✨│ai-help

[sturdy patio]

wait till 20 mins

okok

[simple ore]

it will be done soon enough

[sturdy patio]

fuck my life

i accidentally turned it off

[simple ore]

it will resume

[sturdy patio]

do i need to add the

dataset path and all that stuff again

[simple ore]

no, just the model name

[sturdy patio]

or can i just select the model

okok thx

[simple ore]

then you can run training

[sturdy patio]

okay thanks

wait now i dont se

the compilation thing

wtf

[granite minnow]

i want to delete echo
i can delete reverb but can't delete echo
i try in anvuew v2 and mono but can't
uvr deecho cut off in 17.5khz
any better thing?

[tropic osprey]

anyone knows why the Wokada Deiteris Fork is detected by virustotal as malicious?

[idle delta]

hello, im looking to do an ai cover of a song i like and i need the most accurate most human sounding ai I can use for free. does anyone know one?

<@&1159293204038955078>

[sturdy patio]

jus make one babe

[novel summit]

a

[sturdy patio]

wait uh

anyone know why

this taking so long?

been like this for the past 40 minutes

and i restarted it and it bumped up from 60 -> 94

[keen linden]

Hey, i kinda need someone to generate a video for me.. I can't find any freaking AI where i can generate videos for free

Anyone nice enough to help out?

🙂

[loud ivy]

Hey I heard that the dieteris fork for wokada is much better than the original. Or any rvc gui out there. Can someone explain in what ways it’s better? I tinkered with it for a while and did it back to back with my rvc gui, the rvc gui was faster by a split second. So I’m not sure if it’s speed that’s supposed to be better or quality? But the quality was the exact same for my model. I was only able to open the https version of deteris fork maybe that’s why? But during pc/vr when I go to virtual desktop it does stutter when changing back and forth which is nothing I’ve ever experienced with my RVC gui. Can someone please explain what I’m doing wrong because I keep hearing it’s much better

Also I’m using 2 gpu one is dedicated to rvc

[sturdy patio]

okay so want me to explain simply? or detailed

[loud ivy]

Detailed if you can please

[sturdy patio]

okay so

w okada right

[loud ivy]

The w okada deteris fork I’m wondering about

[sturdy patio]

it has lower latency and buffer managmnet that is a bit better than w okada normal

it kinda feels better with the pitch detections, and better noise reduction for me personally

also like the performance is much better because it reduces cpu load and my memory managment on gpu feels much better

[loud ivy]

Is it possible to open an http interface for it or is it always https in the browser ? I was only able to open the browser version

[sturdy patio]

I don't really know sorry

[loud ivy]

No worries 😌

[sturdy patio]

but the fork

just think of it as an updated, and a bug fixed version

like memory leaks are gone n stuff yk

[loud ivy]

I understand, I would need to mess around with it more but. I actually only experienced bugs with the deteris version. Like switching from virtual desktop to vr in pc/vr it will stutter on a dedicated gpu. This is my experience at least. And when I use my RVC GUI (not o kada) it is smooth as butter transitioning around

[sturdy patio]

huh really

[sturdy patio]

hi noobies...

help me

[simple ore]

you've clicked the window and selected something

[sturdy patio]

what

[simple ore]

ity is waiting for you to press enter to copy the text you selected

and it froze everything

[sturdy patio]

i have to select something?

..

[simple ore]

no

you accidentally clicked the window and selected the text

[sturdy patio]

oh

so i dont click the window

at all

just leave it right

[simple ore]

press esc

or enter

to cancel the text select

[sturdy patio]

FUCK

wait so

its not selected now

right

[odd shale]

First of all, this is a general AI Server.

[sturdy patio]

bro this shit isnt working

wtf

[odd shale]

What's your GPU?

[sturdy patio]

do i have to wait 7 years for this

[odd shale]

Elaborate on your issue, please.

[sturdy patio]

hi leo so

basically like

i've been waiting 4 hours

for a 500 epoch thing

with a strong gpu and a cpu and memory

and like its been stuck and this shi ain moving boss

[odd shale]

If deiteris only opens on the browser, don't worry, that's just a feature.

[sturdy patio]

waited up till 5000 seconds before

[sturdy patio]

noo i changed that

already

[simple ore]

do you even understand what you're trying to do?

[sturdy patio]

like 70% boss

[idle delta]

i just wanted help

sorry

[loud ivy]

I was just wondering if that’s why it stutters on virtual desktop pc/vr transitioning. Browser is the only big difference change I can think of from RVC GUI

[odd shale]

It's fine but well, i guess you can start by reading the docs.

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[idle delta]

thanks homie

[odd shale]

You're welcome bud

[odd shale]

Welp, i'm not really sure about that.

You can also try vonovox.

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

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

• Vonovox

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

• Original Wokada

Not suggested nor maintained, older versions in youtube tuts are even 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

[shadow flax]

im actually losing my mind, i cant get langsmith and my local llm to work

chatgpt is giving me -20 sanity per 10 minutes

[loud ivy]

Is there any real benefit to it? I just use a very basic RVC gui right now it does formant too. It’s not flashy at all but is slightly faster than okada for me

[tough fiber]

is applio good idk i did few models before with RVC that was better than applio. i trained few models on applio and looks really bad quality etc.

idk something is wrong

[sturdy patio]

wait I redid everything to just start it up correctly and smoothly

but its stuck on 0/95

anyone knows why? Do I just need to wait a bit?

