1175430844685484042

[simple mulch]

give me good

idk which one is good

[solar torrent]

You request me like if I know and hear every voice model from the entire library, what.

Think think, use your ears to hear each one of those. It doesn't have always be anyone to pick one for you.

[simple mulch]

:/

you are a helper

so pls help

[solar torrent]

I can help about settings in W-Okada, but picking one good voice model for you is too much, you know.

If you want "Touhou" voice model, search keyword "Touhou" in #1175430844685484042 or use Weights bot in #🔍│find-models

[simple mulch]

ok help pls

[solar torrent]

One more time, homie, if you keep asking me in #🧬│ai-chat instead of #✨│ai-help, or still annoying other people, you won't be genuinely get helped.

[simple mulch]

ok im sorry i will leave 🥺 😢

[amber dove]

Hello

There's a type of videos that I really wanna create but I don't know how!

can someone help?

[buoyant glacier]

anyone familiar with generative audio?

i have some questions but chatgpt/gemini cannot answer me

was wanting to discuss or ask questions to someone

[polar flax]

what kind of "generative audio"? please be specific, that's why it as well as me can't understand what you mean

[buoyant glacier]

i had some question regarding latency differences between Text-to-audio and text-to-speech. like architecture wise they both seem to employ similar architectures but TTS is fundamentally usually like 10x faster and I can't understand why. I get TTS is just lower dimension as a whole i guess but idk. sampling rates and stuff tend to be the same so i don't get the difference. I've been arguing with gemini for 2-3hrs but losing it.

[polar flax]

what is difference of "audio" and "speech" according to your definition?

[buoyant glacier]

so TTS/speech would be:

"I am walking my dog" and that is directly converted to sounds that pronounce that.

TTA/audio would be:

"90s rock music with heavy drums" and we create sound that roughly reflects this label. so this is more similar to image generation. however somehow its much slower atleast on the few SOTA models i tried.

and usually these TTA models cannot do speech, and the TTS model cannot do general audio

[polar flax]

so you meant "audio" as music, and it is prompt-based music generation which is basically Suno, etc.

that's completely different from TTS, you're just comparing a dog with a banana

[buoyant glacier]

yea music or music + soundeffects like "waterfall sounds". but why is there such a difference in latency? ive tried a TTS diffusion model and TTA diffusion model but why is it so different. and the actual models themselves seem fairly similar too.

so it feels weird to me

[polar flax]

what's the point of comparing the generated waterfall sound to the TTS voice saying "waterfall sounds"?

or you mean for classifying the former sound to the text spoken by the latter as the label?

[buoyant glacier]

I'm not comparing them they are for different purposes sure. But it feels weird one is much slower even though it's both just audio.

Because I'm trying to make a system that serves both speech and general audio but because of big latency differences there are load balancing/serving issues

[polar flax]

yeah it's so weird why dogs can walk but bananas can't?

[buoyant glacier]

Yea I get it's different but like how tf it's both just sound

Like image and video makes sense this is just both audio

[amber dove]

hello can I share links here?

Is there any good lipsyncing website or softwares that make these AI interviews?

There's a channel called Mind to Goals

[still sail]

Anyone give me the active invitation code sora 2. I will pay him 20 dollars as a thank you. With much respect.

[buoyant glacier]

Yo I have

[silent moat]

Somebody have a sora code?

[elder willow]

I do

[silent moat]

Would you share it to me pls

[elder willow]

no it's mine

[silent moat]

LOL

Bro i’m just trying to make anime fight videos

Like the tik toks

It’s so cool

[elder willow]

It is not

[silent moat]

Why

[torpid jewel]

anyone maybe has a sora 2 code pls i need it so bad

[elder willow]

I have it! BUT it's mine

[amber dove]

someone has experince with lipsyncing?

[torpid jewel]

😭😭😭😭

[torpid jewel]

Brooo

U cant like give it to more people or sum?

[peak geode]

Echte help me plz

[amber dove]

Why nobody is helping me here???

[limber nexus]

im so sorry, waht do you need help with?

[amber dove]

I'm really looking for someone who is good with lipsyncing videos

to help me out

https://www.youtube.com/channel/UCYVUm41HtniQ2c3nYFaxWEw

to make videos like this

[limber nexus]

ohh, ive never seen anything like that

seems just like it was generated with the audio tho

[amber dove]

i've got the ai voiceover and the actual footage of elon musk in this video but lip syncing it is the hardest part tbh

[limber nexus]

ok, so being honest i don't think i can really help you

im sorry

[gentle trench]

This isn't the help channel

[fickle trout]

Where in this server i can hire people?

[loud frigate]

Is there an AI like Suno AI that I can use?

[limber nexus]

we have no recruitment in this server

[fickle trout]

okay.

