✨│ai-help

[fringe heron]

Acent?

[plucky crown]

Yes he also has an american accent and im using a different language and his accent is seeping in to my pronounciation

[fringe heron]

What

Wdym sleeping

Baked in?

[plucky crown]

Well in the recording im speaking in a different language and when i convert it with the model and i play the converted it made the language im speaking like its being spoken with a foreign accent to it instead of native

[fringe heron]

Whats you acent?

[plucky crown]

American accent speaking my southeast asian language

[fringe heron]

Ik same issues with my ones

[plucky crown]

It sounds like this

But in my recording im speaking like a native i have no accent

And the result has accent when i convert it

[fringe heron]

Yea no wonder

[plucky crown]

Now what hahaha im done for

[fringe heron]

You convert and get the target acent

[fringe heron]

No

Maybr you are in luck

[plucky crown]

How

[fringe heron]

I am working on a way to fix this or to atleast make it way less

But it requires quite some work

Manual work

[fringe heron]

Its a bit complex to explain to someone that has little experience

But the goal would be to first teach the model how your voice is

And my method is not possible to do in the vanilla applio, i had to modify it by myself

But without changing how the model works, so it can be ran in any rvc client

I still need to fully test this

[brazen holly]

It's normal, the model was trained for english voices.i have the same issue in French

[fringe heron]

The model learns acent too @plucky crown

[brazen holly]

well it's doing an english accent for me when I speak in French

[fringe heron]

And speaking style if given enough dataset and time

[fringe heron]

Sounds cursed right?

[brazen holly]

ye

[fringe heron]

Even glitches

[brazen holly]

ye

how do I make it better then?

[fringe heron]

u experienced with code about rvc?

[brazen holly]

nop, just discovered all this some days ago

[fringe heron]

Okay

[brazen holly]

but I know Lua, a bit of c php java

I'm not new to programming

[fringe heron]

Sweet, if you want i can try to explain you in a very easy way

[brazen holly]

would gladly appreciate

[fringe heron]

So the model is not a single "brain"

The file you save is made out of 5 "brains"

The first one is called text enc and sees pitch and phone (the actual linguistic content and speaking style (dominated by pitch)

[plucky crown]

You can use ai, i can give you a google ai premium account so you can make it for free as long as you help me aaa we are all stuck

[fringe heron]

The second one is called post enc and its job is to see the real target speaking style so it has access to every speaker dependend info

[plucky crown]

I searched online and most voice clone in github its mostly text to speech and not speech to speech aa

But you can look into their github's code and reference the folder path in google ai to read the code on how they make accurate voice cloning in very little audio

[fringe heron]

The third is called flow, The flow during training is taught to convert the post enc output into text enc one

The flow is reversible, (during training it will convert text enc output to match what the post enc would have been

Then there is the decoder wich takes thr post enc output and tries to make audio out of it

[plucky crown]

It will be a antigravity premium acc tho the google ai one (same like chatgpt and claude)

[fringe heron]

Then there is the emb g wich conditions everything as a switch, if 0 speak like person 1 if its 1 speaker like person 2 and so on

[brazen holly]

mhm

[fringe heron]

At inference the post enc is not used btw

As its not needed

Now there is a loss that forces the text and post to be close, do this means the text will change its behavior to match the post wich sees real speaker dependet stuff

[plucky crown]

Do u want too lemme knoww anyone up for it? Or is it possible?

[fringe heron]

If trained for enough the text will learn to encode a different acent than the starting point

So to fix this, or partially fix this

We train first the model on our voicr first but not like the original way wants, we lower the loss that make the text close to post and we sample the imput for decoder from text without using the post

Now we can backpropagate all the losses truh the text too (before we chouldnt becouse the dec was getting post)

Now given time and data we will have a model pretty much 1:1 to our voice

Now time for target

Now we freez the text, flow and emb g lower and let only the post (wich is useless in this setup) and decoder update

So the dec is forced to work only with how our own voice sounds and try its best to mstch target

This wont fully remove the target acent but atleast it wont glitch the voice

Or have some wierd mismatches

And very important, train eith a precision bf16 or f32

[brazen holly]

dang

[plucky crown]

So we have to create a new version with new code added ? So it adjusts no matter what language we have?

[fringe heron]

Appriciate

[fringe heron]

I have already done it

But still testing

[fringe heron]

Hopefully

[plucky crown]

Do we wait first or you want to fiddle with the code, i can give you google antigravity pro for free for a month if it helps getting our problems fixed cuz elevenlabs i looked it up they dont accept voice files freely from any sources without verification

[fringe heron]

Ill go for some time so ill see yall later

[fringe heron]

Ill chat later :/

Sorry

[plucky crown]

Hopefully your version works aaa we done for

Who is the creator of vonovox and applio can they be contacted

Are they in this discord server

[plucky crown]

What do u use now to fix that issue

Or nah empty handed like we are

[proven hill]

what for

[fringe heron]

Not needed

[fringe heron]

No, in short you need to record yourself, train a model with the new method on your voice then take the model of you and train it on the target with the new method

[fringe heron]

To be clear i wrote that in a hurry so if its not precise i am sorry, also its semplified 🙏

[fringe heron]

@brazen holly @plucky crown after finetuning the finetuned pretrained (finetuned on my voice) on a voice that has a completly different acent i can tell you that the voice still pretty much keeps the original acent (mine) and speaking style, mostly important it doesnt glitch

I listened to a sample now as i am back home

[wicked fern]

hello im searching for the best uncensored ai
i tried grock and its good but there arent a lot of free messages

[fringe heron]

i would like to help but honestly i dont know

:/

[fringe heron]

sorry for the bad explanation, now that i have more time i re read it and its ass, let me provide you a better one,

the first one is called the text enc wich sees pitch and phone the actual linguistic content and speaking style dominated by pitch
second one is called the post enc its job is to see the real target speaking so it has access to every speaker dependend detail
the third one is the flow, during training the flow is taught to convert the post enc output into the text enc one, the flow is reversible so later at inference you can run it backwards and turn text enc output into something that looks like what the post enc would have produced
then theres the decoder, it takes the post enc output and tries to make audio out of it (but remember at inference the post enc is not used we feed the decoder from the text enc instead, after the flow has done its thing backwards)
finally theres emb g wich conditions everything like a switch if its 0 speak like person 1 if its 1 speak like person 2 and so on
now theres a loss that forces the text enc and the post enc to stay close, becouse of that loss the text enc will change its behaviour to match the post enc wich sees real speaker dependend stuff, if you train like this for long enough the text enc learns to encode a different acent than the starting point
so to fix this (or partially fix it) we do things differently, (and all trainings must be done with bf16 or f32)

we train the model on our own voice first but not like the original way, we lower the KL loss (manually from the config) that is the loss that makes the text enc and post enc stay pretty close together, and instead of feeding the decoder from the post enc we sample the input for the decoder directly from the text enc without using the post enc at all
now we can backpropagate all the losses truh the text enc too (before we chouldnt becouse the dec was getting post)
given enough time and data this gives us a model pretty much 1:1 to our own voice
now time for the target
we freeze the text enc, the flow and emb g, we keep the KL loss low (same value as before) and we let only the post enc (wich is pretty much useless in this setup) and the decoder update
this way the decoder is forced to work only with how our own voice is and try its best to turn that unchangable input into the target, this wont fully fully fully remove the target acent but atleast it wont glitch the voice or have wierd mismatches.

[viral mason]

Vonovox only has two main settings, block size and pitch

Unless u don't have the most recent beta

[west temple]

ye but my vram is limited

[viral mason]

I've never had an issue with that on my end with any software

the 5070ti has 16gb

Vonovox shouldn't be too stressful on ur end either tho

[hardy yew]

TBH if Kuru wants to minimize VRAM usage for whatever reason, then sure, just pick the app that uses less VRAM

Deiteris/Tg develop are probably the same, but Vonovox has various stuff replaced, like different RMVPE etc

So I guess VRAM usage might differ. I wouldn't expect it to be a signifcant difference, but I've never checked

[stoic moon]

how do i make realistic anime videos

[crystal rivet]

-colab

[patent trellis]

📒 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

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Tg-Develop Fork**

by Tg-Develop
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
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

[viral mason]

watcha lookin for? training maybe?

