✨│ai-help

[signal chasm]

But I can't see my audio input (mic)

its just blank

[viral mason]

https://tenor.com/view/bf-gf-girlfriend-boyfriend-fnf-gif-16822914008880281078

[median monolith]

theres no way you noticed it was a fnf related audio with just what I showed

[brittle wing]

Are these the newest graphs & how do I read them?

[median monolith]

im pretty sure i surpassed the overtraining, should I then press the stop training button?

[simple ore]

with 1 minute, probably should not go over 300-400 epochs

even less most likely

[median monolith]

ok so i will press it

oh no

theres a problem

the buttons seem to not be working for some reason

[simple ore]

stop it on kaggle side

[median monolith]

actually, it seems to have already stop on kaggle and tensorboard

really strange that it just visually stuck/keep going on applio

[simple ore]

1000 epochs lol

[median monolith]

yeah, i really wasnt sure what value i realistically should put, so i did what the docs asked

[median monolith]

would you believe if said i have like 3 lowest points in my graph?

should it be the lowest point visually or by the "value"?

[simple ore]

i dont think you can really use a chart for such model.. just test some epochs

[median monolith]

uh, sorry? i know its a really limiting dataset and all, but aint i supposed to seek for "the lowest point" and stuff? so should i go for the epochs with the lowest points and see wich one i prefer?

[analog obsidian]

lowest point info is outdated, in this type of ai the only way to know if your model is doing fine is hearing it

the graphs are more useful when training pretrains

[crude flame]

i mean it can show you a good area on where to start listening

[pure quiver]

hi i have a 9070 what should my settings be for voice changer

[viral mason]

I'm kinda disappointed and impressed
did they get lazy or forget?

no more samples

[oak plank]

Currently, there is a lots of traffic on the website, and that's why things are taking time, but it will be working fine soon

[viral mason]

I use these settings, see if they're good on your end

[pure quiver]

Mines amd however

[viral mason]

same settings probably work :p

[pure quiver]

Thxx

[pure quiver]

Hii everyone my audio cuts out a lot when im talking

voice

[brittle wing]

hello where can i find the onnx models

[viral mason]

they should be in the same selection menu

[brittle wing]

in find models there alt pth models

[viral mason]

?

I can help you later I have to sleep, ask a mod or helper for now

[peak stag]

do you know how to fix it not detect my mic at all

like when i talk it doesnt change my voice or anything

<@&1159293140440723499>

<@&1159293204038955078>

[viral mason]

I'm not sure, I'll be here when I wake up tho ^^

[peak stag]

i watched the duckus vid

[brittle wing]

send me link

[peak stag]

i figured it out i was on the wrong version

[craggy bough]

please dont mass ping

[peak stag]

sorry

[worthy otter]

I need help with my voice changer

everytime i talk, for example I say "hello"

its like "Hel-oo"

its weird

[worthy otter]

<@&1159293140440723499> <@&1159293204038955078>

what do you mena

I just sent this

I just need help with my voice changer

[knotty moth]

I mean don't do like this^

[worthy otter]

but now its not opening at all

alr

[peak stag]

its prob ur settings

bro my shit keeps cutting out before it was so good

[worthy otter]

I watched a vid on youtube

It was by delta

[knotty moth]

any old video tutorial won't work at all

[worthy otter]

do you know any updated tutorials?

[knotty moth]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

[knotty moth]

btw you typed faster than me

[worthy otter]

lol

is this alright?

and this aswell?

[knotty moth]

seems like not enough vram, what is your GPU?

[worthy otter]

rx550

[knotty moth]

that's too weak, at least 6 GB cards as bare minimum or 8 GB as recommended

so RX 580 might be bare minimum

[worthy otter]

I wont be able to use the voice changer?

Its alright if i cant

[knotty moth]

can use cpu mode though it will be slower

[worthy otter]

It wont open anymore ever since I switched to gpu mode

[knotty moth]

^close some other programs or reboot the system

[worthy otter]

alright

do you have some recommended settings I can probably use?

[knotty moth]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#settings

[icy ore]

Can anyone tell me where can I download Love and Deepspace AI voice model ?

[flint anvil]

what's an example of an extremely good model? ("good" being able to replicate target voice without sibilants with a decent range) like just one that is actually maxed out in quality for rvc's capabilities - can be a model of anyone, i just want to test an actual higher quality model

[hazy eagle]

the less latency? more better?

[low shard]

don't spam ping roles

plus moderators aren't always helpers too

[low shard]

You can search rvc ai voice models at:

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

if there isnt one, you can:

• make it yourself with our docs guides

[earnest musk]

:wave: @low shard, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[low shard]

I'm guessing you're talking about the realtime voice changer, wokada deiteris fork that you got from our written guides and not some old ass youtube tut right?

if so, yes

u can also elaborate more for better help

[low shard]

what's ur pc cpu?

[low shard]

you would need an upgrade btw

also, what do u want to do?

[worthy otter]

Have a girl voice

[worthy otter]

I7-3770

[low shard]

elaborate more

pre-recorded audios? realtime?

if realtime, wherer? discord vc? or in games?

[worthy otter]

Realtime in discord vc

And probably in game

[low shard]

ehhhhh you're cooked

3th gen

[low shard]

absolutely not your pc is extremely too weak

[worthy otter]

😭

[icy ore]

But I can’t find any of love and deepspace characters voice models in all of this website

[low shard]

you can either:

• buy a better pc
• use cloud (remote good pc) with limited time, and you need a good internet for this

what u want to do?

[low shard]

if there isn't any, you can make it yourself, we don't allow people to take models request

also remember to maybe use all types of names that the characters could be possible named as, and all the sites i told you

[worthy otter]

I might just get anew pc

For my birthday bc its next week

[low shard]

yeah, what build you gonna get btw?

so like i can tell you if its good enough

like if you upgrade to rx580 thats just bare minimum and will be shitty in games

[worthy otter]

Idk yet but probably a rtx 3060 and a i9-12100k

[low shard]

ohh happy early bday

[worthy otter]

Thanks

[low shard]

nice, just know you won't like play marvel rivals at max settings while using the voice changer

you will need to put the game settings to low so you can have as much low delay in the voice changer

[worthy otter]

I dont like playing games at max settings on any game

I like medium to low

So i should be good

[low shard]

alright then

well, goodluck

[worthy otter]

Thanks

[low shard]

come back here if u get a good pc or want to use cloud instead

dont use video tuts tho

they pretty old

[low shard]

@peak stag @brittle wing don't use video tuts, that one is over a year old

tell ur pc gpu and what u want to do

[icy ore]

Can you tell me how to make my own?

[low shard]

what's ur pc gpu?