[limber nexus]

please ask in #✨│ai-help and i can assist you further

[shrewd sandal]

babamda @vague escarp bu adamın numarası var da aramıyoz

[polar flax]

no promotions, advertisement, and hiring are allowed

[atomic lichen]

we need eboy

[limber nexus]

we?

[polar flax]

https://tenor.com/view/the-office-michael-scott-no-god-please-gif-15249894192487756498

[limber nexus]

im probably gonna end up feeling like that at some point

[charred moth]

Hello, how are you? Any recommendations for AI soundtrack creators? One that's free and unlimited?

[cosmic cedar]

https://ilovesong.ai/ pretty good and it allows you to create 3 free songs to test

[gentle trench]

https://tenor.com/view/stomp-gif-17874349

[limber nexus]

oh the pain im gonna have to endure, mabye not everyday, but most days

[gentle trench]

we need a men kissing channel

[limber nexus]

please no

[ionic sun]

Yes

Talk?

[still sigil]

https://tenor.com/view/staff-furry-rp-is-real-real-gif-1987509878278887091

[brittle dome]

Anyone know any good voice changers I use voice mod but it sounds so bunzzz

[onyx belfry]

what a question

[dusk field]

ElevenLabs

https://tenor.com/view/trust-me-im-a-doctor-trust-me-i'm-a-doctor-trust-me-i'm-a-dr-im-a-dr-i'm-a-dr-gif-2525495326508012402

[sly nest]

Ai

[queen kernel]

Ai

[sterile frost]

does anyone know any good ai voice model creators

[wispy swallow]

do the ai girl voices actually work or do u have to like try to sound like a girl

[thorn meadow]

Hi all

[remote hemlock]

Hi

Hehe

[spark lagoon]

Hey everyone 👋

We’re building an On-Board Courier (OBC) project — basically time-critical logistics where documents or parts get flown across countries within hours.

We want to integrate AI-driven automation for the entire process using Make (Integromat) and other tools.

The idea is to automate:

-AI-based pricing and route optimization

-Automatic quote and invoice generation

-Factoring and cashflow management via API

-Flight booking and courier assignment

We’re looking for someone experienced with Make, automation, or AI workflows who can help us build and connect these processes.

If anyone here has experience with AI automation, APIs, or Make scenarios, please DM me — I’d love to discuss the project and see if we can collaborate.

Thanks a lot!

[spark lagoon]

thx

[pulsar pivot]

QQQQ

[thorn scarab]

I really want an AI for coding but mistral can't do this one

So wich Ai should I use?
Can anyone guide me please?

[calm vapor]

@timid musk

[surreal lava]

hi

[gentle trench]

What are you planning on doing 🤔

[compact river]

does anyone have premade n8n automation templates that i can get?

[frail obsidian]

i got some il hook u up with them

[gentle trench]

What even is a n8n automation template

[wispy swallow]

u dont wanna know

[gentle trench]

U ain't getting the info then 👅

Read the rules a second time

[gentle trench]

catffishers smh

[thorn scarab]

Is Samsung developing an Ai?

I've heard about it before, but I don't know if it's available or not

[paper axle]

helo! how can i fine tuning to lm studio models

[fallen spoke]

I have a question, I have been using this activation function:
If x > 0 then
Return 1
Else return 0

I think this is the derive of relu, but it seems to work fine as an activation function and maybe even better and faster so why not use it?

[chilly lake]

function like that only works in special cases

it is not learnable and can not be used for gradient back propagation

[fallen spoke]

But I used gradient back propagation with it and it kinda worked.
I tested some xor variants on it only tho (like 3 bit, 2 bit and different patterns), and it worked with that, I even tested giving it some string sequences and it kinda worked

[full ore]

Free premium training items are no more. I can no longer make models with them. I can only use credit cards now.

[still sigil]

credits ≠ credit card

[gentle trench]

Just train on applio goofball

Unless it's images or smth

[full ore]

I want to use it for future AI Covers for myself, and back in the day, I loved using them for multi-voice covers

I actually have a 679 cover featuring Dean Wendt Barney, Avenger Chuck E. Cheese, and Kermit the Frog

And can voice models be made on Replay?

[gentle trench]

Nope, but never train voice models using anything related to weights

It's ass compared to actually good training software like applio or mainline

[still sigil]

they can

[gentle trench]

Wait u can train on replay?