[cobalt turtle]

im using mangio RVC to train my model but im running into the problems of the program having a hard time finding the file locations. Was wondering if yall had any other programs that are better

[tough fiber]

well i trained 3-4 models today and all of them bad

at least for me

i want RVC version

mangio was good if i remember correct not that bad

[cobalt turtle]

ah yeah my biggest problem is that it is giving me so many errors casue i just cant locate a file

[tough fiber]

where did u download it

[cobalt turtle]

github

hugging face

[tough fiber]

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[tough fiber]

try applio all of them suggest applio but

idk im trying my last model on applio if that one is bad ill try mangio

[cobalt turtle]

yeah ill try mangio too

the vid a saw about it is using mangio so

idk

ts pmo

[shadow flax]

GUYS IM FLABBERGASTED

im using mistral small, no system prompts or rags or nothing in this case
also using "local deep researcher" bla bla so my LLM can google stuff....lets say the LLM didnt only google what it was supposed to

[simple ore]

and?

[sturdy patio]

bro

@simple ore

ill show you my man boobs

this shit stuck on 0 for the past 30 mins

[simple ore]

did you click the terminal again?

[sturdy patio]

nope wallah

i didnt

[simple ore]

did you run run-applio-amd.bat?

[sturdy patio]

yes

wallah I did

[simple ore]

why did you have 900+ input files in preprocess?

[sturdy patio]

i have lots of

audio recordings

LOL

bro boobies

i mean noobies

wallah im about to crash out

u gyatta help me

[brittle wing]

Do you guys have suggestions I don’t have a GPU

[viral mason]

he's asking about okada btw ^

[sturdy patio]

bro

daddys

help a bbg out (no homo)

[chrome yew]

am i still able to make rvc run on my 5080, if so would anyone be able to help me through it

[bronze mirage]

Applio is a rvc?

[simple ore]

it is fork of RVC (not realtime voice changer)

[simple ore]

https://github.com/IAHispano/Applio/issues/1020

also change the folder name and remove "(1)"

[bronze mirage]

Theres alooooooot i need to learn to slightly understand all this /:

[sturdy patio]

bro my aplio is not starting the training wtf

cpu usage is at 4% and its stuck like this for the past hour

[knotty briar]

Can someone help

i downloaded apolio but the CMD prompt keeps coming up blank

Alternativly if i run it in admin mode i get this

K I Got this

[sturdy patio]

dont move it to a folder with spaces

[knotty briar]

ohhh ok

[tough fiber]

@hallow thistle namari do u have RVC Version for making models

i didnt like applio

i did 5 model and all is bad

[simple ore]

dont run it as admin

[tough fiber]

bruıh on mangio still theres pytorch issues

for 5000 series

bro holly im done with making models its headache

applio isnt good

all models is worse

if someone know how to use mangio with 5000 series can help me /:

[simple ore]

you can try pulling the latest code from main branch, there's an option to make things just like in rvc - single mel loss, normalization for individual slices, etc

you cant make mangio compatible with 5000, it requires new torch and a lot of fixes in the code

[little igloo]

if I train a model with a dataset for 30 minutes, say 600 epochs, how can I improve it if only I will use it? maybe dilute it with my voice or something like that?

[simple ore]

600e would most certainly be too much

[little igloo]

it doesn't matter how many there are, during the training process you can choose the best version, so what about additional training on your voice?

[simple ore]

as long as you save a voice model every 10 epochs, you can pick the best, sure

[little igloo]

either u don't understand me, or no other options except endless trials on various datasets

sorry for pings

[simple ore]

you want to merge some random model with a voice model of your voice? sure you can do that, as long as the model's sample rate is the same

the result may not be what you expect

[knotty briar]

Ok so once i train the voice model

where does it export to

im not sure how im getting that to work

[simple ore]

logs/modelname

[reef agate]

applio refuses to do anything. ive followed all the instructions and when i attempt to boot it up its just a blank command window

[simple ore]

just wait for a bit

you can close it and disable discord connection manually

assets/config.json --> discord_presence: false,

[knotty briar]

@simple ore Not sure what im missing? I'm a bit slow

[simple ore]

did you even run preprocess and extract features steps?

or did you just click train? 🙂

[knotty briar]

I did preprocess not extract i did not know extract was mandatory i thought that was for file cleanup

[simple ore]

what is inside of this

[simple ore]

preprocess slices your audio files, extract features prepares files for training

[knotty briar]

ohhh

i thought was preprocess ok mb

let me start over

[reef agate]

ok so. if my dataset is 19m 13s how many epochs should i train it for. im sorry if thats an annoying question im switching from weights to applio and have no clue what im doing

[simple ore]

there's no answer for that, depends on the dataset's content, better check the tensorboard charts, listen to results

[reef agate]

ok it got done training suspiciously fast and when i checked the log folder there was not a single pth file present

i think i did something wrong

[simple ore]

check the terminal window log

[distant idol]

where can i find that

even the 460e eponch doesnt have sound

[simple ore]

in the model folder, or in eval folder under model folder

it means something went wrong way before

tensorboard may show when

[distant idol]

but tensorboard doesnt work

[simple ore]

how

run-tensorboard.bat

what happens?

[distant idol]

now it works

[simple ore]

yes, it shows NaN all over, it is fucked

look at SCALARS tab

[simple ore]

yeah, probably would be best to use the latest update

something went wrong

[distant idol]

which update should i do?

[simple ore]

adrenalin driver > 25.5.1

hip sdk 6.2.4

and latest zluda

[distant idol]

It says 25.6.1

[simple ore]

#✨│ai-help message

did you use 6.2.4 guide I linked?

[distant idol]

not yet because i had 24....

[simple ore]

but now you have 25.6.1?

and youi're trying to use the old method?

[distant idol]

no i'm updating now to 25.6.1

now i have 24....

[simple ore]

okay, then follow 6.2.4 guide after

[distant idol]

does it work too with 25.6 instead of 25.5.1?