[winter mason]

trying to cosplay micky mouse
1660 super
windows 11
need help in server settings audio

althought client one is not wroking

[severe egret]

I have a question for you lovely people. Do W- Okada still work for these voice rvc models?

[viral mason]

ye they work, that's the main point of them now besides ai covers

[severe egret]

Did W Okada get a new update? because the client mode isn't working and I am very confused

[viral mason]

client mode stopped working back in 2024-25 idk why

[severe egret]

Oh no so what do I do?

[viral mason]

you probably have an old voice changer what gp do u have? (nvidia or amd)

[severe egret]

Nvidia

[viral mason]

is it like, a 3070, 1660, specifics would help better ^^

[plucky crown]

It worked?? Is it convincing?

[severe egret]

It's a Nvidia GeForce Rtx 4060 GPU and the graphics card is a Intel UHD graphics

[viral mason]

u should try vonovox, it's peak

[plucky crown]

Is there a tutorial on this it doesnt work on my end it made the .pth file and index from applio sound worse

[severe egret]

What is vonotox?

[severe egret]

Oh no 🙁 really??

[plucky crown]

Yeah im not sure maybe theres something wrong with how i put the audio file in but i dont think so?

[severe egret]

I just wanna try new voices out 🙁

[plucky crown]

@viral mason is there a tutorial on vonovox maybe im doing it wrong

[viral mason]

most recent up to date realtime voice changer, it's the best one for nvidia currently

https://huggingface.co/dr87/vonovox/resolve/c8034f5f6d50648a8109bb4f847182362e2b779b/Vonovox_beta_17_11.zip

https://software.muzychenko.net/freeware/vac470lite.zip

[viral mason]

it all sounds fin to me tbh

[plucky crown]

Can i see your settings? Can u make a short Video tutorial pleaseee like what to import then settings? Huhuu

[viral mason]

sure

[severe egret]

Send that short video too because there is no YouTube videos about this at all

[plucky crown]

@viral mason can u also send a video on how to do it for @severe egret

So i can watch it too and can be added whenever anyone asks about vonovox

[viral mason]

there is no tutorial for vonovox, it's very simple software

[severe egret]

Really I still would like a video from you if you can

[viral mason]

what would I sbow?

[severe egret]

How to do it

[fringe heron]

Yea

[fringe heron]

Mostlikely his issue is with the model itself converting his voice, not the software

[viral mason]

Oh

[fringe heron]

Always the same issue with acent and speaking style

The converted voice sound wierd

[viral mason]

Yea that makes sense, heavy accents when speaking English or even a odd way of speaking can cause models to sound off

[fringe heron]

Ye

[amber raft]

ooh yall are figuring out sound ai? I need to get into that...

[fringe heron]

I am trying to fix it for my use

[amber raft]

yeah

tts?

[viral mason]

I'd say I'm pretty good at making and working with ai voice models

Personally what I make is for realtime use or singing but can also work for tts

[amber raft]

lol

[fringe heron]

We could call rvc tts

[amber raft]

wait whats rvc XD

[viral mason]

The best tts software I have seen so far is between fish audio and eleven labs

[amber raft]

im new too

[fringe heron]

New new?

[amber raft]

yeah like born 2 mins ago new

[viral mason]

I love this person

[fringe heron]

Sweet

[amber raft]

i mean im not new to ai tho

[viral mason]

Rvc is basically just voice cloning stuff

[fringe heron]

So you want to get into sound ai?

[amber raft]

so like voice style transfer or like training tts on your voice type thing?

[viral mason]

Training a audio file of anything like SpongeBob or just some random person or even sound effects so that it can be used as a voice changer or to replace a singer in a song

[amber raft]

tbh i'm like really deep into image and video ai alr. I like am kinda into llms and stuff... but yeah!

[viral mason]

Or for tts as well

[amber raft]

i see so both tts training and realtime voice changing?

[fringe heron]

Yes

Pretty much

[viral mason]

Ye it does both

[amber raft]

how much vram for running like training (on a voice) and running an expressive tts model?

[fringe heron]

Depends on how big it is and how big the batch is

@viral mason may i ask you how you train your models?

[viral mason]

Oh I use Applio, specifically on the Kaggle website

Applio tho can be downloaded and used locally on your pc if it's good enough

[amber raft]

dear Sapphire... MY BAD 😭

[viral mason]

Ignore Sapphire it's just set to do that forever

[amber raft]

lol yea

[viral mason]

Unless u need actual help the bot is kinda just annoying

[fringe heron]

Yea, but you do somrthing particular

Or no

Settings

[viral mason]

It's dependent on each model since they differ in length

But mostly I go for batch 4

8 for bigger models like 40+ minutes

Models that are just sound effects like the Minecraft ones I go for batch 1

[amber raft]

is a batch how many opperations at once or like... how long of text or what? also are we talking like 2-8gb or more like 24gb vram lol

[viral mason]

I'm not entirely sure but u can read up here on the whole thing

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[viral mason]

The applio docs are quite nice

[amber raft]

lol thanks!

[fringe heron]

So you never touch the config file tho

[plucky crown]

Woah!! How!

[amber raft]

well guys if yall ever need ai image help or anything lmk... i have like 40 more hours of prep before training my model, so have fun!

[fringe heron]

The batch is like a group of files that are loaded in a specific moment that the ai sees

Once that is done the next one is loaded

[amber raft]

so just affects speed right?

[fringe heron]

Ye

[amber raft]

cool

thanks!

[plucky crown]

Also im about to buy voicemod,voicewave,dubbing ai subscription cuz im about to give up

[fringe heron]

I use a batch like 20 or 16

[plucky crown]

Give up on local ones

[fringe heron]

They are even more ass

I believe

[plucky crown]

Did it work how convincing is it

[hallow thistle]

They are.

[plucky crown]

Really

[fringe heron]

I told you already above above

You need to teach the model your voice style first

[plucky crown]

How do i do that

[fringe heron]

Read this

[viral mason]

What config files?

[viral mason]

Holy

[fringe heron]

About my custom code its not ready yet

[plucky crown]

Did u try them or have a subscription i like joined these 3 voice changer's discord servers and asked on the chat to tell me if its convincing and no direct yes yet

[fringe heron]

I used even 40 once

7 hours dataset eated in 1 day, 1105 epochs

[fringe heron]

In the training folder

You can edit cool stuff

Such as segment, loss weight, hops and more

[hallow thistle]

If I say no, it means no.

[fringe heron]

First you need to record yourself speaking

Thats step 1 no matter what

[plucky crown]

So which step is it does that mean we train our voice along with the target voice wont that combine the voice into one?

[hallow thistle]

I'm quite clueless.

[fringe heron]

No

You train a model on your voice

First

[fringe heron]

Follow this if you intrested

[plucky crown]

Then what next

[fringe heron]

Here @hallow thistle

[hallow thistle]

But I already know how to train a voice model.

[plucky crown]

So we train the applio with our voice and make a .pth file and index file?

[fringe heron]

You need to record yourself and once i will give you the modifed code train in a specific way

Then the result you train it again over a target, in a specific way

[plucky crown]

How much length of the recording is needed for our voice?

[hallow thistle]

I run Applio RVC on Kaggle, by the way. The first try I tried to train a voice model, it worked for me.

[fringe heron]

You have a particular acent?

[fringe heron]

78 hours

[hallow thistle]

What does this mean?

[plucky crown]

Woah you recorded yourself for 78 hours?

[fringe heron]

Actually wait yea sorry

[plucky crown]

What do we say in the recordings though

[fringe heron]

The target have your acent?

[fringe heron]

Ye

[plucky crown]

Woah

What are you saying in the recordings

[fringe heron]

Random stuff

[hallow thistle]

I'm so confused. Let try ask something more understandable.

[fringe heron]

Be you

[fringe heron]

You said you train models

So i am asking, the target, the speaker you train the model on has the same acent as you

?

[plucky crown]

Do we need to be in many emotions like shouting,angry,screaming,happy etc etc in the recordings? Since in applio like u guys recommend the dataset should have many emotions

[hallow thistle]

Do you mean like "accent"?

[fringe heron]

Yea sorry

[hallow thistle]

These are voice models I made, by the way.

[fringe heron]

I am not native speaker 🙏 and i am not the best

[fringe heron]

Yea but i have no clue of they speaking so

[hallow thistle]

[plucky crown]

Let us try post it in the models channel cmonnn aaaa

[fringe heron]

Hoew long are the datasets?