I thought it was just only a ai cover thing

[hollow field]

whats the best model for backing vocals/lead vocals

[chilly lake]

https://docs.aihub.gg/rvc/resources/dataset-isolation/#best-models-for-uvr

[hollow field]

so whats the best karaoke model

just says this
Mel roformer karaoke

[hollow field]

can someone give me a link to mel denoise v2

[covert lake]

yup

[polar flax]

the doc doesnt seem to mention which exactly

I'd recommend BS roformer frazer, and then mel aufr33 and becruily

[hollow field]

doc is outdated anyway

better vocal model

FV6

by gabox

[hollow field]

becruily karaoke sucks ass

[polar flax]

that is voc-inst model

[hollow field]

mine doesn’t work

[polar flax]

frazer is better if u dont believe

[hollow field]

oh ok

thanks

[glossy patrol]

Hey folks 👋 — has anyone here played with voice-first AI tools?
I’ve been testing an open-source one called Ito, and it’s wild how fast ideas turn into actions just by talking.
Do you think voice could actually replace typing someday?

[fallen spoke]

Yo, why do we multiply the learning rate by each weight in the network individually, why not just multiply it by the gradient of each output neuron

[solar torrent]

There's a model that says "frazer" in a customized MSST Colab notebook, but I don't remember the name of who made this notebook.

[indigo girder]

yo whats an good model im a boy and want to sound like a girl for catfish lol

[wispy swallow]

u have to speak english for any of them to actually work

and u cant laugh

[indigo girder]

ye i can speak english

not that good but a bit under avg i would say

[wispy swallow]

https://discord.com/channels/1159260121998827560/1403422896680079390

https://discord.com/channels/1159260121998827560/1404856994262810816

[wispy swallow]

this is if u wanna sound like a girl

[indigo girder]

any good settings? i kanda got an bad pc

[wispy swallow]

wait

[indigo girder]

alr

[wispy swallow]

theres a setting called chunk

256 is good

[solar torrent]

Yeah, you two can help each other, but I won't be going for that.

[indigo girder]

alr

[wispy swallow]

if u go above there will be more delay but it will sound better

[indigo girder]

sry

[indigo girder]

okay

[wispy swallow]

but 256 should be good

[indigo girder]

alr

can i test it on you rq in vc?

[wispy swallow]

like ure just gonna speak? i dont speak english

[indigo girder]

oh

alr

[wispy swallow]

just use ur phone

like make an alt and just go in a call with it

[indigo girder]

alr bet

ty twin

[wispy swallow]

np

[indigo girder]

yo can any1 help me with the settings? i just dont get it it sounds weird with my settings

[wispy swallow]

wdym sounds weird

set tune to 12-15

and "in" to like 0.2-0.5

[indigo girder]

can u come in call rq

its like laggy idk

yo can anyq come in vc with me and tell me if it sounds real?

[summer pecan]

What area of AI interest you most?

[indigo girder]

yo can any1 help me? i have to find the perfect voice, i got an deep voice and want to sound like an girl

[queen jungle]

did https://huggingface.co/becruily/bs-roformer-karaoke/tree/main works with msst ?
what repo did you used?

[minor blade]

There's no such thing as a perfect girl voice

[vast gyro]

Is anyone here knows the best way to generate pictures locally ? I just can't see any new guides on yt, only about year ago

[grave stirrup]

Help me..

I trained a model on an NVIDIA T4 GPU using a 40-minute audio file that I downloaded from YouTube as an MP3, then converted to WAV. The training was done with a batch size of 8 for 250 epochs, but the output contains noticeable robotic noise. How can I fix this? I’m sharing the result below.

[wispy swallow]

U gotta try to sound like a girl

[polar flax]

https://tenor.com/view/speech-bubble-gif-27609617

[indigo girder]

idk i cant find a good voice

[gentle trench]

Take woman pills/j

[indigo girder]

can we go in vc and can u tell me if the voice sounds good?

[gentle trench]

[indigo girder]

nuh uh

[slender jackal]

Hey i wanted to ask whats the best ai tool to create summaries for studying at uni that implements different fonts and a nice formatting and marks important stuff (and exports it to pdf or word)

[solar torrent]

I didn't make that notebook my own. I got a custom MSST notebook from #📰│dev-updates message

[minor blade]

Nuh uh

[echo olive]

WHO NEEDS A SORA CODE

i made a basic server of trust for any one to join

dm me if u want it 🙂

[upper jay]

Hlo

[timid musk]

this rule isn't a thing anymore

[gentle trench]

isn't it still illegal to help weirdos who want that stuff

[timid musk]

no

it's not our place to dig into the intentions of people who ask for help

[gentle trench]

if they blantently say it though?