[simple ore]

yes

25.5.1 is the driver that breaks the old method

[distant idol]

I'm updated

gonna use the text now

[distant idol]

I just dont get the: "'**import torch

if torch.cuda.is_available() and torch.cuda.get_device_name().endswith("[ZLUDA]"):
# disabling unsupported cudnn
torch.backends.cudnn.enabled = False
torch.backends.cuda.enable_flash_sdp(False)
torch.backends.cuda.enable_math_sdp(True)
torch.backends.cuda.enable_mem_efficient_sdp(False)''**

Do i neet to complete overwrite all of it whats in the zluda.py?

[simple ore]

you need to open zluda.py in notepad and basically cleanup some stuff

just paste the code that is shown between quotes

just save this over

[distant idol]

Thanks 😄

gonna launch applio now, how can i check if its working now this time>?

[simple ore]

make sure you ran the patch

and it copied the files over

[distant idol]

yes it did

[simple ore]

okay then

[distant idol]

like 6 or 7 copied

[simple ore]

okay, then go ahead with the new run-applio.bat

it is going to compile stuff again

[distant idol]

i will, and tensorboard can show me if its working or not?

[simple ore]

only if it starts training

make sure you deleted the old model folder

[distant idol]

i did!

at logs right?

got an error

[simple ore]

hm

@distant idol i need to figure out why it does not work

for now, could you open rvc\lib\algorithm\commons.py in notepad

find the line def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):

and delete @torch.jit.script prior to it?

then save the file and restart applio

[quiet flint]

How would I distill Kimi K2 into MythoMax?

[simple ore]

????

[quiet flint]

It’s called LLM distillation

It has existed for a while

[mossy hull]

so I am having some weird traceback error
I wanted to merge two voice models for an experiment, and yet I keep getting this error

Traceback (most recent call last):
File "D:\RVCMAIN\infer\lib\train\process_ckpt.py", line 258, in merge
torch.save(opt, "assets/weights/%s.pth" % name)
File "D:\RVCMAIN\runtime\lib\site-packages\torch\serialization.py", line 440, in save
with _open_zipfile_writer(f) as opened_zipfile:
File "D:\RVCMAIN\runtime\lib\site-packages\torch\serialization.py", line 315, in _open_zipfile_writer
return container(name_or_buffer)
File "D:\RVCMAIN\runtime\lib\site-packages\torch\serialization.py", line 288, in init
super().init(torch._C.PyTorchFileWriter(str(name)))
RuntimeError: [enforce fail at C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\caffe2\serialize\inline_container.cc:365] . invalid file name: assets/weights/.pth

I have no idea what's wrong here

[simple ore]

you did not give it a name?

[mossy hull]

they're supposed to have a name?
both models are named, and I assumed that a fused model would be generated

[quiet flint]

#1397787937521012836

[simple ore]

you're trying to stuff an elephant into a matchbox

[simple ore]

when you merge two models you provide a name of a new model

[quiet flint]

Who, me?

Kimi distillation needs to happen at some point

[simple ore]

yes

maybe, once somebody downloads the whole 2TB of it

[mossy hull]

lmaooo I'm dumb
thank you so much!

[quiet flint]

ok then, distill qwen2.5 and mythomax because qwen has the personality of a dead fish

[simple ore]

deepseek distill is garbage

qwen2.5 I only used coder 14b

[quiet flint]

cant i use openrouter for kimi k2

[simple ore]

i did not expect it to have a personality

[quiet flint]

expect what

[simple ore]

https://openrouter.ai/moonshotai/kimi-k2

[quiet flint]

its marginally better than qwen2.5 at persona and roleplay

qwen is absolute dogshit at it.

[oak edge]

Someone help me, after loading backup in RVC, when I try to train it starts from epoch 1, even though there are 124 epochs already there in the logs folder

Are there any ways to force resume from 124th epoch

[wide torrent]

why does the voice keep repating

[snow chasm]

i want a tts which have my model in it and also have emotions detection can any one help

[crude bolt]

you created an output input feedback loop

[fallen venture]

waaa

[crude bolt]

to help out with a plan yes, you will have to learn how to do it though.

• create a rvc model first, you need audio tooling and then a model trainer, like applio.org.
• the same tool (applio) can be used for tts.
• emotion detection; asfaik does not exist.

[latent kettle]

You can use a tts and then inference it in applio

[wide torrent]

how do i fix?

[crude bolt]

by finding the output input feedback loop (do you even know what i talk about? 😂 )

let me give you an example:

• you put the outputcable of your speaker straight back into the input plug of your mic. disconnect the cable
• your speaker totally blast the sound back into the mic. use a headset or lower the volume

Sometimes i see that people have in windows for their recording device (eg the mic) the option on that says "Listen to device" , turn that off.

[crude bolt]

what i said over here doesnt immediately get you in trouble but in some scenarios you accidently create a feedback loop and search like no tomorrow that you overlooked this option.

[somber tusk]

Why is my account logged out of Weights website? (It was suddenly.)

[crude bolt]

log back in. not worth to invest time in this, if it no longer happens then it's fixed.

alright have a good day all 👋 - g2g

[crude bolt]

before i go @somber tusk - if you have a cookie cleaner or cleaned cookies on purpose or some tooling that smartly 'make your internet safer' (by cleaning cookies) then this happens ... happened to me.

anyways out now 😉

[wide torrent]

Well im not used to this?

Smh

[rose marten]

is there i way i can hear my self with the voice changer?

[maiden wren]

Yes, You can you have to find a settings named Monitor Which Is below input and output settings click on monitor setting and select it to your speaker

[simple ore]

as long as there are D and G files in the log folder it will resume from the latest save

mainline rvc is bugged a bit as it resumes from -1 epoch

[simple ore]

I suspect you ran a wrong patch file

[oak edge]

both are there but it starts from 1st epoch

instead of 126th

[simple ore]

mainline RVC?