[peak stag]

i have a gaming pc and 4060 rtx i want to use a girl voice changer and it sometimes becomes choppy

[low shard]

okay, what do you want to do?

alright you edited the message

[peak stag]

and people say im ai in like 2 seconds

yea sorry

[low shard]

you seem to want realtime in discord vc or games

[low shard]

you followed a yt tut right?

you shouldn't

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

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

• **RVC GUI Mainline Realtime**

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

[icy ore]

I don’t know . I use MacBook Air

[peak stag]

yea

[low shard]

uninstall everything

read up the 1st link

wokada deiteris fork is the best

[peak stag]

what does this mean

[knotty moth]

that's good enough, but for gaming and most regular task, you just need decent affordable cpu like i5 12600 or ryzen 5 5600

[low shard]

it's a bot message linking the written guides, read only the first one

there's no up to date video tutorial, so read carefully please

[low shard]

macbooks aren't that great for ai
RVC doesn't support training models on mac

the best thing is for you to either buy a good pc with a nvidia gpu, or use cloud (remote good pc) which is kinda harder to setup and have limited free time

which u wanna do?

[peak stag]

alright thanks ill do this later

[low shard]

yw and lmk, remember to uninstall all things you got off that yt tut first

vb audio cable too, it's not good for windows

[peak stag]

but this guide also says use virtual audio cable

[hazy eagle]

ye i was talking about fork vc

[low shard]

vb audio cable is just one type of virtual audio cable, virtual audio cable is a type of programs not just one
the guide tells you to use vac lite, which is a better VAC (Virtual Audio Cable) for windows

[icy ore]

Really? 😭 Thank you for answering my question

[low shard]

when i say uninstall everything from yt tut, i really mean everything

[low shard]

yeah, wanna try cloud?

[peak stag]

alright

[hazy eagle]

and i don't follow youtube tutorial i follow the guide

[low shard]

do u want to show a screenshot of it so i can check ur settings and give u advices

[icy ore]

No

[low shard]

alright then, i guess you will just hope someone makes that model one day then

just remember that requests aren't allowed here, so don't fall into paid commissions, we dont encourage those nor want you to get scammed

[icy ore]

Okay thanks a lot

[low shard]

yw

[knotty moth]

take a look at existing models in https://discord.com/channels/1159260121998827560/1175430844685484042 and weights.com
other than the Nvidia gpu spec requirement or the cloud solutions, you'd have to learn a lot to do model making

[low shard]

yeah it's not a 1 click button thingy to do

[icy ore]

I found only love and deepspace Caleb voice model .I want other characters of love and Deepspace too 😭

[knotty moth]

as said above, we dont support paid requests/commissions

[icy ore]

Okay thank you

[brittle wing]

For https://www.kaggle.com/code/suneku/voice-changer-public on the cloud, why does the guide reccomend tesla t4 x2 and not the p100? You can only use one of the tesla gpus

it seems like the t4 is closer to a 3050 while the p100 is closer to a 3070

[knotty moth]

P100 is a pascal GPU (same arch as GTX 10-series) that is less supported for AI workloads

[knotty moth]

the raw performance isn't the right way to compare in this case

[brittle wing]

A support thread would say otherwise

From two different people:
"P100- This is the go-to acceleration method for most use cases and notebooks. It is pretty useful for most purposes and is integrated with Tensorflow and PyTorch well but does not use distributed computing

T4- This is also a good GPU but this needs to be used with distributed computing, else one may not use this to the fullest extent."

"GPU T4 x2: The T4 GPU is designed for deep learning and machine learning workloads. It provides fast processing power and supports high-performance computing tasks.

GPU P100: The P100 GPU is also optimized for deep learning tasks and provides high-performance computing capabilities. It offers faster processing speeds and larger memory capacity compared to previous GPU models."

[knotty moth]

you're missing the point of tensor cores, which are included in V100, RTX 20-series, and newer generations

and most AI workloads benefit or may require tensor cores

in this case, T4 is turing arch which has tensor cores

[brittle wing]

i cant find anything about tensor cores in the p100, does that mean it lacks them?

[knotty moth]

Volta (V100) and Turing (RTX 20-series and T4) are the first to introduce tensor cores

[brittle wing]

Oh I see

Thanks for clarifying

[slate smelt]

im trying to train a model using codename's applio fork, but i keep getting this permission denied error:
PermissionError: [Errno 13] Permission denied: b'C:\\Users\\admin\\Desktop\\Codename-RVC-Fork-V3.1.6-rev2\\Codename-RVC-Fork-V3.1.6-rev2\\logs\\test\\eval\\events.out.tfevents.1750078534.HoneyBadgerG6X.21556.0'.
ive tried running as admin and turning off readonly and giving the folder full permissions, but to no avail, the file exists, but it thinks it doesnt. if anyone could help, that would be great and would be saving me hours of anger

[simple ore]

brol... like.. you need 50 more folders down the line

sure that will help

/s

it is not at 250 character limit for a path, but seriously

extract the thing into C:\rvc or d:\rvc

and use a normal user, not admin

[slate smelt]

yea that worked, its very early in the morning, i assumed it was that but it worked fine for one model and didnt the next, but thank you regardless lol

[lost warren]

Hello, someone can help me a little? what is the best rvc to use in online games? i have a RTX4090, i saw the Wokada Deiteris Fork peoples saying is the best, but doing a test with it, the Wokada use 90, 100% my GPU, its impossible play using this, or i can do something ???

[simple ore]

unfortunately there's not much can be done about gpu schduling between games and cuda compute

you can try switching this off

[hazy eagle]

hello can i have the guide for realtime voicechanger?

?

do you have a guide for this?

[knotty moth]

the gpu scheduling is needed for frame generation but ye it can somewhat cause performance issue

[lost warren]

i dont have performance issues if i just using the Wokada Fork, the real problem its because he use 90 100% of the GPU, and its impossible running this with the game, the game consume 60% of the GPU

[knotty moth]

turn up chunk if it is too low

[lost warren]

yea, i doing some tests here, if i use the low chunk i got instant voice with 40ms but my gpu use 100%

if i increase the ms to 120ms, i got 50% of GPU usage, but the voice its not so good like 40ms

😐

[knotty moth]

the stable one (less than 100% usage) is usually preferrable

[lost warren]

yea, ill test with the game now, GTAV more specific, about the audio: Client or Server, which is better?

[remote imp]

Anyone good with AI automation sites such as make.com etc

[knotty moth]

server mode with wasapi devices has less latency

[lost warren]

i using FlexASIO, its better???

[knotty moth]

ASIO is even better if it works on you

[lost warren]

yea, its working on FlexASIO format in cliente with SA 48000

[brittle wing]

is https://3d.hunyuan.tencent.com/ safe?

[simple ore]

safe for what?

it is a too to make 3d assets from ai generated images and texts..

good luck registering ther though

[brittle wing]

safe to download from and use the models without getting a virus?

[simple ore]

dunno, i dont have QQ account or chinese phone number to try

[gusty kelp]

Hey, I have no idea where to ask this, but I'm trying to download okada voice changer, I'm on the Github page and there is supposed to be a place with a grid of stuff to download and its just not there for me

(I might be extrermly dumb but I'm pretty sure i'm in the right place)

[simple ore]

what's your gpu?

[gusty kelp]

Uh what's that

[simple ore]

video card

[gusty kelp]

I have no idea XD

but its fine, I think I got it

[simple ore]

open this

[brittle wing]

does anyone know good anime lora with bold outlines like in anime panty & stocking, which is avaible on civitai?