[timid musk]

depends on what you think the word catfishing means

and what it entails

[gentle trench]

basically this

or to "troll" people on games to get stuff

really sad behavior

[timid musk]

im not forcing anyone to help someone they don't wanna help, just saying asking why they want to use it is not appropriate to begin with

[gentle trench]

👍

[timid musk]

if i go into a poster store and ask for a big poster and they ask what for, im not answering that question

it's frankly none of their business

[gentle trench]

fair enough

99% of the time tho people wanting "e-girl/e-boy" voices is for trolling catfishing behavior

only good reason for it is if they dislike their own voice or are trans

[timid musk]

the entire point is to change ur voice to whatever you want

you don't need a reason

i would hope people would troll with it

[gentle trench]

trolling as like Batman or Darth Vader

[timid musk]

but i also wouldn't conflate trolling with catfishing in its most literal sense

[timid musk]

that's not for us to decide

bad people will do bad things with any tool

[gentle trench]

that's true, but to keep those bad actors from doing it if they ask for egirl whatever I don't help them

someone else can

[timid musk]

ofc, you're not even staff it seems, i wouldn't make you do it even if you were (can't even if i wanted to)

just saying as a general rule of thumb don't ask, if someone is stupid enough to tell you they are straight up going to do it to commit fraud, ofc dont help them, but otherwise it's not your concern.

[gentle trench]

got it 👍

I'm not staff at all I just like sharing the voice changer since it's cool

[timid musk]

for sure

[gentle trench]

but not for ppl who missuse it yknow

[azure breach]

what voice changer do u guys use?

or suggest

[gentle trench]

there are multiple! what gpu do u have?

[azure breach]

AMD 9060XT which is somewhat equivalent to a Nvidia RTX 5060ti

[gentle trench]

is that Nvidia, AMD?

[azure breach]

yeah

I have the 9060xt

[gentle trench]

I'll just use google

[azure breach]

okay lol

[gentle trench]

ok cool, it's AMD

I'd say use Wokada Deiteris fork, it's the 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

[azure breach]

eyah thats wat I was abt to say

I found reviews abt wokada and I gotta say its very realistic

[gentle trench]

should be a amd version linked in the download section

[gentle trench]

oh yea! it's really good

[azure breach]

ohhh okay thanks!

tho idk if the latency is good for yknow gaming

would be nice if it wouldn't get delayed for more than 1 sec

[gentle trench]

that's all settings dependent, once you get teh application setup I could help with them

[azure breach]

okie I probably wont be setting it up rn cuz I dont even know what game imma use this on

but ill mention u if ever

[gentle trench]

alrighty I'll be here

[runic dagger]

yoo brother can you help me

[gentle trench]

Sure!

[runic dagger]

can i dm you?

[gentle trench]

I don't mind yea

I ain't helping with that stuff u shoulda read what I said earlier in this conversation

find someone else

[runic dagger]

alr

[runic dagger]

does anyone use vovox here?

[wispy swallow]

tf is vovox

[runic dagger]

vonowox

[wispy swallow]

tf is vonowox

[runic dagger]

ive heard its best voice changer so

heard*

[wispy swallow]

i mean i doubt its the best

every voice changer that uses ai is basically the same

[timid musk]

they use the same thing as in RVC

[hollow field]

which vocal model should I use

Gabox FV4 or FV6

[plucky quest]

hi guys :) just wondering if ai voice changers are safe to use in CS2? (specifically w/ okada voice changer or if there are other alternatives)

[quick helm]

Hi everyone

I am new to the community

I want to ask is it actually possible to get money selling AI automation services?

[hollow field]

yes

i wouldn’t recommend it but yes

[strange falcon]

Bro i want to train my voice model and then make cover song on it how can I do that

[rustic crescent]

Does anyone know why when I try to run the Http_ file it opens for a second and closes? Even when i re download

It’s like the final step to downloading the Mod

[hollow field]

can you provide a link?

?

nvm found it
https://huggingface.co/becruily/bs-roformer-karaoke/tree/main

also is BS Roformer SW available locally?

[queen kernel]

Yeah it is available in uvr 5 ui

[still sigil]

@crisp flame no advertising

[solemn vortex]

Who wants a free month of perplexity pro? Included gpt+ gemini pro etc

[late inlet]

Aha

[manic gull]

rip RVC 2023-2025 :(

[runic dagger]

any 1 knows how to fix robotic voice

[still sigil]

@errant portal no advertising

[winged kestrel]

im making a medical ai called SanAI if anyone cares

[winged kestrel]

Competition is high. I recommend developing a niche SaaS

[runic kestrel]

looks alive to me :/

[lofty hull]

What does your MedicalAI do?

[winged kestrel]

right now I am feeding it chest xray images

so nothing yet

[pseudo plover]

Actually reminds me I've been curious to know what the most recent RVC developments have been (if there've been much since like, early 2024) I've had RVC GUI since like 2023 so I've been curious about newest training models and such along with the tech itself

[hollow field]

uhh no its not

https://huggingface.co/jarredou/BS-ROFO-SW-Fixed/tree/main

heres a link to the model

[winged kestrel]

I thought you died

[manic gull]

I mean, theres no New things, like rvc 3

[hollow field]

https://tenor.com/view/mj-michael-jackson-kingofpop-thriller-gif-13480530

[hollow field]

rvc v2 is gone?

[winged kestrel]

is it actually you micheal?