[plucky crown]

Since youre a support of this server i bet this is like very high quality

[hallow thistle]

Clueless huh? I have already uploaded some to #1175430844685484042 just letting you know.

[hallow thistle]

Usually few minutes long, some of which were under a minute.

[fringe heron]

Makes very sense of why it still sound fine to you

[plucky crown]

Also @fringe heron did you record in a professional microphone like no background noise? Or you use regular mic and used a background noise cleaner software because i will do what you said about the 78 hours recording

[fringe heron]

I suppose trained on an english pretrained

[fringe heron]

No, it can be a bit nosie, jsut not too much

[plucky crown]

Did you use uvr5 to clean the recordings too or other software? So i can do your step 1

[fringe heron]

No

[plucky crown]

Btw you trained in applio too right ? Is it ok if my applio exe is in an external ssd? I saved up for a 2tb ssd and got it just for the local voice changer's files and dataset

[hallow thistle]

Bro.

[fringe heron]

Trained on an english pretrained (if so) on such small amount doesnt erase the pretrained english bias, if you speak english then it chould be why it sounds fine for you

[plucky crown]

Is it meant to be not in external? Did i do it wrong

My c drive is almost full though

[fringe heron]

Ofc i use applio, i modifed that, you should use on main drive

[plucky crown]

Will it not work if its in external ssd? When i ran applio for 16 hours training it reached 77C° on the external ssd

[fringe heron]

It will work but i raccomand the main drive

[hallow thistle]

I was going to tell you about NVMe SSD, but I just doubt because you were focusing solely on RVC.

[plucky crown]

It is a nvme ssd

Kingston fury 2tb i got it for equivalent 65usd back then. Now i look up the prices it shot up

[plucky crown]

Are there pros if its in the c drive

Its kindof very full now aaa

Well python is in my cdrive

[fringe heron]

You said you where running on a external, i still dont know if its a usb or not

By external at this point idk what you mean

[fringe heron]

This is the main C or

The external

[plucky crown]

External

Wait ill take a pic of it

[fringe heron]

Run it inside of it

I remembered just now that applio has its own runtime

[plucky crown]

Here is the exact model

The one with the circle but i take it out of the enclosure and put a handheld fan over it its the only way to decrease the temp that is 77C

The ssd goes down to 42C if i do that

Thats also why i want it to stay to external ssd

[fringe heron]

Might be a bit slower but wont cause issues with python

[plucky crown]

Since i can lower the temp while its training since its running in external ssd with a handheld fan pointing to it

[fringe heron]

@hallow thistle over 70 degress ssds goes in termal throttling no?

[hallow thistle]

Applio RVC can run from any drive, I tried, as long the batch file can detect PATH inside the folder. But anything else, running Applio RVC from C drive is perceived as more stable.

[plucky crown]

It seems so

[fringe heron]

I have it on a drive that is not C but its not USB "plug and play"

[hallow thistle]

Is this just a closure? And which NVMe SSD did you put in it?

[plucky crown]

E drive the external ssd i put applio and vonovox here

And also the dataset is here too when i was training but python is installed in c drive

[fringe heron]

You will need patience with him

[hallow thistle]

[plucky crown]

I dont get it is it not meant to be in external like it affects the voice aaa

[hallow thistle]

I asked you what SSD you did put to that enclosure, something like WD Blue SN5000 or Kingston Fury Renegrade, and you answered me "it's where you put data to it".

[plucky crown]

Kingston fury renegade yes

@fringe heron also what else did u noticee did it match your accent noww

After training your voice have you tested with short target voice like the latest tts voice clone that only needs small voice to clone it?

[fringe heron]

It matched my speaking style and acent, i can make anytype of wierd noises and the ai doesnt fart

[fringe heron]

I trained the model of myself on 7 hours

Of target

The accent was american and mine is italian

No problem

But I was limited by the quality of the voice model of myself, which was just trained only on 24 hours as a test

[plucky crown]

You can got to open source github of the latest tts ones i think it was voicebox , omnivoice etc maybe you can make upgraded versions of the code that can handle decreased until decreased voice target length amount to clone itt

[fringe heron]

I still raccomand atleast 60 hours

[fringe heron]

What

I got just applio, changed 1 file, and trained

[viral mason]

I've made good voices with less than 10 minutes

[fringe heron]

This is a different story

Not the target

[viral mason]

Oh are you guys talking about pretrains?

[fringe heron]

Yes

Finetuning a pretrained on your voice

In a different method

[viral mason]

Oh ok, I got no clue on those since I've never made one

[fringe heron]

If you intrested ill show you the msg

Wait

[fringe heron]

Game plan @viral mason

[plucky crown]

Just try to check the code of the text to speech voice cloning opensourced githubs (that require short amount of target audio to clone it) for their voice training code maybe we can make it adapt for voice to voice changing in applio code where it can do that in lesser target audio length? There are latest ones released ill send the links wait upo

[viral mason]

Gulp

Long ah message that I'll read in the morning

[fringe heron]

No, the code of applio, the one you downloaded

Modify 1 file

And train

Thats it

[hallow thistle]

For some voice models I made, I used different "pretrain" models.

[plucky crown]

Which file is it

[fringe heron]

train.py

[plucky crown]

Is the change you made like massive and suchh

[fringe heron]

No

100 lines

[plucky crown]

Also lemme know beciful maybe you can make it better i still can give you a google antigravity pro acc if you plan to experiment more on the applio code ill gladly help

[hallow thistle]

I don't need Nitro from you.

[fringe heron]

Yall always mispelling my name like that, I thank you but i dont know if i need it

I have modified the code already

[hallow thistle]

Generally, I do not need to modify any Python file in any Python environment. I simply run as it, unless a fellow model maker or an engineer member tells me to tweak something.

[fringe heron]

Cool

Here is needed for people that need max quality for their voice

Like singing

[plucky crown]

Wow so you are an engineer member wow

I thought you were ai back then

[fringe heron]

:|

[hallow thistle]

No. I'm just a helper. Look at my account roles.

[plucky crown]

Since like for months you refused to answer about my gpu question and i was second guessing maybe im talking to an ai lol

[fringe heron]

[plucky crown]

So you guys coded applio or vonovox?

[fringe heron]

No

The only thing I did is modify 1 file

[plucky crown]

Alright imma find first what i can read with variety of shouting,happy,angry,singing to fill 78 hours, ill let you know immediately once i finish aaaa

[fringe heron]

For the noise in can tell a method that i used

[plucky crown]

Also if you sing in the recording can the voice model sing too? Since if u asked singing i assume the quality isnt good still?

[fringe heron]

I used krisp for noise

[fringe heron]

The point is to make the model handle perfectly your voice

So yea later it should

[fringe heron]

Krisp thought pay attention on loud shots and stuff

You might want to do them without krisp

[plucky crown]

Did u try singing,shouting,etc what did it sound @fringe heron is it still convincing?

[fringe heron]

No singing, i dont sing but yea shouts yea

Vocal fry too

Random noise from mouth

[plucky crown]

How about burping or coughing

[fringe heron]

Any type of noise that is doable from you

[fringe heron]

Ye

But you need a lot for breaths and cough

[plucky crown]

In voice changers some when its just electric fan noise it confused it as a voice

It doesnt do this right like you tested

[fringe heron]

It might do it still

As long as its not in the training files it should be fine

[fringe heron]

But for me in the dataset of the test for my voice i hade keyboard click and mouse and they dont cause the quality to degradate

I also had hiss in the back ground

In half of the dataset

Also fine but if possible remove it

[plucky crown]

Me during recording

https://tenor.com/view/guy-arguing-guy-talking-to-wall-talking-brick-wall-gif-18667615

[fringe heron]

Call a friend and record in the meanwhile

So you dont talk to a wall

[fringe heron]

Also let me tell you as for now with me worked but you might get different results, this is to improve the result a lot but might not be perfect okay,

So in case of it not being perfect as you might think I am sorry this is still in testing for me too

Should work very well and this to me is worth the try

[void flume]

The rules and policies on this discord server requires helpers to ask you "How you are planning on using the voice changer.". That is what the 'goal' refers to.

[void flume]

(edit: removed a number of posts)
I see

I'm sorry, I'm not allowed to help you any further, since we have rules against using AI to troll others.