[viral mason]

ew

[viral mason]

task manager is always a thing

[simple ore]

that too, but it seems we are dealing with a special case

[knotty moth]

get the one from here, not the og repo
https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

[gusty kelp]

no no its fine I got it dw

[viral mason]

wdym?

[viral mason]

omg repo :3

https://tenor.com/view/repo-game-indie-horror-my-reaction-to-that-information-gif-16093430096844533032

[low shard]

smh not that repo

[knotty moth]

what the sigma

[viral mason]

[simple ore]

"(I might be extrermly dumb but I'm pretty sure i'm in the right place)"

[knotty moth]

https://tenor.com/view/please-respect-the-sigma-please-respect-the-chika-chika-rizz-sigma-chika-rizz-chika-gif-14729089138670968928

[viral mason]

ohhh a special case

got it

[zinc fern]

how create rvc models? i can't find

[low shard]

You can search rvc ai voice models at:

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

if there isnt one, you can:

• make it yourself with our docs guides

[earnest musk]

:wave: @low shard, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[low shard]

did u first try searching in the sites i told u?

[zinc fern]

i want to create with my voice

[low shard]

what's ur pc gpu

[zinc fern]

ryzen 5 4600G
amd radeon rx 5700 xt

how i can make?

[viral mason]

just use Kaggle applio

[prime shell]

how i can download applio?

[simple ore]

what's your gpu?

[prime shell]

this one

[simple ore]

4GB?

[prime shell]

yes

[simple ore]

https://huggingface.co/IAHispano/Applio/tree/main/Compiled/Windows

you may be able to run inference on small files, but no training

[prime shell]

okay

where do I give it?

[median monolith]

I just wanted to share the results of different choosen epochs of my voice model, but im not sure where I can do that

also, if I post a model on #1175430844685484042 , and Im not sure wich epoch.pth to share, could I maybe share a drive link on the post to a folder with all the different .pth files I downloaded? (it would be 9 different pth files for one model).

[vagrant raven]

heyy i have a question about weight ai i might purchase there subscription have a couple of queries

can i train 2 models on a subscription?

and is there a hindi feature on tts?

[hazy eagle]

?

[simple ore]

https://rentry.co/forkvoicechangerguide

[median monolith]

you can make as many voice models on weights without the need of a subscription, and if you want to do a "premium training" (a better sounding model), you can get a free one by getting a 5 day streak on the site (making atleast one creation, like a cover, per day).

[nova juniper]

Guys when I Download a voice model which pitch should I use on it

it doesn't say which settings are best for each model when I download it from #1175430844685484042

[viral mason]

is it a male or female model, and is your voice high pitched or low pitched

change the pitch until it sounds like the model

[nova juniper]

Oh okay

Is there anything else that I need to change other then the pitch

my Voice is mid pitch not high not low its in the middle

Its Adele's model

[viral mason]

not really, just experiment with every model

most models are good at pitch 0

[nova juniper]

Perfect got it appreciate you 🙂

[viral mason]

np ^^

[silent iron]

I'm stuck on start_https

I cant get anything to load

<@&1159293204038955078>

[pastel oak]

Delete what you have

Whats your gpu

[silent iron]

NVIDIA

[pastel oak]

Full name

[silent iron]

how can i check

nvm

uhh

cant find it

[pastel oak]

Task manager

Performances

[silent iron]

NVIDIA Geforce RTX 4050 Laptop GPU

[pastel oak]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

Download virtual cable and the first nvidia link you see

the nvidia-b2332 one

And follow rest of the guide, write a message here if something doesnt work

Ignore mac, linux, online steps ofc

[silent iron]

cant find the nvidia file

you mind going on screenshare?

[pastel oak]

💀

[silent iron]

im a bot mate

[pastel oak]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#download-nvidia-on-windows

[silent iron]

ok

thanks for spoon feeding me im sorry lol

alr what do i do with the file now

oh

😭

i got it

downloading rn

it says all weights are loaded

nothing loaded

what do i do

hello?

[peak stag]

i did this and it takes me to a website not an app

[silent iron]

dud

u need to download it

[peak stag]

i did download it

and it just takes me to a website

[silent iron]

bruh im boutta miss the interview

cause this shit wont work

[peak stag]

what interview?

[silent iron]

i got some shit to attend

and i need ts to work

why cant they just make ts an app

wait do i have to use the voice changer in the website?

[brittle wing]

Why is the graph so weird?

[silent iron]

is this normal

[peak stag]

<@&1159293140440723499> can i have help deleting the voice changer please

like fully

[viral mason]

https://tenor.com/view/cat-meme-funny-editing-photoshop-gif-4010364909273162247

[brittle wing]

It's adorable

[silent iron]

guys

i need some help cause my shit is way too laggy

are these good settings?

[silent iron]

is there a way to lower the lag

i have 400 ping

4000 total

@viral mason

[viral mason]

?

[silent iron]

look at the ping

its super laggy

yo omfg ts is so hard to use

do i really have to use a web browser??

[viral mason]

onx is bad why are u using onx rmvpe

[silent iron]

i dont know

nobody helps me

im stuck

they tell me to read

like wtf??

im an idiot

[viral mason]

same lol

[silent iron]

you think you could help me set mine up?

ive been trying for about an hour

[silent iron]

pretty please?

[viral mason]

sure yea! my dms are open

[crude flame]

if you dont want to use the web browser you can download vonovox (nvidia only version of w-okada. you have a nvidia gpu so you are fine here) and use that instead

[silent iron]

im getting so many things told to me, but i dont know where to get any of this

[crude flame]

everything is in our docs https://docs.aihub.gg

[knotty moth]

you need #outdated-model-maker-role before posting models there

[vast elm]

https://tenor.com/view/cowl-cowls-gif-2563789475502710348

[vast elm]

delete the folder where you extracted the voice changer, thats pretty much it :/ ...

im not sure which installation of real time voice changer your talking about but pretty sure that it has the python and installed libraries within the folder so everything is in there, you dont have to delete anything else, just the folder of installation, couldnt be any simpler

[knotty moth]

• close programs and background processes that may degrade the performance
• switch to high performance instead of power saving profile (assuming you're not running on the laptop battery)
• use an external cooler in case it overheats and throttles

[median monolith]

yea, ik

when i have that, can i do the thing i said?

[alpine flare]

i have the same problem but still with it. and i dont understand why its lagging, even if GPU cant reach 10% of using

like, if the GPU was using 100% i would understand the lag

[crude flame]

no, post only one link with the best epoch in your opinion

[cyan ledge]

cooked settings?

[hearty basin]

Alguien me ayuda

[wise token]

anyone know how to fix constant voice cracking

[brisk trail]

Today morning I woke up to this error and I can't fix it

[wise token]

clcik reload

[brisk trail]

I did didn't fix anything

I even restarted my pc

To no avail

[median monolith]

(Sorry to interrup, but...) If I have a more of a trivial question, can I do it here?