[hollow field]

yes

[manic gull]

Idk (it's I dont know ok Michael?)

[hollow field]

oh ok

[mortal oyster]

skibid itoilet RIZZ

[winged kestrel]

67

[mortal oyster]

https://tenor.com/view/chicken-stars-gif-10427224816068102349

[winged kestrel]

my model is learning fast and getting better at detecting pneumonia

@lofty hull , since u were askign

[hollow field]

How do I make UVR export FLACs in more than 44.1khz/16bit

it'd be nice to have Hi-Fi exports

[winged kestrel]

i can tell u

[hollow field]

how

@winged kestrel

[winged kestrel]

im typing it in ur dsm 1 sec

dms*

[hollow field]

kk

can it do 192khz / 24bit

[chilly lake]

if the model is trained to produce 44k/16bit, how would you get more?

lol

ridiculous

[winged kestrel]

can i ask a question to someone that knows ai training a little

[runic kestrel]

what context

Coding as in Websites? orrr something like RVC?

[winged kestrel]

huh

no like

can i dm u

wait its fine i cracked it

[runic kestrel]

I'm not educated on sophisticated AI training for coding websites or something complex like a service or whatnot

But I can make voices

[hollow field]

btw what model is THE BEST for isolating sfx from a youtube video/scene from tv show or movie (5.1 is out of the question)

[dusty shore]

hi

[queen kernel]

Updated uvr ui ?

[hollow field]

what is that

[queen kernel]

Where you checked btw ?

[hollow field]

https://huggingface.co/spaces/TheStinger/UVR5_UI

[scenic blaze]

Why does the voice changer cut out and sound bad

[queen kernel]

I see, I talked to the the devs and they said you have to wait for a week. They are going to push updates soon.

[hollow field]

should I download all of UVR5 UI models and some other models like MVSEP exclusives?

what model is this
model_mel_band_roformer_ep_3005_sdr_11.4360
i forgot

[covert lake]

Please elaborate in #1192011222023950368

[covert lake]

Hello! HRU?

[chilly lake]

"Mel-Roformer-Viperx-1143"

https://github.com/Eddycrack864/UVR5-UI/blob/fe21c3110a9b488b7458b38235d1b1b233ec1672/assets/models.json#L14C5-L14C31

[polar flax]

mel 1143 is an old model released 1.5 years ago and before BS roformer 1296 and 1297

judging by the quality, there's no reason to use it anymore

[lofty hull]

Hello, has anyone ever worked on an agent that gets its responses from a firebase database?
Any documentation about it?

[winged kestrel]

my ai can now detect pneumonia via xray imaging

[chilly lake]

at a cost of burning a bag of coal at the power station vs a regular radiologist doing it in 0.5 seconds flat

[cold robin]

https://tenor.com/view/daebom-war-thunder-strv-103-lde-gif-25759682

[cerulean field]

Hello

[winged kestrel]

is this true

[elder willow]

does anyone have a Sora 2 invite code?

or does anyone know how to make AI vids w/out the 'policy guidelines'?

[chilly lake]

what true? that using AI is a waste of resources for something that a human do with ease?

[winged kestrel]

its not a waste of resources at all

[floral hollow]

hi

[worthy smelt]

anyone needs AI related work or projects done? im quite skilled in ML and RAG. and ill do the work with no funds, my payment is going to be the experience

if so just dm!

[spark agate]

How many epochs would yall recommend for a 2 hour dataset model

[grave gate]

Funny question

Considering we use RVC, which modifies/manipulate your actual voice to the target voice with AI, it is prone to limitations as a deep tone voice will have problems changing it to higher pitches like a feminine voice.
Is there any technology that instead of changing your voice, it actually listens to what you said and in real time replicate it with the target voice?

[minor blade]

There's no such technology like that sadly.

[minor blade]

None, you must figure yourself by training the model and testing/listening to the pth's.

[glad nebula]

rvc doesn't modifies or manipulates your voice, it's listening to what you said and replicating it with the model's voice lol

[minor blade]

this

[glad nebula]

it takes your input, then it runs a rvc inference in realtime

using that

[grave gate]

Huh if it's the AI doing the voiceover, why do your own pitch matters so much the

[glad nebula]

this type of ai audio uses a f0, unlike TTS which doesnt (so the pitch change they make are completely randomized)

you can technically train a no f0 rvc model, but it sounds very bad

(used to be a feature in mainline rvc, removed in applio)

[glad nebula]

so rather than "imaginating" pitch, it tries to follow the reference audio pitch instead

rmvpe as a f0 is nearly perfect, most of the time models learns voice cracks during the training

there are other realtime solutions besides rvc, like ddsp-svc and so-vits-svc

but so-vits is super slow and eats your vram alive, and it actually has more voice cracks/artifacting than rvc

and ddsp-svc sounds like a downgraded version of rvc

lol

[glad nebula]

plus it needs a big dataset in order to produce good results

just like so-vits

i think there is another realtime solution that i cant remember the name uuhh it was zero-shot

but the quality was also bad and limited to 22050 sample rate only iirc
edit: ok they released a 44k model, nice i guess lol