[fringe heron]

@plucky crown if havent done already also download audacity, you will need it

[void flume]

I feel like a jerk now. >.>;

sigh

[fathom blade]

do you think pharma actually has potential on AI?

because idk

privacy

[fringe heron]

Same

[fathom blade]

are they actually safe from AI

because realistically

you put too many sensible data on ai

[fringe heron]

Its less likely to be fully replaced by ai

[fathom blade]

sure

im not referirng to that

they've trained models for years now

[fringe heron]

True

[fathom blade]

wait sei italiano?

[fringe heron]

Si

[fringe heron]

My bad you are right, you are intrested in privacy mostly

About privacy idk but i dont count on that

[proven hill]

no way

[sacred temple]

Guys,it's been months,I was making voice model trainings,and now the rvc2 disconnected doesn't work on google colab.What should I do?

[viral mason]

Oh yea the disconnected colab has been dead for a while, you should switch to applio

Kaggle is the easiest option

Colab is ok but not as good as it only gives like 4 hours max

[proven hill]

whats your gpu?

[sacred temple]

Not having gpu at the moment

[proven hill]

then use applio on kaggle

[dim shuttle]

What's up, guys I was looking for a voice changer in real-time that is also an open source and has the ablity that make me put a voice model btw i have an integrated GPU so any recommendations?

[viral mason]

integrated gpu means u basically can't use it locally, only option in using one on browser

[dim shuttle]

like google colab

?

[viral mason]

Kaggle is better but yea

https://docs.aihub.gg/realtime-voice-changer/cloud/tg-develops-w-okada-fork-cloud/#kaggle

this is the best I can offer since I have zero knowledge using any voice changer not locally

[dim shuttle]

thanks

[viral mason]

no problem! if u need help @ the helpers role

[void flume]

I basically have to ask due to the rules and policies, but what do you want to use it for? (feel free to check the rules in id:guide )

||hopefully its something normal... x.x;||

[dim shuttle]

oh just trolling

[viral mason]

what kind? like what models are u wanting to use? maybe some Anime ones?

[dim shuttle]

yeah something like that. I'm not like taking it professionally

[viral mason]

ok so ur normal

and don't want to use those weird egirl models

[void flume]

pranking is fine, so long it's not harassment, psychologically abusive or erp basically

[dim shuttle]

I'm not into those sadistic things

[void flume]

sorry, trolling can have a range of meanings with people and... unfortunately people do use voice changers for those reasons (and even post videos of their actions on socials)

Some just have fun with friends (sudden spongebob voice for example), others ... oof

[viral mason]

most people say trolling and mean they want to use the egirl models to mess with people and then either scam them or catfish them

it's basically required to ask what the person means when they just say "trolling"

due to the bad rep

[dim shuttle]

fair point

[plain cloud]

greetings to my fellow sentient clankers , i have questions regarding GPT-soVITS please ping me twice if you have any idea about it

[fluid harbor]

how can i use the voice models etc

[proven hill]

just why

[marsh tapir]

has anyone used Kimi on desktop? is there a way to get it to access my files please? for some reason it comes it cannot access my files

• Goal (e.g., TTS, AI Covers, Roleplay): Get Kimi to access my filesystem
• Specific Issue: Kimi is not using my filesystem
• Full GPU Name: RTX 3080Ti
• Operating System: Windows 11 Pro
• Tutorial Link used: none.

[viral mason]

there's multiple ways to use them, what are u wanting to do

[whole carbon]

How to make an AI Cover model?

[clear vine]

hey guys is there a tutorial on how to use the voice models on a discord vc

[grave oasis]

whats the best settings for the Every Song of The Album "Thriller" By Michael Jackson
Link: http

[plain cloud]

Vonvox does not feel close enough to the sound

I need more accurate sound

[zinc sequoia]

Hi guise can n1 tell me how use Kimchi on CloudCod

[proven hill]

gpt sovits is realtime???

[plain cloud]

No

[proven hill]

then why?

[plain cloud]

Huh as I said vonvox does not sound accurate

So I am now going to use the best tts

[fringe heron]

What do you need to do with it?

[plain cloud]

Vc

[fringe heron]

Yea but for what

[plain cloud]

Uh sorry but could you rephrase that uh

For games ? Like role play / cosplay type

[fringe heron]

Like singing

[plain cloud]

??

[plain cloud]

No no

General talking

[fringe heron]

Okay

[fringe heron]

Give me a sec good sir

[fringe heron]

Okay so, you said you tried vonovox and you are not satisfied

[plain cloud]

Yesss

[fringe heron]

May i ask you what the model was

[plain cloud]

Ultimately the input is the problem

[fringe heron]

True

[plain cloud]

Model as in the voice ?

I made them

[fringe heron]

Even better

[plain cloud]

I believe gpt thing won’t be as fast as vonvox but for me quality beats speed

I just need to figure out what transcriber to use to convert my voice to text

Then constantly fill gpt with the text from the transcriber

And the it converts and outputs it in the fly

[fringe heron]

Sorry but i have a thing to fo

Please wait

Sorry

[plain cloud]

Take your time it’s alright

[fringe heron]

Will be quick

[fringe heron]

I am back

[fringe heron]

Gpt wont be provably as good as you think

But I dont know for sure

[drowsy hedge]

Has mmvc update or is it still the same? i haven't used this for years and i don't know if i have to update it or not

[fringe heron]

I dont know, always used the new version, idk about older ones

[fringe heron]

May i ask you about dataset lenght, pretrained used and if target have same accent as you?

[drowsy hedge]

What's the newest version?

[fringe heron]

I dont know, i always used just one , never bothered checking olders

I am not on pc so i cant just check

[plain cloud]

The target does not have same accent as me and as for datasets they were like 2 hours length

[fringe heron]

The converted audio sounds wierd on your voice right

You get accent of target?

[fringe heron]

Also you dont know the pretrained used?

[plain cloud]

Uh the what use of pretrain ?

[plain cloud]

No

Quite different actually

That’s why i wish to switch tts

[ocean maple]

whats the best for amd gpus

i got a 9060xt

[fringe heron]

So i guess you get the accent of the pretrained

[ocean maple]

best voice changer

[fringe heron]

When you train a model you use a starting point tyat already knows how to handle voice right

[hallow thistle]

Vonovox is exclusively for NVIDIA GPU. What will you use the voice changer for?

[ocean maple]

anything

i just wanna be able to use like all the models i find interesting

[plain cloud]

No my accent overrides the accent of the pertrained

[proud wing]

hey

can yall help me set up my voice changer

im intel

[void flume]

It is simply policy and part of the rules, something we basically have to ask: 'How are you planning on using the voice changer and those voice models?'
This rule is in place because people use the tool for notorious reasons, some even for hostile ones. We have rules against those. Feel free to glance at the rules in id:guide
If your plan is normal, it's fine.

[iron aurora]

Hey everyone, I'm searching for the right settings to get that AI 'Baby Voice' from Playboi Carti. I'm using the same model as BABY BOI, but when I upload my audio file, it sounds terrible. Any tips? I use Applio.
I use my MacBook M1 Max

[viral mason]

what kind of rp? the video is weird..

[dawn gazelle]

So I am now using the AI voice app called VoiceBox to make AI voices. Is it like with Weights and that I can make a model and use it safely just as long as I don't try to sell anything?

[viral mason]

I have never once seen or heard of that

use Applio to make voice models

[dawn gazelle]

Idk how to use that to make voice models.

[viral mason]

I've made a video guide on how to use it via Kaggle which is a browser site

are u wanting to make covers and stuff or realtime voice changing?

[dawn gazelle]

Kinda. Using it for content that I do not wanna sell.

[viral mason]

ah okok

[dawn gazelle]

Like having 2 different characters having a conversation.

[viral mason]

definelty Applio then

it can use already existing models for that

[dawn gazelle]

But can Applio make a voice model out of a character?

[viral mason]

it can yes, all you need is the audio of the model and then follow this

[dawn gazelle]

I have the audio of the character and/or model.