[brisk trail]

I did it 7 times and the 8th time fixed it 😂

[atomic wind]

@unkempt turtle

[unkempt turtle]

how do i download the voice changer

[wise token]

anyone know how to fix constant voice cracking

[viral mason]

Probably should ask the question lol

[viral mason]

I can help you out in a couple hours probably

[viral mason]

Try using suppression

[pastel oak]

Its a local webui, its only to display everything. Its not a website. Read the guide it mentions that in faq

To delete it fully delete the folder thats all

[pastel oak]

Crackles?

[pastel oak]

Ask instead of asking to ask

[pastel oak]

Send screenshot of the full voice changer

[nova juniper]

Guys what is the best woman sounding model there is at the moment ?

[blissful escarp]

what is gemini doing

QApplication.setAttribute(Qt.ApplicationAttribute.AA_แล้วไงต่ออ่ะลScreenResolution)
QApplication.setAttribute(Qt.ApplicationAttribute.AA_EnableHighDpiScaling)

where did it get AA_แล้วไงต่ออ่ะลScreenResolution

[elder coral]

bruh pls fix mainline

on colab i can't use it it still keep saying this site can't be reached

[pastel oak]

Use kaggle
https://docs.aihub.gg/rvc/cloud/mainline-kaggle/

[elder coral]

how do i use kaggle

[digital cairn]

Hey guys, i'm testing out different models and I noticed that some models take higher performance hit (requiring bit higher delay) than others, am i wrong to assume that it is because of higher sample rate? and if it is the case can it somehow be reduced, so that i dont have to increase the delay

[knotty moth]

perhaps, or it takes a bit more memory usage

[digital cairn]

if you mean RAM then it's the same from the looks of it

[low shard]

I saw you got helped. Next time don't ping moderators, moderators dont have to be helpers

just wait for an helper next time

[viral mason]

I'm glad the ability to @ all helpers instead of a specific user is a thing

I'd be nervous to do it

[low shard]

this server is english only, speak in english and elaborate

[low shard]

what's ur pc gpu? which tutorial link did u use? be sure to not use video tutorials, they are old

[low shard]

you can either use this channel for casual questions or #1192011222023950368

[low shard]

elaborate, what's ur pc gpu, what do u want to do, what tutorial link did u use?

be sure to not use video tutorials

lol the tag

[low shard]

it won't be fixed, the creator of the colab is busy since months

Plus applio is more suggested

[digital perch]

hey why does the http file for the voice changer only work once, I cant open the menu again

[low shard]

it's another cloud service provider, the other helper literally linked you the guide to read

also, what's ur pc gpu and what do u want to do?

[low shard]

elaborate:

• ur pc gpu
• what do u want to do
• the tutorial link you're using

if you're using a youtube video tutorial, you're cooked, because those are old

[digital perch]

amd 6700xt
im trying to setup the voice changer
https://www.youtube.com/watch?v=SxdnGxicJOg

the tutorial is 2 weeks old?

[viral mason]

he's cooked bro

[low shard]

i just checked the tutorial, the version that he uses is over a year old...

[digital perch]

ah ok so i did something wrong

[low shard]

this is the same shit u find in 1 year old plus tutorial lmao

[low shard]

the thing you did wrong is using video tutorials

[low shard]

absolutely uninstall everything off that youtube tutorial

forget it even exisated

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

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

• **RVC GUI Mainline Realtime**

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

[low shard]

read up the 1st link

wokada deiteris fork is way better than the over 1 year old version of original wokada in that video tutorial

there's no up to date video tutorial

[digital perch]

ok thank you

[viral mason]

u can check ur gpu with task manager btw

ez

[low shard]

he already said his gpu

[viral mason]

o

👍

[low shard]

remember to uninstall everything before using the new written guide, meaning vb audio cable and the old original wokada

then read the guide and lmk

[elder coral]

why did you rejoin in 2024 did you just take a break?

[digital perch]

u da goat i got it working

[low shard]

want me to check your settings?

[digital perch]

i cant post images

[knotty moth]

!give-media-perms 30m @digital perch

[digital perch]

wait actually for input audio theres an error when i select my mic

[knotty moth]

see the console window

[digital perch]

this?

[digital perch]

says this when i try to put my mic on input

[low shard]

what browser are u using

[digital perch]

firefox

[low shard]

gpu: ur amd gpu
chunk: 100ms
extra: 2.7

[low shard]

try chrome or any other browser, be sure to give microphone perms

[digital perch]

ill try that now thanks

[digital perch]

this is why ur my goat, ai voice changer is working 100% no errors thank you bro

[low shard]

be sure to put
input: microphone
output: line 1
output: headphones (to hear urself, optional)

u can also optionally set force fp32 mode on in advanced settings for kinda better quality but kinda more delay

be sure in the discord vc or game setttings to put the input as the line 1

do u need any other thing like optionally find a way to lower a bit the delay, even tho its kinda more hard

or is it all fine?

[digital perch]

its good now

i jacked the extra all the way up my gpu can handle it

[viral mason]

don't use onnx

[digital perch]

which one should i use then?

[viral mason]

rmvpe works fine ^^

just not the onnx version of it

[unkempt vale]

Is Hina_Mod_AICoverGen a broken project at Google Colab? Does anybody know?

[knotty moth]

I suppose both should work for Nvidia gpus without issues

perhaps slight performance difference

[viral mason]

I heard onnx of any of the models is worse, honestly not sure tho as that was said a while ago and I forgot who told me that

[unkempt vale]

I need change voice in mp3 file with Hina mod for Colab. It worked perfectly 2 minths ago((

[viral mason]

???

what are you trying to do, explain in words that make sense please

I might be able to help

[unkempt vale]

Me?

[viral mason]

yes

[unkempt vale]

Ok

I have mp3 file with my voice in russian. I need to change this voice to another voice model. I have this ptn and index file. I used to Hina at Google Colab for this task constantly two months ago.

Aplio doen't match my expectation becuse of bad pronunciation of some ukrainian letters (sounds). Hina was perfect.

But now it doesn't work with mistakes,

Traceback (most recent call last):
File "/content/Hina_RVC/src/webui.py", line 8, in <module>
import gradio as gr
ModuleNotFoundError: No module named 'gradio'

OR

Timer: 00:09:36Traceback (most recent call last):
File "/content/HRVC/src/webui.py", line 10, in <module>
from main import song_cover_pipeline
File "/content/HRVC/src/main.py", line 22, in <module>
from rvc import Config, load_hubert, get_vc, rvc_infer
File "/content/HRVC/src/rvc.py", line 5, in <module>
from fairseq import checkpoint_utils
File "/usr/local/lib/python3.11/dist-packages/fairseq/init.py", line 20, in <module>
from fairseq.distributed import utils as distributed_utils
File "/usr/local/lib/python3.11/dist-packages/fairseq/distributed/init.py", line 7, in <module>
from .fully_sharded_data_parallel import (
File "/usr/local/lib/python3.11/dist-packages/fairseq/distributed/fully_sharded_data_parallel.py", line 10, in <module>
from fairseq.dataclass.configs import DistributedTrainingConfig
File "/usr/local/lib/python3.11/dist-packages/fairseq/dataclass/init.py", line 6, in <module>
from .configs import FairseqDataclass
File "/usr/local/lib/python3.11/dist-packages/fairseq/dataclass/configs.py", line 1104, in <module>
@dataclass
^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 1232, in dataclass
return wrap(cls)
^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 1222, in wrap
return _process_class(cls, init, repr, eq, order, unsafe_hash,
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 958, in _process_class
cls_fields.append(_get_field(cls, name, type, kw_only))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.11/dataclasses.py", line 815, in _get_field
raise ValueError(f'mutable default {type(f.default)} for field '
ValueError: mutable default <class 'fairseq.dataclass.configs.CommonConfig'> for field common is not allowed: use default_factory
Timer: 00:09:37