[oak edge]

Applio

I train with colab

[simple ore]

In colab you need to restore files from the backup

[oak edge]

backing up files with Drive

[oak edge]

yup I did that, and I can access epoch 126 for interference

[simple ore]

you can check that the files got loaded back into logs using file explorer

[oak edge]

yup they are loaded as well :(

[simple ore]

you can access epoch 126 voice model, not D/G, in inference

[oak edge]

I checked D,G wait ahh lemme send ss

[oak edge]

here is the d/g files

[simple ore]

that's the backup

now in colab

this folder button

[oak edge]

okiee I'll show that one as well

[simple ore]

then Applio/logs

[verbal light]

guys
in rvc, my voice changer sounds like hes choking or moaning
sometimes he speaks gibberish
when i stay quiet
what do i do 😭

[pastel oak]

Sounds like your mic is picking up every noise in your room. Move the threshold slider to the right, enabling sup2 can help too

[simple ore]

as long as you did not click 'fresh training' on UI, it should resume

[oak edge]

nope I haven't done that either

😭 I'm not even sure why it's not loading up from where it left off

[oak edge]

I had to force stop the training by pressing stop here, since the browser public link got "error connection timeout"

[oak edge]

yea it starts from epoch 1 there

but the weird thing it says, skipping the epoch_X since epoch_x exists

[simple ore]

that's for saved voice model

[oak edge]

also I'm sharing one folder between 3 google drive accounts

but the model gets to access it and it loads inside colab

so

[simple ore]

let me check

[oak edge]

and it starts from here

[oak edge]

okayy

[simple ore]

I imagine you f'd up something, it backed up e1 file

now you're restoring e1 file

since you're using 'save only latest' it saves only one copy of D/G named 2333333

that's what it is

it resumes properly, otherwise it would've said loading ... pretrained /f0xxxx.pth

[oak edge]

it did that for once, starting from epoch 120

[verbal light]

no lol, its a mic built in my headphones and in a vc, it basically feels like i stay muted but i dont... so i think thats not the problem

[oak edge]

then after I stopped using the stop button and resumed

it went back to 1

and now yup not going back to 120

[simple ore]

because you messed up your save and backup

[pastel oak]

You just confirmed it is the issue
Picks up your headphones volume aswell

[oak edge]

😭 so no way to force go back to epoch 126?

[simple ore]

I can recommend going back to google drive and looking thru the history of D/G files and recovering the version that is not e1

[verbal light]

hm

[oak edge]

wait how do we go through history?

[oak edge]

or are there any manual ways to edit that file

[simple ore]

find the date your last epoch was saved (same as voice model file, I assume)

and revert both D/G to the same datestamp

[oak edge]

okayy I'll try that

[knotty briar]

Went to sleep 8hours ai is still training

i did use my GPU just cpu

would there be a speed difference if u sed my GPU too?

Im wondering how long this is gonna take

[oak edge]

@simple ore okiee I found the version how do I restore it

[distant idol]

I run the 6.2 one

[simple ore]

lets try again.. uninstall torch, install cu118 torch, run 62 patch, show a screenshot

[distant idol]

I will later today, i’m not home today

[simple ore]

there should be an option under ...

[oak edge]

i deleted newer versions so current version is now 97

[simple ore]

okay, do the same for g

[oak edge]

💀 i think I did some step wrong

[simple ore]

then restore backup

[oak edge]

now all the stuff gone from the drive

[simple ore]

go to trash and recover it

[oak edge]

there is nothing in the bin

[simple ore]

not sure why it is called that

[oak edge]

nothing is in there 💀

126 epochs gone idk how T_T

[simple ore]

do you have the colab running at the same time???

dude

[oak edge]

💀 yes

[simple ore]

[oak edge]

💀 wym

What happened to those 126 epochs

💀 💀 💀 💀 💀 💀 💀 💀 💀

i clicked load backup

[simple ore]

kill the colab tab

[oak edge]

after editing out the D/G

[oak edge]

okiee

[simple ore]

to go google drive, rvc_backups, right click on it,

there will be activity

find your D/G files from that epoch

download them to your PC

[oak edge]

💀 I'm seeing lot of an item was placed in messages

like all of the stuff

[simple ore]

find the set saved

[oak edge]

ok so all of the stuff that should be in the backup folder is now scattered in my main drive

[simple ore]

it should not be scattered

[oak edge]

😭 there are already many stuff going wrong

It is scattered T_T

[simple ore]

make sure that in RVC_Backup/ARK the folders with sliced_audios and everything else is populated

[oak edge]

💀 lemme see

[simple ore]

you fked up something again if that happened

[oak edge]

yea sliced is populater

like folders inside of backup is populated

[simple ore]

cleaning up those hundreds of files from your main drive gonna be ass

[oak edge]

but index epoch all the main stuff are gone

[oak edge]

💀

[simple ore]

okay, from activity did you find D/G/model file saved at epoch 128?

[oak edge]

😭 that I found already

[simple ore]

download them to your PC

[oak edge]

okiee

[simple ore]

then upload them into backup/ark folder

[oak edge]

but all the other epoch's are gone T_T

[simple ore]

who the f cares

[oak edge]

even the index file is gone

like no other files is there T_T

[simple ore]

you also need model info, config and filelist

[oak edge]

wait so can I just start training with D/G files only?