[viral mason]

https://mvsep.com/en/home
this part is for cleaning audio to get the best results

this is actually using applio in Kaggle https://www.kaggle.com/

[void flume]

I guess I'll remove the posts, I just wanted to let you know you don't need to wait for them basically

Universe is fine at least ;p

[viral mason]

oki

[tawdry shell]

• Goal (e.g., TTS, AI Covers, Roleplay): AI Cover
• Specific Issue: Trying to figure out why my AICoverGen is doing this weird thing where it'll error out in the middle of processing despite no errors on the cmd side of things. It starts doing "connection errored out" after the error trying to process the result.
• Full GPU Name: NVIDIA GeForce RTX 5070 Ti
• Operating System: Windows 11
• Tutorial Link used: https://www.youtube.com/watch?v=pdlhk4vVHQk

[quartz moon]

hi
i wanna use a voice changer but im not sure which one to choose(maybe if there is a cloud one for linux i know there is one for windows)

im on cachyos i have an rx 570 4gb and an i3-8100
uh my goal is just to play games with it

[void flume]

Hm, the goal part is basically to determine if someone is planning on abusing it for notorious reasons. For whatever reason, people are happy to admit to this when they do, which is the odd part. I guess no one reads the server's rules.

[quartz moon]

my goal is considred bad?

[void flume]

No

[quartz moon]

ah mb i missunderstood u

[void flume]

I mean the question may seem odd

There is a guide on AI-Hub Docs for linux users, but it is sorted per voice changer. So you just have to pick one.

[quartz moon]

thanks

[void flume]

If you're going for the cloud option, then any of them would do. Local, you can only pick one of the w okada ones due to having an AMD card.

[quartz moon]

hi
i ran this command
pw-loopback --playback-props='media.class=Audio/Source node.name=AI_Voice_Cable node.description="Virtual_Microphone"'

to make a virtual device but it doesnt appear in rvc(cloud)
im on linux cachyos
it does appear in discord

it also appears in other websites that use microphones so thats weird

it appears in input but not output?(even the monitors appear in input but not output)

nvm i used some other command and now it works but whenever i press start server it says select an audio input device

even tho i did select one

[sacred mica]

hey I was wondering if I want to do TTS, should I go with Fish s2? Currently running a gguf version of it on a 2080 ti 11gb vRam; I don't mind spending money on runpod or smt to train models but I'm trying to accurately create TTS voices that have human emphasizes voice acting like almost
Its js getting the prompt right for it seems impossible is that the best way to goa bout doing it
also is there a reverse speech to text? that can tell me what tags a certain audio would fall for
so i can learn wat i need to do
tyty

[simple ore]

speech to text is usually somethin like Whisper model

for tts there are plenty, fast and decent - kokoro

if you need voice cloning there are other options

[sacred mica]

?

[simple ore]

google Whisper ASR

[sacred mica]

im trying to replicate a certain style w fish s2 model

so i was tryna see if

a stt model

could detect

the tags yk you can do

[balmy cradle]

If I wanted to use the ai model for a documentary/youtube channel sort of thing, which is the best to use? or just the best to use overall

[tawdry shell]

am wondering if the rtx 5070 is too new for most of the ai builds cuz rn am struggling to get the GPU to run a kernel for voice changing

[rustic pumice]

bruv most of the rvc models i want are from weights.gg

none of them are in this server

yes

[proven hill]

whats your goal

[cyan spear]

Im SO confused, im trying to get mmvc for my computer, I have an rtx 5070 so im trying to do the thing, everytime I extract the zip it says I

Ive tried extracting it to different places but it does the same thing or says

[valid arch]

I wanted to share my project in the project showcase. I know I'm just new here, but what do I need to do to get permission to share my project in the showcase forum?

[proven hill]

you need level 5

[valid arch]

I don't know how to do that?

[proven hill]

chatting

[valid arch]

Thanks. I think the template may not apply to my question either.

I’m not asking for technical support, troubleshooting, model setup help, GPU help, TTS, AI covers, roleplay assistance, or anything illegal/NSFW.

The project is a published Amazon maze book series created with a heavily AI-assisted human production pipeline: concepting, layout assistance, algorithmic maze generation, editing/proofing, cover/art workflow, and publishing prep.

🧘‍♂️

[proven hill]

what do you need it for? trolling? catfishing?

[hardy yew]

Smooth xd "if this purpose is against your rules, then I'm doing it for a different purpose"

https://tenor.com/view/tkt-smart-gif-20642718

Chill, I'm not even sharing my opinion whether that's OK or not, just pointing out it's quite funny

[broken urchin]

guys

im having some trouble

how can i make it so the virtual audio cable isnt heard through screenshare

whenever i share my screen on discord to someone and turn "Allow to hear sound"

[sand iris]

We still using RVC for voice conversion ? Nothing relevant appeared since ?

[broken urchin]

the Virtual Cable from the voice changer is heard through the desktop sounds

[broken urchin]

i dont think so

[proven hill]

yes

[sand iris]

damn

[shrewd whale]

so I got an RVC folder inmy pc but i forgot how to set it up

go-realtime-gui.bat or go-web.bat doesnt do anything

[tawdry shell]

hello! I tried installing uvr5 but it's been hanging every time I've tried to process anything on it(Like it would sit at 10% for 2hrs with no change while trying to harvest vocals). Any help would be appreciated!

GPU is rtx 5070ti, I don't plan on using the CPU to process anything.

[shrewd whale]

i reinstalled windows, i forgot how to set them up

[hallow thistle]

What is your PC GPU? What do you use the RVC or voice changer for? Because it looks like you're trying to run realtime mode of original RVC GUI one.

[shrewd whale]

yes, i have been to do that before, i just reinstalled windows, thats why i forgot

last used RVC in 2024 iirc

[shrewd whale]

i used to just click any of this, depending on which i want to do

[hallow thistle]

[shrewd whale]

i think i just need to run the python stuff which i have forgot

[shrewd whale]

can u help me?

[hardy yew]

you kinda got asked a couple questions and completely ignored all of them xd

[shrewd whale]

i even have voice models already installed in the logs and weights folder

[shrewd whale]

i already answered them

[shrewd whale]

which is this one

i use both before

i used to run them on 1060 3gb now im on 4060

[tawdry shell]

nvm got it, found a patch for the rtx 5000 series 😄

[shrewd whale]

thats good then!

[shrewd whale]

i have used the web gui and realtime before using 1060 3gb, now im on 4060 and just want to be reminded on how to set them up

• Goal (e.g., TTS, AI Covers, Roleplay): Realtime
• Specific Issue: Just forgot to set them up, I already have voice models on my logs and weight folder in RVC
• Full GPU Name: 4060
• Operating System: Win11
• Tutorial Link used: https://docs.google.com/document/d/1KKKE7hoyGXMw-Lg0JWx16R8xz3OfxADjwEYJTqzDO1k/edit?tab=t.0 or https://docs.google.com/document/d/1vBJIJfq_UzWhDTRtYL1EUlQQNY2NaFGANeuTjDUezgI/edit?tab=t.0#heading=h.e9y3sfgsycb8

Apologies, I'm an older user of RVC, last used in 2024, I have just forgot how to set them up considering I have reinstalled my windows and upgraded from Win10 to Win11, and haven't touched RVC again until now

[hallow thistle]

W-Okada and Vonovox are dedicated realtime voice changers, they work way better than whatever realtime mode within RVC GUIs. Applio RVC also has realtime mode but that's about it.

[shrewd whale]

Idk the RVC version that I have but I'm used to it, is there any way for me tohave it working like before?

I remembered I have copied some codes and run them using python I think before downloading the stuff for RVC

[hallow thistle]

Nah, you should go for Vonovox if you mean solely realtime. Mangio RVC is outdated; Applio RVC is better as non-reatime AI cover maker.

[shrewd whale]

Oh okay, will the same voice models that I have work for the ones you mentioned?

Some of my voice models are custom trained too

[shrewd whale]

What will u suggest if I do both realtime and the voice changer that isnt realtime?

[hallow thistle]

Mate.

[shrewd whale]

Applio?

[hallow thistle]

Let make things simple, ok? If your voice models are "RVC", then they should work either in Vonovox or Applio RVC anyway.

[shrewd whale]

My custom train vocals are trained using weight.gg back then

and the rest of my downloaded ones are downloaded from the old ai hub server and some from weight.gg too

[shrewd whale]

Is there any benefits if I use a dedicated realtime voice changer vs Applio that ALSO has realtime?

[proven hill]

applio realtime kinda sucjs

[shrewd whale]

So its better to use a dedicated voice changer for realtime and non realtime?

[hallow thistle]

I don't have time to discuss today. I'll go sleep, and you might like to read AI Hub's website for more information.