[viral mason]

it may not work because most google collab stuff is dying

I really only know of weights.gg, applio, and mainline that could be used t do what you want

[unkempt vale]

but Applia works right know. What's the difference,

[viral mason]

it's up to date with its code, Hina might not be

[unkempt vale]

It doen't pronounce correctly some sounds,,

Oh it's a pity

[viral mason]

rvc isn't ever perfect know

people mess up words all the time no matter the language even when speaking normally

[unkempt vale]

Last time Hina helped me to fix this issue

[viral mason]

so it's kinda natural

[unkempt vale]

No

It's not an explanation

[viral mason]

are they in this server?

[unkempt vale]

Who are they,

,

?

[viral mason]

is Hina in this server

[unkempt vale]

Sure

[viral mason]

it's a yes or no, do you know if they are???

[unkempt vale]

I'm not sure if it was Hina. But man or woman helped me

I'm sorry I need a specialist

[viral mason]

you can probably get better help from one of the helpers than me

there's two online rn, see if they're available

[unkempt vale]

Yes. You are right

thanks

[knotty moth]

it may be contentvec limitation, or depend on the language, or try articulating better

[nocturne sinew]

Does anyone know how to set up something like AI-Dungeon but on my computer?

[neat meadow]

where is the download link??

[digital void]

Is it possible to do a mommy voice without using Okada Voice Changer?

[crude flame]

yes, its called doing vocal training

[low shard]

using more than 2.7 can cause cause issue of cutting

[low shard]

don't say that, he's on amd, onnx is specifically for non nvidia gpus

[digital void]

BRUHHHH

[low shard]

rmvpe_onnx, dw

[low shard]

it's broken and replaced, it will NEVER come back,check #📰│dev-updates message

[low shard]

#📰│dev-updates message

@unkempt vale what's ur pc gpu? what do u want to do?

[low shard]

Hina is busy since months

[low shard]

in the guide, you have to read it

[digital void]

Is it possible to do a mommy voice in weights?

[neat meadow]

idk where is download link

[crude flame]

nope, they dont offer realtime voice changing

but for normal inference you can

[digital void]

What would be the best pitch for me to put in?

[crude flame]

for male +12-ish for female 0

[digital void]

Thanks bro

[crude flame]

np brosidion

[neat meadow]

December 7th 2024 is the latest version?

[brittle pagoda]

hi guys im tryna make a song w a female voice but because im a male, I need to transpose the voice up by 8 for it to be in her range so she can actually have energy while singing but that puts her vocals out of key with the beat, is there any way around this? I cant pitch up the beat cause it sounds like some garbage

how do guys make songs w female ai models in their range?

[low shard]

what's ur pc gpu? what do u want to do?

[low shard]

try just playing with the pitch

[viral mason]

I wasn't ever told that lol

[neat meadow]

download this?

[low shard]

i asked you, what's your pc gpu? what do you want to do?

[neat meadow]

my gpu is nvidia rtx3060

realtime voic3echanger

[low shard]

for calls right?

[low shard]

then yeah, this is the latest wokada deiteris fork for nvidia gpus

be sure to read all of https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

[low shard]

i think its said in https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/ along with convert to onnx

onnx is slightly worse in quality iirc but has better performance for amd

[neat meadow]

I downloaded it, just run the file?

[low shard]

no, be sure to read the whole guide

you need VAC lite too, it's in the virtual audio cable part of the guide

please dont skip steps, else it wont work

[neat meadow]

right this?

[low shard]

no, that's for nvidia rtx 5000 users

you're on the 30 serie, not 50

[neat meadow]

What should I do then?

[neat meadow]

i downloaded this

[low shard]

https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#virtual-audio-cable

you can skip mac/amd/intel/rtx 50 serie steps, but you need to follow the rest of the guide

if you dont read it, it wont work ofc

[swift thunder]

Does anyone know an alternative to Weights images? He's hungry and overweight.

[neat meadow]

Do I just run MMVCServerSIO and set it up? Is there anything I should do when installing?

[low shard]

what?

[low shard]

Yeah, you should read the guide and do the Virtual Audio Cable part

it isn't a 1 click program

[neat meadow]

I know how to set setting

I was wondering if I could just set it up and run it

[low shard]

no, you need to do the virtual audio cable part

A VAC (Virtual Audio Cable) makes a fake audio device, used to re-route the audio of different programs

In Wokada context, it's used to get the output of wokada as the input in other programs

[swift thunder]

On weights.gg, you have to wait 1000 people to be able to take an image or train a model, I've still been waiting for 3 days

[low shard]

if you dont do it, it wont work in other programs

[neat meadow]

yes i know i have vb audio cable

[neat meadow]

and i use audio interface do i need to use server?? I know the client is better

[brisk trail]

I had fork by deitris, and I've been working with it for months, it just happened today morning

[hollow sail]

anyone know GOOD ai voice changers

[low shard]

no, vb audio cable is bad, it may give random issues on windows as users reported

uninstall it, forget everything from wokada youtube video tutorials

get vac lite from the guide, you need to read the guide to use the program

[low shard]

is it wokada deiteris fork b2332? also what browser do u use? what's ur pc gpu? does the issue still persist?

[low shard]

elaborate:

• ur pc gpu
• what u want to do? realtime for calls? pre-recorded?

[hollow sail]

rtx 4060 ti
for a stream and to troll people in game i want it to sound as realistic as possible yk i tried odaka but im having issues with it

[median monolith]

I just asked to question because I needed to be sure I could make a less of technical question here, before I do that and someone goes like "yOu AreNt SuPpOsed To bE asKiNg thIS HeRe" or something like that.

[pastel oak]

All good, you can ask anything here

[low shard]

yk i tried odaka but im having issues with it
youtube tutorials are old

forget everything you get about it in a video tutorial

also, no there is no such thing as completely perfect way

[hollow sail]

can you help me its really choppy and i cant seem to put hte playback

[low shard]

there's limits, ai can't laugh well for example

[hollow sail]

like i dont hear the voice coming through on the other side

[low shard]

delete everything you get off videos

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

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

• **RVC GUI Mainline Realtime**

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

[low shard]

read the 1st link, wokada deiteris fork

[hollow sail]

alright but cna you also tell me how to output hte audio

[low shard]

wdym output the audio

[hollow sail]

like i mean i dont hear it at all on the other end

say i want to use the vc on discord right how do i make it go through discord

[low shard]

you need to uninstall everything off youtube

then read the guide

the guide explains it

[sacred hill]

about voice models, whare can i find beatrice?

[edgy topaz]

Do I need a Laptop or a desktop for voice changer?