[oak edge]

💀 i think all of it are gone

or scattered

this is how the backup folder looks

there should be epochs and other stuff in this main folder

filling it up

but all of em missing

[simple ore]

you need at least the model info, others can be recreated

[oak edge]

model info lemme search

okee I think I got him somewhere

idk where

[simple ore]

and filelist

copy them back into backup

[oak edge]

location is mydrive

okayy

💀 gosh I'm done Imma clean wipe

[simple ore]

so in your rvc_backup\ARK folder there should be

• D/G files

[oak edge]

okiee

[oak edge]

WELL the f0 voiced

f0 is empty

[simple ore]

anyway, you can always take your dataset file and re-process them

wont hurt if you use the same parameters as before

[oak edge]

😭 fr Imma clean wipe this drive

uhm how do I fully wipe the drive

HOLY SHIT there is no way to wipe T_T

[simple ore]

it is easier if you install g-drive on PC

easier than manually clicking files with the mouse

[oak edge]

🥲 i manually deleted all of em

[pastel oak]

You should only be using gpu, not cpu

[simple ore]

CPU is extremelly slow.. like 30 min/epoch instead of 3 min/epoch

[distant hollow]

Can someone tell me how I can make an ai voice module of my own
i know its written somewhere im just really lost 😭🙏

[knotty briar]

How do i make it use my GPU?

[knotty briar]

How do i make it use my gpu

[pastel oak]

Whats your gpu name, whats the program you downloaded

[knotty briar]

Nvidia

But like

do i put it like this

or another way?

cause i tried and this didnt work

[knotty briar]

^

[pastel oak]

Your gpu is recognized so im not sure how you managed to fuck it up

Did you change anything related to gpu after that

[knotty briar]

No

Let me try again pls wait 5 min let me set up again

[knotty briar]

I get this

[flint vapor]

Did you run start applio.bat?

[knotty briar]

What kind of Did you turn on and off quetion is that

[flint vapor]

I said, if you open the .bat for run applio right?

[flint vapor]

Calm down

[knotty briar]

I Am calm that quetion doesnt make sense

How would i get the CMD or interface

without running the .bat

[flint vapor]

I asked you which bat have you run

If you run that or another

[knotty briar]

No just the normal bat

[flint vapor]

Ok, are your drivers of your gpu updated?

[knotty briar]

Let me check actually

[flint vapor]

Sure 😄

Yes you can ask ;D

Depends, like, of course first of all depeneds from model and how many chunks have you set

Which is your gpu?

Which rtx?

Which series numer

Sure 😄

Go on task manager, performance

And you can see

Ok nice, which voice changer did you download and from where?

From there there are only models,

But which voice changer did you download?

Not model

This is a model

Not a voice changer

What are you trying to do with that model?

E troll? Roleplay? Catfish? Testing?

[knotty briar]

@flint vapor How long does it take to train a model?

8hours? more

cause mine was on hour 9 and still not done

I have a 64gb 4070 its a laptop but still a 4070

I Dont know if this thing was using my gpu or not

[flint vapor]

A 4070 can't be 64 gb, you mean your normal ram?

[knotty briar]

8GB VRAm

normal ram is 64gb

@flint vapor

[flint vapor]

Yes wait, is ok normal ram don't count now for applio, 8 gb vram is not so much for training

Especially also because you have laptop version

How long is the dataset?

Depends also from the epoch you put and lenght of dataset

[knotty briar]

@flint vapor Wana VC with me pls

so i can see

what you mean

This is the raw WAV

[flint vapor]

I can't vc at the moment, yes is pretty long this dataset, so it will require a bit of time, how much epochs did you put

[knotty briar]

I left it at default but how many You reccomend?

Default is 10

[flint vapor]

Is impossible you put 10 epochs and is slow 8 hours

You put 500

Is a lot

[knotty briar]

Thats default

[flint vapor]

But for 20 min it should be fine, but you can reduce them a bit at least you can try with 350

Is normal if you havve 8 gb of vram

With long dataset will be slow

With many epochs

[knotty briar]

well again i slept and woke up many hourslater

so 300?

350?

[flint vapor]

Yes you can put 320

[knotty briar]

Ok ill try

[simple ore]

this should never happen:

It never says the name of the GPU

unless you made some changes in the code

why did you paste the name of the GPU here?

[tender ingot]

whenever I tab out the voice change just goes to shit

like utterly goes to shit

[low shard]

This is a General AI Server, AI has many fields, so we can't know your issue

Please Elaborate:

• your PC GPU
• your operating system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[knotty briar]

When does it end? Expectedly?

[knotty briar]

I Wasnt sure if it was using my gpu or not i set to make the model last night i slept 6-8hrs its still going

[snow chasm]

please suggest me some good tts which are free and support emotion

[knotty briar]

But apparently i had alot of epochs so i reduced i still dont know if im running slow or not

[west rampart]

how do you train rvc models, whats the program called

[simple ore]

you should not touch it, it should've been 0

[knotty briar]

If you saw the chat i already put it to 0 infact the first attempt that was hella slow was 0

[simple ore]

open task manager/performance, click on your gpu on the left hand side, then make a screenshot

[knotty briar]

and its back on 0 now

[simple ore]

close applio

re-start

go to training tab, before you start training, in the training section expand the advanced settings tab and check [x] chekpointing

[knotty briar]

ty

[simple ore]

right now with batch 8 you endede up using shared memory, that is a huge performance hit

also make sure this is off

you can also lower the batch size

[knotty briar]

Fresh Training

Cache Dataset in GPU

SHould those be checked?

[simple ore]

do not check cache

uncheck

[knotty briar]

fresh training?

[simple ore]

fresh training should only be checked when you want to wipe your training attempt

and start from 0

[knotty briar]

Also this is my raw dataset

The raw wav

What should this be ^

@simple ore

[simple ore]

that's fine

[tough fiber]

can u help me how can i do this

[west rampart]

is this meant to be 0 if following ai hub guide?