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[proven hill]

yea

[shrewd whale]

Okay, thanks!

W-Okada, Vonovox for realtime Applio RVC for not realtime?

[proven hill]

vonovox only for realtime and applio for non realtime

[shrewd whale]

So Vonovox is better than W-Okada then?

[proven hill]

yes if you have a nvidia gpu

whats youe goal?

[shrewd whale]

Realtime and the non realtime (voice to voice conversion)

[west valley]

Hi, where can I find the free TTS or voice cloning bot in this server to make a voice text-to-speech? Thanks!

[void flume]

She means like, how are you planning on using it?

[shrewd whale]

I dont understand the question, sorry

[west valley]

I want to upload a 14-second audio sample of a voice, and then type a text for the bot to speak it using that cloned voice (Text-to-Speech).

[shrewd whale]

Like is there any other way for me to use them besides realtime and voice conversion?

[west valley]

@Rumi Yes, exactly. I just want to clone a specific voice from a 14-second file and use Text-to-Speech. Is there a free bot inside the server that can do this now?

[shrewd whale]

beta 17 11 is the one to download as thats only from 3 months ago right? no need to download patch or fix?

[void flume]

One tool is designed for real time conversion. Another tool is designed for pre-recorded conversion.
How, as in, in what setting? For example, music production or for discord call voice conversion or, idk.

depeding on the usecase, you may need different tools, etc

[shrewd whale]

Ah okay, my bad, so ill be using realtime for calls, and the non realtime for music production

i will also use the virtual audio cable

[void flume]

You mean like a discord bot? No.
Actually most of the bots are a bit broken due to the server costs and such.

[shrewd whale]

Is it better to run the precompiled setup for nvidia gpu or the manual setup for vonovox?

[void flume]

precompiled is easier

[shrewd whale]

Just download the latest here? no fix or patch?

[west valley]

"Thanks for the honest answer, Rumi! I appreciate your time and help."

[void flume]

I am not too sure, since I never used it. The patches are likely for upgrading pre-existing installed builds though. Maybe someone who has more understanding of this should answer.

[shrewd whale]

Yeah considering the latest ones should have the fix or patch on them from the previous versions

[void flume]

Just wondering, do you plan on trolling on discord with it?
(lots of people come in here for help for that reason.)

[shrewd whale]

Nope, i dont even use discord to call

[void flume]

oh ok

yeah, sorry for asking. It's just, because of the widespread identity fraud, harassment, etc. It's basically against discord's guidelines, our server rules. I'm just hoping people use it for normal reasons, and not those kinds.

[shrewd whale]

I supposed I can now delete my old RVC folder, but make a backup for my existing voice models right?

and do i need to keep both the files in logs and weight folder of my RVC?

[void flume]

I never used vonovox, so idk.

[shrewd whale]

Okay, no worries! Ill just test them out

[void flume]

If you wait a little, someone else may be able to help out. (I should be checking the logs and such anyway)

[hardy yew]

yeah, just that zip, without any other files. That's the most recent one.
There's a bug with index files in it, though. If you want to use index, you need to fix it by running ./runtime/python.exe -m pip install faiss-cpu

[shrewd whale]

So I have run the start bat but it doenst do anything? Or amn I missing something?

I just downloaded the precompiled setup

[hardy yew]

probably need to wait a short moment at first startup

[shrewd whale]

Oh okay!

I dont need to install python or smth?

Ahj the guii appears now

Thanks guys!

https://tenor.com/view/thanks-thanks-alot-thanks-a-lot-thanks-so-much-thank-you-gif-18225743498217953407

[hardy yew]

everything's bundled with the app, so all good

[patent trellis]

AI HUB | Technical Support Desk

📋 Required Help Template

To receive assistance, you must provide your system details. Copy and paste the block below into your reply and fill it out.

⚠️ NO INFO = NO HELP

👉 Fill this form:

- Goal (e.g., TTS, AI Covers, Roleplay):
- Specific Issue:
- Full GPU Name:
- Operating System:
- Tutorial Link used:

📍 Quick Checklist Before Asking

• Check Docs: Many fixes are in the AI Hub Docs.
• Be Specific: Say "RTX 3060 12GB", not just "NVIDIA".
• English Only: Keep all discussions in English.

⚖️ Community Guidelines

• No assistance for NSFW/Porn or ANY Illegal Activities.
• Read the [Full Guidelines](#1402790586028789830 message).

Helpers are volunteers. Be patient and don't ask to ask.

[proven hill]

no i mean to troll? catfish?

[shrewd whale]

ah def no

[proven hill]

so what then

[shrewd whale]

i would just like to test things out in calling some friends

[proven hill]

ok

[shrewd whale]

how would in know what to use in the embedder

The models that im using was made back then

also it seems that the pitch is always on 12

by default

[hardy yew]

unless the model was made with spin (very rare case and definitely mentioned in model's description), contentvec is the default andd most common

and default pitch of 12 is just due to male->female conversion being the most common use case..

if you want a different pitch value, just change it

[shrewd whale]

Ariana Grande then lol

that's like one of the earliest models back then

weights.gg isnt up anymore, it keeps prompting me to downloa replay

[proven hill]

shut down

[shrewd whale]

so where do i find voice models besides here?

[proven hill]

no where

[viral mason]

This site, tho the models aren't always the best quality
https://voice-models.com

Only place I know of since weights died

[shrewd whale]

abt the epochs? which is better lower or higher number?

[viral mason]

Neither

That's just how many times the AI went over the audio until the model sounded good

It's all up to listening and testing when training

[gusty moth]

-colab

[patent trellis]

📒 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

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Tg-Develop Fork**

by Tg-Develop
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
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

[viral mason]

u lookin for anything in particular?

[proven hill]

this is just a rip of ai hub

[shrewd whale]

so some index files arent working? what do i do? just let the pth file do its thing?

warming up voice conversion takes awhile now?

i think my gui bugged out awhile ago i closed the app and then now its stuck on warming up

[viral mason]

Wdym some index files don't work?

[viral mason]

It's more of a better weights

[shrewd whale]

some of my voice models index file doesnt work

[viral mason]

That's weird

All index files are the same besides the information that they contain

Shouldn't work any differently

[tall bobcat]

how to clone my own voice

[viral mason]

use applio, I have made a tutorial on how to clean audio and how to use Applio on a web browser site called Kaggle

https://mvsep.com/en/home
this part is for cleaning audio to get the best results

this is actually using applio in Kaggle https://www.kaggle.com/

[proven hill]

whait what

[viral mason]

ye it just has a ton of voices and no pay to win stuff on it

[proven hill]

really?

[tall bobcat]

you are the goat

[viral mason]

though it has been fucked lately where it won't upload any of my models anymore automatically

and the quality of the outputs stinks

so there are major flaws

[proven hill]

how come?

[viral mason]

dunno, randomly stopped

[proven hill]

wow

[viral mason]

the discord server for it is very sad to look at, no helpers only one guy running it and he's super inactive

I dunno if he even runs the site tbh

[proven hill]

funny enough this is ai hub in my eyes

[viral mason]

at least we know how to help people make models and use the voice changer :D

we're not a total mess

[proven hill]

eh

[viral mason]

thx, if u have questions just ask me or any of the helpers

[shrewd whale]

Why some voices are hard to match, is it because of the mannerism of the voice or its just the pitch isnt matching (male to male)

[viral mason]

it depends, some models have very specific ways of speaking that u cannot just raise the pitch to match, some also are just for singing so use in realtime would depend on if u can sing good

[shrewd whale]

makes sense, also speaking voice is different from singing voice

[viral mason]

true yea

[nova isle]

Dumb question but is custom pretrained in applio, gone or

[viral mason]

nah it's still there

you may have glanced over it

[nova isle]

where huhu

im using applio

[viral mason]

one sec lemme start it up

[nova isle]

alr

[tacit ermine]

I created a web gui with kraggle and ngrok however when i upload my first model i get this error "No empty model slot available. Please clear a slot or manage existing ones. " What should i do?