[brisk trail]

Yes it's that version and the problem still presists, and my pc gpu is: RTX 4060 (8GB)

I use Google Chrome

[sonic night]

RTX 3050 (sorry for answering too late)

[brittle wing]

Why is training those RVC voice models so slow

[digital perch]

i had no issues when i used it that high

[viral mason]

nice jpgs

https://tenor.com/view/epic-embed-fail-ryan-gosling-cereal-embed-failure-laugh-at-this-user-gif-20627924

[cold belfry]

What would yall say is the best realtime voice changer

Also take into account I have a RTX 2070 so no very high end ones.

[low shard]

Beatrice v2 models are Faster and more Lightweight than RVC v2 models, but have Less Quality and there are Less Publicly Available models

TL;DR: you shouldn't use Beatrice models, DO NOT USE VIDEO TUTORIALS FOR RVC AND WOKADA

They are old

What's your PC GPU and what do you want to do

[low shard]

Realtime? Of course, that's the bare minimum to even run cloud, a phone doesn't have a virtual audio cable (VAC) which is needed for it to work in other programs

Or do you mean pre-recorded?

[low shard]

What did you do specifically that caused this issue? Also try other browsers like firefox

[edgy topaz]

Sure I think realtime

[low shard]

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

You can:

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

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

[low shard]

Alright, you got at least any type of PC? Would be the best to have a good one

[low shard]

AI is an intensive task, don't expect it to be fast

[low shard]

It might on some models, but sure you can do whatever

[edgy topaz]

I want to my a PC but I don't know If I should get a laptop or a desktop

[brittle wing]

Hmm so what's the matter it takes three days on average to train such models a waste of time

Also why does the chart look sow weird

[low shard]

WOKADA deiteris fork

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

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

• **RVC GUI Mainline Realtime**

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

[low shard]

1st link

[crude flame]

3 days?

[low shard]

read it

[brittle wing]

Yes

[low shard]

Depends on the specs, generally speaking a desktop is better if it has a good GPU

[crude flame]

rvc models take max like overnight

[low shard]

3 days? What's your PC GPU

[crude flame]

some can be finished in like 5 hours depending on gpu

[brittle wing]

Hmm yes

[brittle wing]

Colab

[edgy topaz]

What GPU is the best so I can get it?

[cold belfry]

Fork? I have the natrual WOKADA onyxgpu-cuda where can I find fork?

[low shard]

What's your budget?

[brittle wing]

Hmm so I'm getting bad GPUs in Colab, got it

[crude flame]

ah, that makes sense

[low shard]

I gave you the link, it's the 1st, read it

[cold belfry]

ok thx

[low shard]

Are you sure? Colab has 4 hoyrs max of GPU for free daily

[crude flame]

a amd 6700xt gpu is faster than colab

[edgy topaz]

I don't have one yet

[low shard]

Are you using the free tier? What GPU are you using

[cold belfry]

Also, will a 2070 be able to run decently smooth on it?

[brittle wing]

Tesla T4

[low shard]

Yeah, just set game graphics to the lowest if u gonna game

[cold belfry]

Mmk, appreciate the help! Have a good day!!!

[low shard]

Then it's hard to suggest to you, I can just tell you to get a pc with an RX 5090 (which alone that component costs over 2k USD) if you want the best lol 😭

You need a budget because the top of the top is expensive yk

[crude flame]

i mean a a100 would be better but yk

[low shard]

Oh yeah of course you can also suggest ai specified GPUs that cost over 10k dollars

This is why you need a budget to make a PC, there's no other limit than your money

[edgy topaz]

Can't I get one for like $150?

[low shard]

An entire PC for 150$?

[edgy topaz]

Yea

[brisk trail]

Nothing actually it just happened, I woke up to the app being no avail

[low shard]

Hell no, generally speaking it's probably gonna have low end specs, prob gonna be some crappy shit, at max it could run cloud (meaning using a cloud computing site for using a remote good PC with limited free time)

Are you gonna buy a PC just for realtime voice changing?

[edgy topaz]

So $2,000?

[edgy topaz]

Nopee there's more therefore

[low shard]

Like?

Depends also on what you wanted to do

[edgy topaz]

Yea I also what like real-time video

[edgy topaz]

What about Dell Optiplex 9020 Desktop Computer PC

[rare lintel]

Hi which version should i get? https://huggingface.co/wok000/vcclient000/tree/main

I have a rtx 4060

[untold rapids]

where's RVC v2

[knotty moth]

that sounds nothing more than a raspberry pi retro rig or a low budget phone

[edgy topaz]

Is that an insult orrr

[knotty moth]

unless you meant 1.5k budget

[edgy topaz]

Wait is budget money?

[brittle wing]

Why is the graph weird?

[analog obsidian]

did you stop the training and then resume later?

[alpine flare]

can someone help me plz?? why my Res is around 40000

[brittle wing]

Yes several times

It is cause i ran out of gpu

Also someone said this graph is outdated

[analog obsidian]

two possibilites here:
you resumed the training with the wrong batch size, let's say you started the training with batch size 8 but then resumed with batch 16
or
the graph bugged because you stopped training before completing an epoch, in this case, it is a harmless visual bug (technically it is not a bug)

and yes that graph is outdated

[brittle wing]

I always trained w 8

[brittle wing]

I know & how do I seriously identify the real lowest point?

[analog obsidian]

the lowest point of the graph is not your best epoch, that is outdated information

[brittle wing]

Really then how do I know?

[analog obsidian]

you save every 10 or every 5 epochs then hear them until you find the best sounding one

[brittle wing]

So basically i can train as many as i want now

Im using a pretrain

[analog obsidian]

well it's more complex than that but to make them simple, train 200 epochs and maybe save every 10 and hear them
in my case, today i trained a model for 100 epochs because the dataset was big and i know the model will converge before 100 epochs

[brittle wing]

Does the pretrain guarantee that the model is gonna finish faster?

[analog obsidian]

yes, always train with a pretrain, if you dont use one the model will sound like a robot

[brittle wing]

So nowadays its recommended

[analog obsidian]

original pretrain

the best we have

[crude flame]

I mean it can give lazy people an area to listen to epochs in

[brittle wing]

Yes but im using one for kor. Langauge

Original is mostly english

[analog obsidian]

the model learns the new phonemes regardless of the language of the pretrain

it learns them in like 5 epochs

[brittle wing]

Yes but I have preferences

[analog obsidian]

also it's a known fact klm does not produce good speech results

but if klm makes u happy go for it

[brittle wing]

How do I calcolate the epoch per dataset minutes

Is there a dataset length and epoch correlation

[analog obsidian]

yeah

[brittle wing]

So 30 minutes.

[analog obsidian]

hmm maybe try 200 epochs and save every 10

[brittle wing]

Will it be 500 epochs

[analog obsidian]

no thats too much

[brittle wing]

Thats what im doin but im at 300

[analog obsidian]

sure thats fine

you know the model is overtrained when after a certain point the model starts to sound very robotic

[brittle wing]

How is it too much?Explain for 30 mins

450?