[simple ore]

download https://github.com/IAHispano/Applio/archive/refs/heads/main.zip

unzip to a new folder, run run-install.bat

[simple ore]

usually it is 2 files for 30-60min set

if you have clean dataset without noise and enoigh silences already, you can set it to 0

[west rampart]

I cut the silence on audacity

like the guide said

my dataset is 16 mins

[tough fiber]

@simple ore im not sure with this which one should

[frozen harness]

does anyone know to do a thorough search on voice models that will let us filter attributes like dataset, steps, batch size and epochs individually?

[low shard]

you can't do that unless you search like epochs in the search bar, and that's not needed, they dont mean the quality lmao

the only way the check the quality is listen to the samples or use it urself

they are just training info

[frozen harness]

aw I was hoping there would be a way to better filter through them so I didnt need to download 100 of them and test them individually

[low shard]

nope, the only way is either testing or listening to the demo, no other way

[frozen harness]

sadge

thanks anyways

[low shard]

yw!

[simple ore]

contentvec is default, for spin you'll need to download a pretrain

[normal drift]

Hi, I haven't trained a SD lora since 2023, what's the meta Colab notebook for that purpose now, and is there a great guide on it ?

Also I forgot what I used back then, but what's the meta "dataset scraper" tool now ?

[wild bobcat]

using old rvc gui as my local gui, it takes up alot of my c drive space for no reason when opening it up

[wild bobcat]

so can someone tell me: is there a folder on the c drive that contains a bunch of assets used for the rvc gui?

[viral mason]

do you use Nvidia or AMD for your gpu

[low shard]

Please Elaborate:

• your PC GPU
• your operating system
• what you want to do

[low shard]

dont use rvc gui, its outdated

[wild bobcat]

no no the thing is

it was working just fine before

but now it EATS my space for whatever reason

[low shard]

you shouldnt use it, it's not updated from 2023 and runs on Hopes and Dreams 😭

its unstable

[wild bobcat]

wasnt unstable until recently for me but

im assuming youll tell me to download applio right

[low shard]

yeah it's up to date, stable and works better than that

[viral mason]

https://tenor.com/view/undertale-asriel-dreemurr-tobyfox-flowey-frisk-gif-7429666

[wild bobcat]

🥹

could you tell me where to download it then?

#✨│ai-help message

[low shard]

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

[wild bobcat]

alright thank you

[viral mason]

https://tenor.com/view/star-wars-general-grievous-all-too-easy-easy-the-clone-wars-gif-18491449

[wild bobcat]

does it really have to be on the C drive?

(i wanna download it)

[low shard]

we would suggest tho, not sure if it still makes issues on other drives, would be better to ask that to @simple ore

[wild bobcat]

asking cause i have space issues on c drive

i have an external drive

[wild bobcat]

alright

[tough fiber]

well i tried applio again

and many new data

its really worse

i prefer old rvc versions

instead applio

tf

[simple ore]

did you use post normalizing method?

[tough fiber]

i guess no whats that

[simple ore]

in preprocess

[tough fiber]

well i didnt use

but i guess not because that whole

model is worse idk

its like robotic and too bad

oh i found ill try this

https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/releases/tag/2.2.231006

this one was too good

[simple ore]

feel free to make a model there and compare

same parameters and stuf

[tough fiber]

lets see if its work with 5000 series

[simple ore]

nope

[lavish kindle]

anybody can help me? it worked fine and now i just opened it and it doesn't show anything in the input and output list

im using the b2332 nvidia-cuda

[simple ore]

runtime\python -m pip install ... same torch thing

but it will likely break 20 other places where it tries to use torch.load() without weights_only=True you'll need to fix the code for all that

[olive plover]

hello

what this why????

[knotty briar]

My thing still going

is it meant to be like this?

how long do i got left

@simple ore

[spare eagle]

any cover ia for android?

[knotty briar]

NVM my Model just finished training

So what do i do with these

Which one is my model?

@simple ore

[tough fiber]

is it this u talking about

[lavish kindle]

does anybody know why is input and output not showing my mic or headphones?

[tough fiber]

ah this pytroch version thing again

[tough fiber]

NVIDIA GeForce RTX 5070 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.

do u know how can i fix this

i upgrade it but still

[tough fiber]

i really need help about RVC Mainline training with rtx 50 series gpu

this pytorch version print annoying

wanna throw this 50 series gpus bro nothing work correctly

[simple ore]

if you ran 300e max, then 50 more epochs.. so about an hour

[simple ore]

sm_120 - requires pytorch install cu128

same way you did it with applio, just a little difference which python to run

[barren hatch]

Hello i want a voice changer for free

[knotty briar]

You missed my message it finished

But what do i do with the files

[simple ore]

click refresh button on inference tab, then give them a try

also use run-tensorboard.bat to see the tensorboard in order to pick the best epoch

[knotty briar]

Where is the inference tab

[simple ore]

the very 1st tab in applio

[distant idol]

I did! but it gaveme this warning

To be sure, i did add them to Path now 🙂

[simple ore]

no need

now the patch 62 screenshot

[distant idol]

i followed this

[simple ore]

okay

did you edit that modules script?

[distant idol]

not yet!

[simple ore]

dont

give inference or training a try

[distant idol]

I will! Keep you posted! (With training there is no where i can select v2 in applio is that correct?

[simple ore]

it is v2 by default

[low shard]

Please Elaborate:

• your PC GPU
• your operating system
• what you want to do
• what tutorial link are you using
• a screenshot of the program

[barren hatch]

https://www.youtube.com/watch?v=SxdnGxicJOg

[low shard]

that uses an over year olod version of original wokada, vb audio cable creates issues on windows

delete the zip folder and uninstall from windows app settings

also i saw its about ai girl voice, are you looking to do e girl trolling/catfishing?

[barren hatch]

Bro never mind

is so many things to do and i don t understand

[distant idol]

starting now!