[nova isle]

nvm I saw it

brah

[tacit ermine]

• roleplay
• I created a web gui with kraggle and ngrok however when i upload my first model i get this error "No empty model slot available. Please clear a slot or manage existing ones. " What should i do?
• gpu t4 x2
• os ngrok chose?
• https://docs.aihub.gg/realtime-voice-changer/cloud/tg-develops-w-okada-fork-cloud/ kraggle

[viral mason]

srry it took so long lol

[viral mason]

what are you wanting to do?

are u using applio? or maybe wokada tg fork?

such little info is confusing

[tacit ermine]

I cant upload screenshots but after I spun up the ngrok server i went to the url. I continued with the tut and clicked the plus on the left to import model and i entered the pth file in model file

But when i click save, it gives the error above "No empty model slot available. Please clear a slot or manage existing ones. "

[viral mason]

I'll assume you're using wokada tg fork then

what gpu does your pc have?

[tacit ermine]

yeah i am

[tacit ermine]

i'm using cloud

https://docs.aihub.gg/realtime-voice-changer/cloud/tg-develops-w-okada-fork-cloud/ the tut url

roleplay
I created a web gui with kraggle and ngrok however when i upload my first model i get this error "No empty model slot available. Please clear a slot or manage existing ones. " What should i do?
gpu t4 x2
os ngrok chose?
https://docs.aihub.gg/realtime-voice-changer/cloud/tg-develops-w-okada-fork-cloud/ kraggle

all the details

[viral mason]

yes but if you have a decent gpu it would be better to use Local instead

[tacit ermine]

my gpu is not decent lol

[viral mason]

I'm not familiar with the cloud versions of the voice changers btw so I can't really help here

<@&1159293204038955078> give this man some help

[nova isle]

how do I fix this

Pretrained model sample rate (40000 Hz) does not match dataset audio sample rate (32000 Hz).

[void flume]

What kind of roleplay? egirl trolling? Catfishing? Erotic Roleplay? or something else?

[nova isle]

My pretrained model is 32K bra

[viral mason]

well did u check the sample rate of your audio?

[nova isle]

yeah

[viral mason]

hmm

[nova isle]

its 32000 Hz

[viral mason]

and you selected 32000 in applio?

[nova isle]

yes

I used this custom pretrained

https://huggingface.co/SeoulStreamingStation/KLM49_HFG/resolve/main/G_KLM_HFG_32k.pth?download=true

[tacit ermine]

its for a gta server

[void flume]

oh ok, then it's fine. (sorry for the question -- just has to do with the rules)

[tacit ermine]

@void flume do you know what i can do for this

i wish i could upload screenshots but to describe it with text when i click upload model at the top right it says no empty model slot

[viral mason]

I've never used that pretrain so I'm not sure if it's any good but if it's 32k it shouldn't cause any issues when training a 32k model

[nova isle]

I used the pretrained the one you downloaded, and used 32K pretrained it keeps saying its 40K 😭 wth

[viral mason]

hmm

[dense drift]

How to make ai covers on mobile?

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[viral mason]

@analog obsidian u got a clue on this guy's problem?

[dense drift]

Idk if this works on mobile

[dense drift]

Guys pls tell me

[viral mason]

I have no idea how to do anything ai on mobile I'm sorry

[dense drift]

Oh ok

[proven hill]

using a website

[dense drift]

Which one

[proven hill]

kaggle applio

[dense drift]

Do you have link

[nova isle]

any help guys? My dataset and custom pretrained are 32K and in the interface I chose 32K too

[proven hill]

https://docs.aihub.gg/rvc/cloud/applio-cloud/

[proven hill]

you have to wait

[nova isle]

wait for?

[viral mason]

someone who knows how to fix your issue

the best bet is to make a post here https://discord.com/channels/1159260121998827560/1192011222023950368

[analog obsidian]

what pretrain are you using?

[nova isle]

Hifigan, contentvec:

32k G: https://huggingface.co/lyery/mode4/resolve/main/G_15.pth?download=true
32k D: https://huggingface.co/lyery/mode4/resolve/main/D_15.pth?download=true

[analog obsidian]

go to the custom pretrain tab and show me this

[dense drift]

How do i use model?

[proven hill]

follow the guide :)

[dense drift]

Doesnt explain

[proven hill]

search for a guide in the site i gave you

[nova isle]

oh wait wrong one

[proven hill]

you have to input the path not the link

[nova isle]

sorry

😭

[proven hill]

oh it was the download one lmao my bad

[analog obsidian]

select 32000 here

then click train and see what happens

[nova isle]

it works now

k ill sleep while training this bai

[proven hill]

not even a thanks

[nova isle]

sorry, thank u guys huhu

[dawn gazelle]

Well as I asked before. If I make an AI character voice model. And use it to make content without selling anything. Would I be safe?

[tacit ermine]

roleplay
I created a web gui with kraggle and ngrok however when i upload my first model i get this error "select audio input. " What should i do? I have everything setup for inputs
gpu t4 x2
os ngrok chose?
https://docs.aihub.gg/realtime-voice-changer/cloud/tg-develops-w-okada-fork-cloud/ kaggle
all the details

[dawn gazelle]

• Goal:TTS
• Specific Issue: Using character voice models. I wanna see if it's safe to make a model of a character and use it for content. I don't plan on selling anything.
• Full GPU Name: NVIDIA GeForce RTX 3050 Ti Laptop GPU
• Operating System: Windows 11
• Tutorial Link used:

[proven hill]

yes

[dawn gazelle]

You sure?

[proven hill]

i mean

you have bad intentione?

[dawn gazelle]

@viral mason Can you help me with this?

[dawn gazelle]

No. I wanna make content. But not sell anything with AI characters.

[proven hill]

then youre safe

use applio

[dawn gazelle]

Well i am using another thing for it. But if you say I'm safe. Okay then.

[proven hill]

what are you using?

[dawn gazelle]

Voicebox

[proven hill]

what the hell is that

[dawn gazelle]

It's a app tha lets you make AI character models. It has QwenTTS included, And I think it might be better than Applio

Sure it takes some space but still.

[proven hill]

oh anything is better than rvc for tts tbf since rvc is sts

[tacit ermine]

My cloud version is working the only issue is the virtual cable. I get nothering audio wise from it. When i switch output to speaker, i hear the audio. When i switch it to my virtual cable to use it, the audio dies. So its a virtual cable issue, does anyone have a fix

[proven hill]

vac lite is better

[unborn goblet]

Can I improve the results I get from AI by speaking first-order predicate calculus instead of natural language?

[proven hill]

what

-realtime

[patent trellis]

🔊 Realtime RVC

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

• Vonovox

A Realtime Voice Changer with similar performance to Wokada Tg-Develop Fork, with extra features, but it supports only Nvidia GPUs on Windows 10/11 unlike other options that have wider support. and without cloud options

• Local 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.

• Local Guide
• Cloud Guide(s)

• Applio Realtime

A Realtime Voice Changer with similar performance to Vonovox & Wokada Tg-Develop Fork, with extra features.

• Local Guide

• Wokada Deiteris Fork

Deiteris' fork (modified version) of wokada that doesn't get updates anymore.

• Local Guide
• Cloud Guide(s)

⛔ 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

[proven hill]

use tg developed okada

[unborn goblet]

Which part confused you?

[proven hill]

everything

https://docs.aihub.gg/realtime-voice-changer/local/tg-develops-w-okada-fork/

[unborn goblet]

Okay. Thanks for your help anyway. I'll wait for another responder

[unborn goblet]

You're not allowed to use AI to cause harm to your friends in the form of trolling

That's a harmful behavior to troll

Well we still have to obey ethical principles

But you can use it for having fun with your friends. Or joking to your friends

Is that what you need it for?

Okay. I'm not a specialist in that area. Is there anyone who can help Zack

[void flume]

I need to process my disappointment again.

[shrewd whale]

another question, what should i use for training voices?

[clever gull]

Hey everyone, I'm 15 and I’ve been building with local LLMs lately. I got frustrated with coding agents leaving messy // TODO stubs and structural duplicates, so I built an AST-based CLI that hooks into Ollama to auto-patch that debt. I'd love to get some feedback from people who know more about LLM-integration than I do.
​Here's the repo: https://github.com/zenapta/BloatHunter

[shrewd whale]

is GTX 1660 applicable for the older nvidia gpu under vonovox like the extra steps for gtx 10 series?