[latent kettle]

You can check tensor board

To prevent overfitting

[brittle wing]

The graph is messy and outdated

[latent kettle]

What do you mean by outdated

[analog obsidian]

there is not too much info in your dataset and you're using a pretrain model, which already contains a lot of information, you dont want to replace the pretrain knowledge with the knowledge of your 30 min dataset, the pretrain is trained with over 50 hours worth of data, you dont want to lose that

by training 500 epochs you're telling the model to forget the pretrain

thats bad

[brittle wing]

Okay then tell me how many epiche for 30 minutes.

[analog obsidian]

200 and save every 10

batch 8

[brittle wing]

I have already completed 200

Epochs

[latent kettle]

Then stop and do inference

[analog obsidian]

stop and try the model

[brittle wing]

My saving frequency is every 10 batch size 8

[brittle wing]

Whybso 200 is enough???,?

Is that a joke

[analog obsidian]

no

[brittle wing]

200 is enough?

[analog obsidian]

yes

[latent kettle]

Most of the models are trained on 200-250 epochs

[brittle wing]

I have the 200 epoch weights saved

[latent kettle]

Also you can resume training dw

[brittle wing]

Yes but my 30 minutes dataset is clean and high quality

[brittle wing]

Im still training going for 350

[analog obsidian]

bro, it's simple, if you wanna train a super high amount of epochs, do it

its your model

[latent kettle]

Not a problem. Just try your model

[brittle wing]

Okay and

[latent kettle]

It will ruin your model

[brittle wing]

But you say 200 is enough do you mean thats cause of the pretrain

[latent kettle]

More epochs is not = to good quality

[brittle wing]

Hmm sometimes it is

[analog obsidian]

well i already explained everything, you can ignore that and train the way you like, it's fine

[brittle wing]

Is it cause of the pretrain

[latent kettle]

Everyone use pre trained to make the training faster. Training a model without pre trained will took days

[brittle wing]

Cause what i noticed with this pretrain it starts sondino good at 100 already

[brittle wing]

I remember

It took longer before

[latent kettle]

Yeah you have 2 ways

1. Trust us
2. Look at tensor board

[analog obsidian]

she's using mainline graphs so they wont help

[brittle wing]

She*

[analog obsidian]

sorry, edited my message

[latent kettle]

So I suggest you to stop training and try the model

[brittle wing]

You say 200 is enough cause i used a big pretrain?

[latent kettle]

Well you can try and find it out

[brittle wing]

Uh ok

[knotty moth]

you have resumed training with different batch size than before

[brittle wing]

No i didnt

It was always 8

[knotty moth]

another possibility is you lost a tfevents file for the second training session

[brittle wing]

Could bey

[knotty moth]

assuming you have really run three/more training sessions

[simple ore]

@brittle wing how many steps per epoch you got?

[brittle wing]

How do I understand that.

[brittle wing]

Formula

[simple ore]

4345 steps

[brittle wing]

And

[simple ore]

tell me the number

[simple ore]

okay, so 90steps per epoch, okay

[brittle wing]

And?

Whats the lowest point

My graph is messy

[simple ore]

just checking that you're not doing something silly

show avg_50 charts

[brittle wing]

WOW WAIT

[brittle wing]

How do I read them

[simple ore]

just start checking models saved at 200 epoch onward

if you actually did save them

[brittle wing]

Weight?

Yes

I have them

[brittle wing]

By listening?

[simple ore]

what's after this ?

by running inference and listening

[brittle wing]

I know

[simple ore]

can you like not do it from a phone with tiny screen?

did it save the actual model, not g/d

[brittle wing]

I font have another device

[brittle wing]

The pth yes

[simple ore]

okay then

[brittle wing]

They said 450 epochs is too much for 30 minutes

Is it true

[simple ore]

you can only find out by running a bunch of tests and listening which one is good and which one is bad

[brittle wing]

Okay.

[simple ore]

450e seems unnecessarily too much

[brittle wing]

What if it sound normal at 350

[brittle wing]

How do you know?

Im all ears

Some ppl train 16 mins on 500

[simple ore]

and get garbage

I ran 5 min set to 1700e

so what?

[brittle wing]

Too much

[simple ore]

how do you know?

[brittle wing]

Cause it sounds logically true

Is 250-300-350 okay for 30 minutes

[simple ore]

You. Can. Only. Know. By. Listening. The. Results.

[brittle wing]

Okay

[edgy topaz]

My GPU is 10000000000 /j

[latent kettle]

Still seeking for help. We already told you

[flint anvil]

i dmed emojikage and got no response, is there anyone else i can contact to help? i want to edit the source code and run the gpu-enabled deiteris fork locally, but due to a lot of technical reasons that ppl might be able to help with, i cant

i want to continue the deiteris fork project but forking the repo doesnt let you build gpu compatible versions for some reason

[tight ether]

I'm not sure that's entirely true... where is the data showing that it performs poorly in speech tasks?

[faint mason]

Are there any models from Google that are faster and smarter than gpt 4.1 mini?

[knotty moth]

you don't

if you want to use for gaming, you'd want to upgrade your gpu vram

even many modern games demand more than 8 GB vram

[tight ether]

If what you're saying is true, then any models using the og pretrained model shouldn't sound robotic. but clearly, most of the models currently uploaded don't meet that standard, which means they aren't uniformly high quality. that suggests the bigger issue lies in the dataset you're using. I just don't want the blame to keep being shifted entirely onto the pretrained model.

even if you train a model without using any pretrained models at all, the results will still vary depending on your dataset. sound is extremely subjective what sounds good or bad can differ from person to person. so there's no absolute answer. on top of that, everyone uses different hardware, which makes it even harder to define what something like "sounds like."

all of the models I've created were trained using KLM. If the pretrained model were solely responsible for robotic artifacts, then all of my models should consistently show the same issues. but unfortunately, the differences always depend on the dataset I'm training with. even as the person who created KLM, I can't definitively say where the problem lies. so to see people confidently claiming what's "wrong" as if it's a fact. it feels a bit frustrating.

[jaunty star]

the raspiness of the voice changer keeps coming back every now and then, what's the cause of this? do i need to find a better voice model or do i reinstall VAC lite?

[simple ore]

whym? it runs locally

[golden walrus]

Guys, can i ask if there is a big different between 32k and 48k

I mean the quality

[simple ore]

for speech use 32k

[golden walrus]

Ah okay, because i saw on vonovox, it has 48k option only

I have no idea if i use 32k, it gonna be worse or not

Thank you. Let me try training in 32k

[simple ore]

I'm not sure whether it re-samples audio or it is just the audio output setting

you can use any model, I think

[golden walrus]

Ah, i forgot. Can i ask if the pre-train you made is good for asian languages

[knotty moth]

for voice changer and most practical use cases that don't demand audio quality/fidelity, 32k is enough

for model making, keep in mind of this

https://cdn.discordapp.com/attachments/1159290139609137264/1344111265253032017/image.png?ex=685360aa&is=68520f2a&hm=63b21941cc950f43938147d9d8496b51dce1d882d5e430c28de06f965ebc74ff&

[golden walrus]

Ahhhhhh, i get it

So no need 48k

[simple ore]

normal speech recordings do not go over 16k usually.. sometimes even 12k

thus x2 SR for the model - 32k and 24k

[low shard]

I tried googling it's specs and it seems to not have an integrated GPU, so no

[low shard]

None of this, this is original WOKADA

[low shard]

You want a realtime voice changer for calls?