[glad nebula]

seed-vc yea lol

tldr; rvc is very powerful actually, the other realtime solutions simply do not compare to what rvc can provide

[grave spruce]

Hey guys, whats the best software to change voice by ai nowaday? was using w-okada/voice-changer back in the days, but I got a new pc and would like to try again (free and locally ran only please)

[chilly lake]

throw away 75% of your set, train with the remaining part for max 200e

[grave gate]

To me the best one which seems less hardware intensive, simple and has almost no delay is vonovox

[covert lake]

there's no recommneded epoch

[covert lake]

I'd suggest you to tell your hardware in #1192011222023950368

[covert lake]

Vonovox is only for modern Nvidia GPU on Windows 10/11

[polar flax]

voice manipulation and voice replication/reproduction are two different things

for the former, roughly you can use a VST plugin to do pitch and formant shift

or something like Vocoflex by Dreamtronics

[gentle trench]

what's this do

[polar flax]

search in youtube

[near nymph]

[shrewd parrot]

Hey, curious if there are any tutorials or people who could give me a rundown on how to setup Applio. I’m trying to train models locally on my pc. As you can see I’m a noob who has no idea what I’m doing

[scenic blaze]

does the voice changer actually make u sound like a real girl

[gentle trench]

I'm scared of what u want from it

[shrewd parrot]

That didn’t come out right

It’s ironic, that there are questions like my and then ts

[scenic blaze]

i wanna sound like a woman

[gentle trench]

please say miku and not an egirl :(

[scenic blaze]

idk the voice changer never works for me anyway

[gentle trench]

what gpu do u have

[scenic blaze]

amd radeon 6650xt im kinda confused hwo to use deiteris

[gentle trench]

all u gotta do is download it and start the mmvcserversio exe

[scenic blaze]

When i do the voice sound bad

dosent sound like a real girl it sounds like ai lol

[keen folio]

Hello

[slim maple]

bro I havnt been on this thing in like a year does anyone know a realistic girl voice. Has this advanced at all?

[shrewd parrot]

Unfortunately yes

[slim maple]

wym? explain in simple terms lol, I joined this server 2 years ago so should I use the same download or the newest one? Also which voice model is the best girl voice

[gentle trench]

Why, just why :(

Please have a actually good reason 💔

[slim maple]

me?

[shrewd parrot]

all these gooners man

[gentle trench]

Yes

[slim maple]

??? I have a yt I want to start voice trolling again

[gentle trench]

Please troll with something that isn't sorta illegal

[shrewd parrot]

its always the same

[gentle trench]

Like literally anything else but e-girl/eboy

[slim maple]

thats why im asking u guys cause I tryed this when it was first coming out 😭

[gentle trench]

There's some new stuff but u gotta find someone else to help u in your freaky activities

[slim maple]

gng what 💔 What can u do with a voice changer

alr 😭

just asking for help but ok

[gentle trench]

Don't end up like this ok?

Please

Stay away from the dark side

[slim maple]

bro I already said I was doing videos?

If u dont want to help its fine but

[gentle trench]

I probably missed that I'm kinda blind

My bad

[slim maple]

oh

[gentle trench]

I can help u out in https://discord.com/channels/1159260121998827560/1159290139609137264

[slim maple]

ok thanks

[gentle trench]

Np

[keen folio]

hi

[remote badge]

Yp whats up boys, im really new to ai but I do marketing and im trying to learn how to build ai chatbots to add to my premium marekting packages. Do yall have any advice on the best open source software to build em on?

[still sigil]

theres an entire channel called #✨│ai-help

dont mass ping there

[solar torrent]

Ok, you didn't have to ping every moderator like that.

If you mass pinging other staff one more time, you won't be genuinely getting helped, homie.

There are #✨│ai-help and #1192011222023950368 for help topics, yet you keep asking in #🧬│ai-chat which is not even a help channel.

[solar torrent]

I see a member from this server has the phrase "no AI art" in their server username. Ironically, "AI Hub by Weights" is a Discord server about AIs, which also including AI generations and AI arts. Because I count "AI-generated music" as a type of AI art, so does it?

[thorn scarab]

Give me some new AIs

I'm running out of Ai...

Why isn't any company unveiling new AI?

[gentle trench]

Technically music is a form of art it's just not like pictures

[polar flax]

just say it #🏙│ai-images

not only realistic graphics but also illustrations in wide range of styles

[charred orchid]

Some1 has an code for sora ?