[knotty briar]

@simple ore I Do not know how to understand these statistics

How do i tell which is the better model

[simple ore]

show the charts, loss_avg_50 on scalars tab

like this:

[analog obsidian]

oh your mel diverged, that explains the weird results

[knotty briar]

So this ?

[simple ore]

S C A L A R S

[knotty briar]

No

[simple ore]

click

there

my dude

[knotty briar]

Yeh my bad

[analog obsidian]

lmao

[simple ore]

do you even read

[knotty briar]

Maybe I just missed when you said scalers

[analog obsidian]

yellow scalars, not the gray one

[low shard]

dw i can help you, are you looking to do e girl trolling/catfishing?

[simple ore]

right there

[barren hatch]

i just want to have a voice changer so i can troll my friends to make funny and things you know

[knotty briar]

Sorry my autistic brain kepts look up and right

[barren hatch]

i want to have voice from dexter i have it but i don t have voice changer

I understand nothing from all tutorials and like that.

i seen tutorial 5-10 minutes and i looked 30 minutes

and i understand nothing

[simple ore]

give these models a try

[barren hatch]

probably i will pay 100$

for pro and that it

[knotty briar]

k ty

[knotty briar]

wait you got that from looking at the graph?

[simple ore]

and perhaps you can do another run with a different batch size

[barren hatch]

@low shard if you are gonna help me is good

if you can t then idk.

[simple ore]

4 or 6 instead of 8

[barren hatch]

do you have voice changer?

how!

you paid

or?

i can t understand

5-10 minutes then i looked for 30 minutes

i swear to god

Can you help me?

[simple ore]

@knotty briar make another folder in logs, copy config.json, model_info.json, filelist.txt from the old model folder over there, then pick this folder on the training tab and run training with the same parameters, except lower batch size

[barren hatch]

okoda?

-re

-rt

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

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

• Vonovox

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

• Original Wokada

Not suggested nor maintained, older versions in youtube tuts are even 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

[barren hatch]

@loud hill what do i need to install first

For Windows
Download this: VAC Lite (Virtual-Audio-Cable by Muzychenko)

i looked on this video

for 30 minutes

i can t understand it!

everything

[knotty briar]

Btw when testing the voice this do i unselect these and if so how cause is this overlapping the voice samples when i do text to speech

My TTS sounds like there is an un wanted accent

[simple ore]

is this the actual language you're using?

[knotty briar]

No no im in english ok i just realized they are the Initials for languages

[simple ore]

arabic algerian?

[knotty briar]

it's actually english ok i switched it

model sounds awesome

now im wondering

If there are other english neurals that work better

cause there are alot of english options on that list

Are they accent based?

[simple ore]

no, they are very basic generic tts reads by microsoft

[knotty briar]

OH so they dont matter as long its in english?

[simple ore]

not really, just pick a male tts voice if your voice model is male

[barren hatch]

very good is directly virus ahahahaha

ahahahhaha

[knotty briar]

I see np ty its female but ty

[barren hatch]

@loud hill

very good

install it

and nothing happend at cmd

👍

look at the whole video you said no?

[low shard]

so, e girl troll your friends? or like spongebob?

[knotty briar]

What is a good application to run the pTH through so i can test Voice to Voice quality in real time?

[low shard]

thats old asf

dont suggest it

[barren hatch]

Bro i think

you trolling me i swear

bro the cmd is not working

i open then instantly closed

[low shard]

Im just asking you to be sure what program you need, if you need an e girl or things like spongebob to troll your friend

I need to know that first

[barren hatch]

Bro

i just want to make

the app voice changer

first

then i will say what i want

[low shard]

yes, but you need to tell me first, else I cant help

[barren hatch]

DDEXTER

DEXTER

BLOOD

BLOOD

you understand now?

[low shard]

delete all you had from thst video

its outdated

what's your pc gpu? the program version depends on it

[barren hatch]

... bro

my head is exploded right now i swear.

[low shard]

yep dw, I just need to know your GPU so i can send you the correct GPU version

[low shard]

AI is not 1 click 😭

[low shard]

@barren hatch no need to spam it, that's just task manager

i need to know the GPU name

[barren hatch]

for what?

[low shard]

for me to give you the correct version 😭

[barren hatch]

RTX 5070

[low shard]

EVERYTHING depends on the GPu

[low shard]

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

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

• Vonovox

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

• Original Wokada

Not suggested nor maintained, older versions in youtube tuts are even 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

[barren hatch]

and rtx 3060

[low shard]

1st link, wokada deiteris fork

theres also vonovox, but the only difference is the User Interface

[low shard]

the rtx 5070 is better

[barren hatch]

never mind fellas i uninstall

you give so many bullshit things and i can t understand it

[knotty briar]

@simple ore Last quetion....What application can i use to test Voice to Voice with the model?

[barren hatch]

so no thanks

[low shard]

yeah you always had to, that program was outdated, you gotta read the wokada deiteris fork guide

[simple ore]

inference tab

of applio

[low shard]

AI is complex and intensive, dont expect it to be a 1 click program, you just gotta read the guide

[simple ore]

load wav, select the model, run

[barren hatch]

Do you have idea how can i install the voice changer?

because i can t open the cmd

i open then closed directly Noobies

[low shard]

the only way is by uninstalling from yt tuts, and reading the wokada deiteris fork guide 😭

[barren hatch]

Bro

[low shard]

like there isn't any simpler way

[barren hatch]

what you don t understand

come with me

voice

and i will show to you

[low shard]

I can't VC, i already told you, you were using an old version

delete the old version

and read the guide

dont trust yt tuts for voice changers

they are all old

[knotty briar]

No i mean live voice to voice

Sorry i guess to say thats why im making these models in the first place

[low shard]

realtime voice changer for calls? that's not what Applio is for, you need wokada deiteris fork

[knotty briar]

Oh my gosh

[low shard]

you talked about female models, are you looking to do e girl trolling/catfishing?