[low shard]

What's your PC GPU? What do you want to do?

[low shard]

elaborate:

• your PC GPU
• what you want to do
• the issue
• what tutorial link are you using

[low shard]

Show your WOKADA settings

[knotty moth]

https://discord.com/channels/1159260121998827560/1339155300720054316

there is also a version for spin embedder

but not so recommended unless you're interested on testing and contributing

[tidal mango]

im trying to find the name for this ai txt to speech model name can someone dm me so i can dm them the mp3 ??

[tidal mango]

its a very common one

[brittle wing]

Sound overtrained at 350, sound i go for 300

[brittle wing]

Yes tell em

[brittle wing]

Okay and it was right

Like 350 epichs sounded robotic as hell

Now 300 are okay but moderate robotic

So 250 maybe will do

[brittle wing]

Can i DM simeone so they tell me the best checkpoint?

[golden walrus]

Ah, i tried using this. But the outcome is a bit, not as good

Since i heard breaking and cracking, a bit robotic too

I will try KLM 6.3 for spin

[knotty moth]

if you don't find it good using default pretrain as well, more likely the dataset issue

[golden walrus]

I mean, default is fine, just default is mainly English, so when it comes to Chinese or Vietnamese. It's a bit

Mispronounce ?

Like can't spell certain words

With spin, that issue has decreased

But still exist

[simple ore]

it hasn't been trained on chinese or vietnamese audio

@golden walrus

[brittle wing]

How come the 200 epochs model checkpoint sounds the best 😭

Just asking

Should I go for that one...

[dry meadow]

do you guys know where to get voice models for german females?

[analog obsidian]

they sound robotic because ppl overtrain their models training it for like 500 epochs

[analog obsidian]

i can run another test run today
but i said that because ive seen a couple of people claiming klm made their speech models sound a bit metallic
as for me i also remember experiencing that, but i can confirm that today

[brittle wing]

Look at what I asked pls

[analog obsidian]

no

[brittle wing]

How come my 200 epochs checkpoint sounds the best

[simple ore]

we told you and you did not listen

[analog obsidian]

i literally told you yesterday why training 500e is a bad idea

today you're seeing what happens when u overtrain the model and are surprised

[brittle wing]

But the 200 epochs chkpt is the best.

[brittle wing]

Well after listening yesss also it was today morning in my case...

Timezone

[tight ether]

ah I see I see.

[brittle wing]

I checked the checkpoints below 350 & I understand

[analog obsidian]

yup gonna ping you after the comparison is done

[brittle wing]

Okay I'm gonna choose the 200 epochs checkpoint and shut up

And apply for model maker like...it's time

[analog obsidian]

good luck!

[brittle wing]

THANKS

[dry meadow]

because i dont find like german female voices in the voice models category.. could someone help me?

[viral mason]

There's no german tag so good luck

[eternal quest]

how to use weights.gg?

[viral mason]

Are you using it on mobile or PC?

[eternal quest]

pc

I mean is there a way to use the voice models from weights.gg?

[viral mason]

Yes you can, what are you trying to do?

[eternal quest]

how can i get the model and index for the voice changer?

because I was that you can search for models and I got one interesting, so I want to use it but I do not know how 😦

[viral mason]

When you click on a voice model next to the use button should be three dots, click them and there's a download button

If u need any more help finding it I could send you a video of what to do

[eternal quest]

Thank you so much I got it. Appreciate it 🙂

[viral mason]

You're welcome

[fluid briar]

how do i fix my mic either lagging or talking with 5-10 sec delay?

i don't understand allat chunks extras stuff

[viral mason]

There's always going to like a small delay but chunk at 360.0 is pretty good and extra at 2.7

[fluid briar]

i have a problem though, my chunks don't show me values like 360, but stuff like 38400 [0.8 sec]

same with extra

lagging actually got fixed, now it's just a delay

i've seen some people that have it adjusted either without delay or just like 1 sec delay, which isn't too bad

[analog obsidian]

uninstall what you've got and instead download this: https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/

[viral mason]

I think you're using an old version

Download the Lyery

The one he sent

[fluid briar]

should i use the official or deiteris fork?

[analog obsidian]

the official is slower

fork is faster but it opens in your browser (which is actually better because w-okada GUI uses more resources)

[fluid briar]

could we go to DMs? i have few questions and need to upload a screenshot, which i can't do here

[analog obsidian]

sorry i don't accept dms, but please try to read the guide, everything you need to know it's there, trust me

is a 5 min read

[fluid briar]

it's about downloading, i'm using ADM, should i use the ones that say DML or cuda in the end?

i got confused on that question

[analog obsidian]

@low shard can you please give clay image perms pls?

[analog obsidian]

is your gpu nvidia or amd?

[fluid briar]

AMD

[analog obsidian]

download DML

[fluid briar]

alright, got it, thank you ❤️

if i'll have any more questions that won't be in the guide, i'll make sure to tell you ( if you don't mind, ofc )

[analog obsidian]

sure no prob, i'll be here

if i dont answer im busy

[low shard]

You can search rvc ai voice models at:

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

if there isnt one, you can:

• make it yourself with our docs guides

[earnest musk]

:wave: @low shard, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[golden walrus]

okay, it's fine. The quality is improving.

[low shard]

Alr

!give-media-perms 1h @fluid briar

[viral mason]

https://cdn.discordapp.com/attachments/747186322745196665/1353792072498286612/ezgif-1c927995919a40.gif

[simple ore]

there is an interesting pretrain for chinese though

https://huggingface.co/PluginsKers/Convbased-Studio/tree/main s2m

[golden walrus]

it exists ?

noice

[simple ore]

but you need to change the config in the model's folder

set "spk_embed_dim": 476

[wheat copper]

anybody know how to fixx when i load a backup and get this error

[simple ore]

well, you did not load it it seems

[wheat copper]

i did

do i load it after starting applio or before

i did it before

and it said it was successful

[golden walrus]

but let me try again with klm6.1v3 first, since Seoul Streaming Station trained it in Korean, i think it might help me out

[simple ore]

those are languages used there

[golden walrus]

but isn't it already 476? .___. i thought we don't need to care about config

[simple ore]

default is 109

[simple ore]

check the files

[analog obsidian]

how did you found this lol?

[viral mason]

Noobies, can u help someone who's kinda like annoying me about some issue I have no idea how to fix, it's @solid vortex they despereatly need help

I'll let them explain

[golden walrus]

now that you mention it

he pretty much know a lot

but i have no idea how he knows

[simple ore]

some goof came to Appio's github and demanded to add a revolutionary new architecture.

turns out it was just a pretrain with a lot of chinese

did not even use chinese embedder, I think?

[golden walrus]

but sir, it's 476 by default

[simple ore]

or maybe they did

no

[analog obsidian]

have you tried using their rvc version?

[simple ore]

same thing

no difference

[analog obsidian]

lmao