[thin granite]

will a fifine mic work well?

usb am8

[shrewd parrot]

Hey, curious if there are any tutorials or people who could give me a rundown on how to setup Applio. I’m trying to train models locally on my pc. As you can see I’m a noob who has no idea what I’m doing

[drifting rain]

how do i make the voice changer work on discord

[limber nexus]

please ask in #✨│ai-help

[strong dune]

It's funny how every person whom just blatantly admits they do it to get 'robux' or any type of currency also automatically is admitting fraud. under u.s. law that falls under deceptive misrepresentation and can even qualify as wire fraud under title 18 section 1343 of the united states code

[limber nexus]

which is also the reason that if they say they are going to do any of that we don't help them

[gentle trench]

Literally

It's so sad

[strong dune]

definitely the right approach. I respect that in a way i can't even type in words. Since on discord people tend to care a lot less when it comes to money. Providing assistance could be seen as facilitating that behaviour and you guys make sure not to.

[limber nexus]

truly is the right way

[spark agate]

Wait why should I delete 75% of my dataset

[gray ridge]

@wooden jolt

#✦│chat

[wooden jolt]

cant rn

[gray ridge]

when u got time for vc

[wooden jolt]

few minutes

[gray ridge]

alr ill wait at support vc

[wooden jolt]

im here

[wooden jolt]

https://tenor.com/view/que-gif-27530657

Im gonna leave if you dont come back

[gray ridge]

am back i was at kitchen

[restive rune]

Jr.

[warm otter]

Sexo

[gentle trench]

https://tenor.com/view/all-roads-lead-to-rome-clock-ticking-white-rabbit-alice-in-wonderland-gif-11005866172848930328

[chilly lake]

because 2 hours will be excessive

but it is fine, your model will be done much sooner than you may expect that way

[spark agate]

i can see where you are coming from tbh, but the whole dataset includes various ranges and its all studio quality audio

so i think theres nothing in there that could make the model worse

like the quality remains the same all thru out

[chilly lake]

training with a huge dataset for many epochs erodes the model's knowledge

30 minutes is more than enough

[spark agate]

could you elaborate a bit more? i just wanna make sure i understand so i dont like, make any mistakes and stuff

[chilly lake]

what else to add? the model is trained with 100+ voices to learn the difference between them, you only need a small sample, like 15 minutes to retrain the default voice characteristics

with 2 hours you dump 8x more audio that is being used by other speakers, and since youre training on top of the pretrain the model slowly loses information about other voices as it tries to realign towards the new data

the trick is to train the model just enough so it realigns the default speaker but does not forget how other voices work

[spark agate]

ahh i see

[spark agate]

but what if i want the model to have a very good range (from lows to highs when singing) and ive tried training 30 min model with the dataset but i felt like 30 mins was a small time for the model to learn very well all the ranges

and therefore it was a bit more inconsistent when hitting different ranges

[chilly lake]

well, the default model has low and high voices known

it should be able predict how a new voice sounds with different pitches

[spark agate]

wouldnt having more dataset make the model more consistent tho? (considering the whole dataset is all great quality, raw vocals and good mic)

like, theoritically shouldnt the model be better and have less glitches and stuff and be better at pronouncing if it has a longer dataset

thats how i think about it, but do correct me if im wrong lmao

[chilly lake]

that's not how it works

the model learns phonemes and pitches in relation to individual voices and how does a spectrogram look like for each

when you train a regular voice model with speaker 0 it realigns it to march an existing voice from speakers 1-108 and learns how it differs

then when you infer it takes the newly learned voice, phoneme and pitch from the input audio and predicts a spectogram (how the audio should sound with a new voice)

and based on this prediction it generates the new audio

[spark agate]

ah i see

so you are saying that a 30 min model will sound better than a 2 hour one 100% of the times?

(obv considering that the data is all good and the same quality etc...)

[chilly lake]

I can't say... give it a try

use 15 min, 30 min, 1hr

train, compare

ideally you want a consistent voice in the set

[spark agate]

I’ve done a 25 min and a 1 hour one and it relatively sounds very similar but the 1 hour one is better with higher ranges and stuff

[chilly lake]

dont think having singing and speaking will do any good

[spark agate]

yeah ofc

It’s only rapping/singing

I’m just varying the ranges and the tone

[spark agate]

I also forgot to ask, what specific problems would you think could come with me training the 2 hour model

If you have any in mind atm

I’ll train it anyway just to test it

But I’m curious

[edgy helm]

Wow

[main nest]

<@&1159293140440723499> some mod pls take a look at that.

thanks for the swift reaction :)

[lament mulch]

you cant handle the most basic things in 2025 bro

[still sigil]

eating

[still sigil]

you cant read in 2025 bro

[main nest]

hate speech isn't "the most basic thing".