[nova isle]

uhm

[rotund dew]

Hello! When trying to do inference this happens:

Exception in callback _ProactorBasePipeTransport._call_connection_lost(None)
handle: <Handle _ProactorBasePipeTransport._call_connection_lost(None)>
Traceback (most recent call last):
File "C:\Applio\env\Lib\asyncio\events.py", line 88, in _run
self._context.run(self._callback, *self._args)
File "C:\Applio\env\Lib\asyncio\proactor_events.py", line 165, in _call_connection_lost
self._sock.shutdown(socket.SHUT_RDWR)
ConnectionResetError: [WinError 10054]

[rotund dew]

• Goal (e.g., TTS, AI Covers, Roleplay): AI Cover
• Specific Issue: WinError 10054
• Full GPU Name: RTX 4060
• Operating System: Windows 11
• Tutorial Link used: https://docs.applio.org/

[nocturne field]

Hi my gpu is 1660 super.
im using RVC Deitris fork
windows 11
Goal : roleplay

I'm getting my voices back like it just lags and next person can hear my audio but rigged/lagged/echoed in a robotic form...

Need help in it.

[proven hill]

use vonovox

[nocturne field]

whats vonovox

[proven hill]

a better voice changer for nvidia

[nocturne field]

no like theres a charac nezukos thats one

[nocturne field]

can u give the guide

[proven hill]

https://docs.aihub.gg/realtime-voice-changer/local/vonovox/

[void flume]

Oh ok

[proven hill]

no donations

what is your gpu?

whats your goal? catfish? trolling?

not allowed, we will not help, @void flume <@&1159293140440723499>

because its not allowed by the #📜│rules

for trans people and roleplay

@viral mason this convo might interest you

[viral mason]

<@&1159293140440723499>

keep an eye for anyone weird joining

[viral mason]

it does interest me, because that user is so kind to give the most obvious reason to ban immedietly the next person asking for the voice changer

[proven hill]

lmfao

[void flume]

I have role mentions enabled by the way

[proven hill]

perfect

[void flume]

thanks for the report

[proven hill]

no prob hun

[viral mason]

and then they kiss :D

[proven hill]

smooch

[brisk trail]

Goal : Roleplay
Specific Issue: Choppy, distorted and robotic voice

• Full GPU Name: RTX 4060 (8GB)
• Operating System: Windows 10
• Tutorial Link used: https://docs.aihub.gg/realtime-voice-changer/local/vonovox/

my audio keeps sounding choppy and distorted, and I'm pretty new to vonovox any help would be appreciated here's my settings my CPU is also Core i5 13400K
and idk what to do I want the most quality AND the least delay that I can get, what should I change?

[brisk trail]

here us how it sounds

[broken urchin]

you sound radio

yo

when will realtime voice changer become perfectly realistic with real emotions

i mean like laughing etc

i can't wait for that for real

[brisk trail]

not that soon probably

[hardy yew]

not exactly, there is some people that do tons of experiments and put in lots of effort to improve it further

obviously probably still far from any breakthroughs

[void flume]

I see

[void flume]

I have to ask due to policies and rules here, but what kind of roleplay?
Like trolling?, life-action roleplaying?, ERP?, Game-character immersion?, creation of a separate online identity?, that kind of thing.
( in case you wonder about that: see #📜│rules or discord's community guidelines. )

[broken urchin]

what

so it wons progress no more?

i don't believe you at all lmao

[void flume]

maybe I should have removed the post. According to capybar there is progress

[hardy yew]

I mean, i just clarified that there is some people that try to improve it. But RVC has been more or less the same for years now so it's not like any changes are likely to happen in nearest future

That's what I meant, sorry if the first comment was misleading

[brisk trail]

for HUNK roleplay in garry's mod Resident evil RP server

the character HUNK y'know

[void flume]

yeah

[brisk trail]

that one, i don't do erp at all so yeah 😃

[void flume]

sorry for the question. I'm glad it's something normal 🙂

[brisk trail]

that's completely fine friend! you're doing your job and I deeply respect that!

[void flume]

Yeah, idk how to resolve the choppiness, but Delay is mostly determed by Block Size, you want to make that as small as possible, but not too small- (you'll notice when your resource usage goes up too much or it starts to get truly choppy). You can increase the Extra (ai-hub generally recommends 2.0) for better quality. I'm not too sure on vonovox to be honest, but I'm sure someone with more knowledge on it may pop up online soon
Anyway also try to adjust the crossfade to be somewhere between 0.08 and 0.15 (higher is better quality wise)

The audio sounds more or less like it has been 'pushed up' to a limiter; which may be normal for the model, no clue.
the micro choppiness might be due to an aggressive noise filter

I recommend trying to see if it happens without any noise filter enabled, and also disable your noise gate.

[analog obsidian]

dr recommends using the default block size value of 0.3

[dim wharf]

yo i'm not much updated by how often okada devs update but anyone knows when is it gonna release support for rtx 50 series? mine is 5050

[hardy yew]

realtime or offline inference/training?
(anyway, in both cases RTX 50x0 are supported)

[dim wharf]

wait what? it didn't work for me though

[hardy yew]

realtime -> Vonovox/tg-develop w-okada fork
offline/training -> Applio

[dim wharf]

is there a guide how to install anywhere? i might have installed it the wrong way then

[hardy yew]

which one?

[dim wharf]

ill check rq

[hardy yew]

i mean, what purpose

[hardy yew]

.

[dim wharf]

prob realtime

[hardy yew]

Vonovox will be suitable then: https://huggingface.co/dr87/vonovox/resolve/main/Vonovox_beta_17_11.zip
(if you want to use index, it also needs to be patched with the following command: runtime\python.exe -m pip install faiss-cpu)

[dim wharf]

alr ty

[hardy yew]

just download, extract, then just run start.bat, no setup needed

[dim wharf]

tysm bro

[hardy yew]

(except for the index fix as mentioned above)

[dim wharf]

W support

[frigid ore]

Hey all, I just installed hermet with codex and it keeps missing the first byte? Claude was also giving me greif

Sorry, Hermes*

• Goal (e.g., TTS, AI Covers, Roleplay): I just wanna get Hermes working reliably on my T480
• Specific Issue: Keeps missing the first byte?
• Full GPU Name: Not relavent, It's the top model of Lenovo T480 Laptop
• Operating System: Ubuntu
• Tutorial Link used: Not relevent, haven't used one

File "/t...]
⚠️ No first byte from provider in 45s (codex stream, model: gpt-5.4). Reconnecting.
⚠️ API call failed (attempt 1/3): APIConnectionError
🔌 Provider: openai-codex Model: gpt-5.4
🌐 Endpoint: https://chatgpt.com/backend-api/codex
📝 Error: Connection error.
⏳ Retrying in 2.6s (attempt 1/3)...

[manic hearth]

hey can anyone help me pls

i got an ai conversion model

and it says

tuple index out of range

any idea on how to fix this

[tall bobcat]

is there a free software that can clean audio? bcs all these websites want subscriptions

[proven hill]

uvr5

but you need a good gpu

what gpu do you have

[proven hill]

what software

[manic hearth]

vocal essentials

it was the one quinton used

[proven hill]

is it rvc?

[manic hearth]

yea

[proven hill]

use applio

[manic hearth]

Every time I do the voices sound nothing like what they’re presented as

They sound opposite and very robotic / static

[proven hill]

in real time?

[manic hearth]

How do I setup real time

Never knew that was a thing

[proven hill]

whats your goal? catfish?

roleplay? whatever

[manic hearth]

Nah I’m tryna make songs for like Carti ai

[proven hill]

then you dont need realtime

[manic hearth]

How do I get it working then

Also preciate the help

I been stuck on figuring this out

[proven hill]

https://docs.aihub.gg/rvc/local/applio/

[manic hearth]

Ty

Goat

[latent kraken]

heyo,
you guys know any other ai voice websites that isnt weights.GG

because I lowkey am struggling

[grave light]

What are you trying to do? Podcasts or music or what

[proven hill]

what is your gpu?

[latent kraken]

music

[latent kraken]

how do I check that

i am on task manager

[proven hill]

performance

[latent kraken]

I have an NVIDIA GeForce GT 1030

and if apps don't work then are there any websites i can use that will allow voice cloning and isnt a paid membership because i am broke

[proven hill]

https://docs.aihub.gg/rvc/cloud/aicovermaker-cloud/

[proven hill]

https://docs.aihub.gg/rvc/cloud/applio-cloud/

[grave light]

https://aimusicgen.ai/

[latent kraken]

its asking me for a cloud service what do i say ;-;

[proven hill]

this is not what he asked

[proven hill]

use kaggle