[knotty briar]

All these time wasted

[barren hatch]

ok fine

[knotty briar]

No no its actually for RPing in a tabletop

[low shard]

no you can just use the models in the wokada deiteris fork, they use the same type

[barren hatch]

Download NVIDIA on Windows
The latest version as of December 7th 2024 is: nvidia-b2332 (click here to download)

[low shard]

RVC = Retrieval-based-Voice-Conversion, the best Few Shots Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.

Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)

Hopefully you understand by this

[low shard]

get the nvidia rtx 50 serie, for your rtx 5070

[barren hatch]

i DON T use

on rtx 5070

i use on rtx 3060

[low shard]

thats great

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

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

• Vonovox

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

• Original Wokada

Not suggested nor maintained, older versions in youtube tuts are even way worse. GUIDE

• RVC GUI Mainline Realtime

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

[low shard]

1st link, wokada deiteris fork

[knotty briar]

I'm on the page looking for the download area

[low shard]

whats your pc gpu btw?

[low shard]

i mean, if you want that then yes

[knotty briar]

oh nvm

you gave me that

oh wait thats the same page i found

[low shard]

basically Applio/RVC = making models, use models on pre-recorded audios, TTS
Wokada = Realtime Voice Changer

[knotty briar]

yeh k looking for the download

[barren hatch]

very good bro

fuck the vokdodo voice ai

[knotty briar]

That i got im just cconfused on where the download for windows is

[barren hatch]

nevermind i will uninstall it

[low shard]

that's a false positive, the program is Open Source, you can even read and modify the code yourself

😭

[barren hatch]

i don t want viruses

[low shard]

please read carefully

it's not a virus, you can literally look at the code yourself

[barren hatch]

just shut up bro

i understand you make

[low shard]

what's your pc gpu? it depends on that

[barren hatch]

your work

[knotty briar]

4070

[low shard]

@barren hatch No need to be offensive, I'm explaining you, the App Is Open Source, here's the full code, read it yourself if you dont trust me https://github.com/deiteris/voice-changer

[low shard]

What?

[barren hatch]

yea very goodd you give me a github

[low shard]

get the nvidia on windows download then

[barren hatch]

to look on 2-3 videos

with app

very good

[low shard]

what?

that's the source code..

[low shard]

every single line of code is here, if the app was a virus, it would have been taken down, since github is by microsoft

[low shard]

the github was just to prove you the app is open source 😭

[barren hatch]

yea very good

fuck the vokodo app

and finish it

[low shard]

there's NO difference from the code in github, to the huggingface download.. except you have to compile yourself the github one

[barren hatch]

so many things to do for this bullshit voice ai

i just wanted to make funny on fivem with the voice dexter

[low shard]

@barren hatch I'm warning you, please stop being aggressive and cooperate, AI isn't easy

[barren hatch]

BRO I DON T UNDERSTAND YOU understand NOW

[knotty briar]

@low shard So I tested TTS that sounded awesome

I Tested VTS on the inference tab and that sounded horrible did i do something wrong or is that just cause the model is not made for it

[barren hatch]

i stay 30 MINUTES FOR A TUTORIAL ANDD the tutorial have 5 minutes

[low shard]

Im trying to explain you, ALL VIDEO YOUTUBE TUTORIALS ARE OUTDATED, the one i gave you is the best one, it's open source, free and safe, you just need to use that one

[barren hatch]

yes very good

i think

you are ...]

never mind

you give me something to look on github

for nothing

i looked to see a open source

and i thought you gave it to me voice changer directly

[distant idol]

@simple ore It has been like this fot half an hour now

[low shard]

you wanted proof of the app being open source and safe

[barren hatch]

because you told me was the last version and is not working

ok bro

[low shard]

there's no 1 click way, it's a complex setup

[barren hatch]

now I WANT to install voice changer

[knotty briar]

Boop^

[simple ore]

it is compiling

[barren hatch]

https://www.youtube.com/watch?v=xwdbwbtO-FQ&t=60s

look bro

8 minutes

[low shard]

then keep reading the guide i sent, you just need to click more info and run anyways

[barren hatch]

30 minutes i look on this bullshit tutorial

sent me again

[simple ore]

unless you accidentally clicked the terminal window and selected some text and froze it

[barren hatch]

and if it s working

[low shard]

that is an over year old video...

[barren hatch]

you will give me support directly

[low shard]

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

[barren hatch]

bro

[distant idol]

Now i didn't

[swift thunder]

What do you recommend? Contentvec or Spin?

[low shard]

yep, you also need that

AI is complex and intensive, read the guide carefully

[knotty briar]

Nick could you answer the one quetion please Im getting wokada but download is sllow so im just using the inference not sure if I should be worried or not

[barren hatch]

yea fuck the ai

[knotty briar]

with the results

[barren hatch]

and finish that was the tutorial fuck the ai

i just wanted to make some funny but you destroy my day i swear to god

[cloud pebble]

contentvec imo. spin is harder to get good results from imo, where contentvec is much more figured out and tested

[swift thunder]

give, thanks

[viral mason]

Bro's too lazy to set it up

Let him figure it out

[barren hatch]

ok i fixed nick

Now how can i fix the delay

[low shard]

please share a screenshot of ur settings

!give-media-perms 1h @barren hatch

[lavish kindle]

does anybody know how can i fix audio on client not showing my mic or anything at all?

i also made a thread on forum #1398010031131066438

[low shard]

uninstall vb audio cable from windows app settings, get vac lite from the guide
extra: 2.7
chunk: 90
f0: rmvpe without onnx
input: microphone
output: line 1
monitor: headphones optionally to hear urself