[solar torrent]

Not-embeded image from newer member especially is always suspicious, but I'm kind of not allowed to delete one's message here.

[still sigil]

what are you, an 1800s peasant?

[lament mulch]

i mean im trolling in most servers rn

im not but my friend is a guy thatd be willing to buy

[still sigil]

read the #📜│rules

[signal laurel]

whats the best tool for text to audio ?

[lament mulch]

nah who reads those anyawy

pointless lmao

[main nest]

That wasn't trolling, that was just being a basic lame-ass bigot. :-P

[still sigil]

you when i give you a mute to read them

[solar torrent]

For even more severe incident, you can use the modmail bot to contact the moderator in this server.

[lament mulch]

im probably never gonna interact with this server again

[solar torrent]

Your choice.

[lament mulch]

im back

[main nest]

<@&1159293140440723499> Scam advertising here
(somebody please tell me if you rather have stuff like this reported in a different way, instead of just pinging all of you ;) )

[timid musk]

generally nothing but time

at a certain point your hitting diminishing returns

you can do it it's just not worth

unless you want to train a model for extra 4 hours for a negligible difference

and a model will be different every time you train it, trial and error

[minor blade]

Small tip: Mixing a lot of varied tones of a same voice won't work well always.

With certain singing voices it's better to just make models focused only on their distinctive singing/rapping voices of the same person. Even more if the target sounds like a completely different person when doing rapping/soft singing/aggresive singing/etc

Believe me, i know this thanks to experience.

[glad nebula]

idk why this server recommends not training big datasets

you replace more things of the pretrain and rely less on it with a big dataset

relying on a voice that it's not from the model will never be good

and the losses just get much better, gradients become smaller

you also get less robotic/metallic moments, since these are actually coming from the og pretrain knowledge (the dataset used to train the og pretrain has close to 0 harmonics past 4k)

[glad nebula]

fm gets much more stable, mel loss is lower

there is no diminishing returns wtf

you get a more stable model

lol

[glad nebula]

also, the discriminator gets super strong with small datasets, which makes the model even more metallic/robotic :U

obviously if u train a big dataset it has to be good tho

[minor blade]

Yup, for sure

I encourage a lot using big datasets only if these are well made.

[fathom wharf]

new here
what are u guys even using AI for? im hearing abt gazzilion of new routine killer AI every day but after looking at them i just cant imagine scenario where that would be useful

[spark agate]

Do larger dataset models tend to hit the lowest point earlier than models with less dataset?

[glad nebula]

bro this "lowest point is the best epoch" is just plain wrong

its just noise

lol

[spark agate]

oh

That’s jus what I’ve always heard

What’s the best epoch then?

Do you just have to test every epoch

Or

[glad nebula]

|| 👍 ||

basically the best epoch is the one you like the most

no loss graph can tell you which one is the best

[spark agate]

Got you

[spark agate]

How did the “lowest point” thing come about then lmao

Cuz that’s what I’ve always heard

[glad nebula]

same source who told everyone to train 200e ig

[glad nebula]

i know where this came from tho, back then when so-vits-svc was a thing they trained 500 epochs to 1k

[spark agate]

So what’s basically like, the correlation between epochs and how good the model is, like as the epochs increase

[glad nebula]

understanding epochs is quite simple
an epoch is everytime the model has seen it's full dataset
when you train from scratch (no pretrain) and you use a big dataset (100 hours) you might wanna train a lot of them, like 500 or even more, thats because the model has no understanding of speech or anything

but when you train with a pretrain (og pretrian, legacy, klm, etc) the model already has a base, you're just meant to do small adjustements so the model forgets the pretrain voice, but doesn't forget everything else

by training 1k epochs with a pretrain you're just telling the model to forget everything it learned, rather than just learning your dataset voice

you don't need that much of them when you use a pretrain, it takes a small amount for the model to forget the pretrain's voice

[spark agate]

Ahh I understand now

So you would think that a 2 hour dataset models should theoretically be good with lower epochs than a large amount

[glad nebula]

what i do is save every 10 and train for 100epochs, but some want to train up to 200e

[spark agate]

Say: 200 vs 600

[glad nebula]

correct

at every epoch, the model extracts more features from your dataset, but also, forgets the pretrain even more

[spark agate]

cuz I left my model training over the weekend to 650 epochs, you think this may be too much?

[glad nebula]

you want a sweespot where it sounds like your intended voice but at the same time doesn't fully forget the base

[glad nebula]

tbh i have tried doing multiple tests with the epochs and everytime anything above 100 epochs overtrains

[spark agate]

Oh wow

Regardless of the dataset time?

[glad nebula]

yea regardless of dataset size

[spark agate]

Oh wow

[glad nebula]

for context my pretrain took 110e iirc and that was a 50 hour dataset

[spark agate]

Insane

[glad nebula]

i used the og as a base

so imagine that