✨│ai-help

[fossil sage]

yes it is

[teal ferry]

you see this?

[fossil sage]

yes

[teal ferry]

hold control and click the gradio dark link in the terminal

whatever it may say

or copy and paste it into your browser

if youre using brave browser tell me

[fossil sage]

im using firefox

[fossil sage]

again thats 7851 not 7852

[teal ferry]

Don't go to 7851

Screenshot your terminal

Show me

If you can hit the API at 7851 then you can go to the output at 7852

[teal ferry]

Ok you have a torch issue

We need to uninstall pytorch then reinstall the right version

[viral mason]

if u want to make covers sure, models? don't waste ur time or money on that garbage

[fossil sage]

ok

[teal ferry]

E:\xtts\alltalk_tts\alltalk_environment\env\python.exe -m pip uninstall torch torch vision torchaudio

[fossil sage]

just finished uninstalling it

[teal ferry]

Ok then go

[fossil sage]

reinstall it got it

[teal ferry]

No wait

[fossil sage]

wait wtf it rquires python 3.9 or higher

[teal ferry]

E:\xtts\alltalk_tts\alltalk_environment\env\python.exe -m pip install torch torchaudio torchvision --index-url https://download.pytorch.org/whl/cu128

Wait

[fossil sage]

i have cuda 12.1

[teal ferry]

Ok now it works

Don't worry about it

Just follow instruction

[fossil sage]

now i just gotta wait

[teal ferry]

The python version is specified by your environment

[knotty moth]

doesnt it need 3.11?
and who tf still uses 3.8/older?

[teal ferry]

Cuda may be an issue but it may not 50-50 shot

Yeah 3.9 is really old. But it doesn't matter we're not using the system python

Once torch is done installing close the terminal and start all talk again

Should occupy port 7852 this time and therefore you'll see the link in std output

Also if you have to go to bed then tell me to shut up. It's fine

[fossil sage]

i wthought it meant 3.09

hahaha

not 3.9

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 0/4 [sympy] WARNING: The script isympy.exe is installed in 'E:\xtts\alltalk_tts\alltalk_environment\env\Scripts' which is not on PATH.
Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location.
━━━━━━━━━━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1/4 [torch] WARNING: The scripts torchfrtrace.exe and torchrun.exe are installed in 'E:\xtts\alltalk_tts\alltalk_environment\env\Scripts' which is not on PATH.
Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location.
Successfully installed sympy-1.13.3 torch-2.8.0+cu128 torchaudio-2.8.0+cu128 torchvision-0.23.0+cu128

what wtf

[teal ferry]

tap windows key and type env hit enter

[fossil sage]

wym

[teal ferry]

i dunno how else to say that

use brain

[fossil sage]

ohhhhhhhh

do i edit environment vriables

[teal ferry]

yeah then click path then edit

[fossil sage]

alr

[teal ferry]

system variables so the bottom box

click new

then put
E:\xtts\alltalk_tts\alltalk_environment\env\Scripts

[fossil sage]

and im browing an directory not a file

done

[teal ferry]

click ok and exit ALL of the windows we just opened.

ok three times

close all terminals

then start all talk again

[fossil sage]

👍

wow

[teal ferry]

oh no

ok so this would require editing the source code of a couple of torch files. specifically the wirghts_only variable

there is no way around this because your GPU requires a certain version of torch

we could downgrade torch in theory but im guessing because the last error you got with compute score

it wont work

[fossil sage]

well looks like i should just stick to chatterbox

[teal ferry]

so you can open. E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\TTS\utils\io.py and

[fossil sage]

and i've tried using chatpgt

[teal ferry]

i think thats it.

[fossil sage]

its an infinite loophole

[teal ferry]

this will work

[knotty moth]

is yours RTX 50-series? if so it'd need cuda 12.8 and latest version of torch

otherwise GTX 10-series might need the legacy cuda

[teal ferry]

theres no such thing as an unsolvable problem

[fossil sage]

unfortunately it didn't

[fossil sage]

fyi i am are install alltalk2 tts

[teal ferry]

just open this file

E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\TTS\utils\io.py

or send it to me

[fossil sage]

https://github.com/erew123/alltalk_tts/tree/alltalkbeta

[fossil sage]

yes i have the 5070

i am having the cuda 12.1 cuz i think chatterbox needs it ig

so far its the best out of index tts and vibevoice

[knotty moth]

you need cuda 12.8 for sure

[fossil sage]

oh well i can't remember

chatgpt was messing everything up

[teal ferry]

swap this one in for the one you have

i may have to add another argument

but i think you hit the else statement regardless

else within the if statement*

[fossil sage]

ok time to run it again

ALLL great

[teal ferry]

if you get the same error then use this one. i guess i couldve added the variable regardless to begin with

oh thats a new error

[fossil sage]

yes

[viral mason]

btw @tawny radish how bad are my settings in okada

[tawny radish]

horrible

absolutley

fucking horrible 😭

[viral mason]

gimme good ones

[tawny radish]

ur extra cant be higher then 2.7

[fossil sage]

this is realtime voice changer ?

[tawny radish]

put it at 2.7

ur chunks, idk about that just mess with it

[viral mason]

wokada deiteris

[teal ferry]

send me E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\torch\distributed\elastic\agent\server\api.py

[tawny radish]

mainly i put my index at 0.5

or 0.7

[teal ferry]

actually if loop didnt exit you may be up and running

[ashen patrol]

If 2.7 it's good if 3.5 not great

[teal ferry]

no nvm it did

[tawny radish]

2.7 is the max you shoudl do

never go higher

or youll get problems

[tawny radish]

@fossil sage are u trying to make voicemodels

[viral mason]

what would you recommend?

[tawny radish]

hmm

idk

i dont use forked anymore

i do like 63 chunks but thats bc i have a good gpu

[teal ferry]

wrong one sorry i need this E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\deepspeed\elasticity\elastic_agent.py

[fossil sage]

yes thats the short answer but the logner is that im currently testing all tts inference things see which is best then learning how to train model then going ahead putting it through appoilo and finally having a good voice

[tawny radish]

making good voicemodels isnt really rocket science

[viral mason]

idk what that means tbh

[tawny radish]

so id say your goal would be so easy

[tawny radish]

my graphics card

i have a 4080 super

ur settings vary chunk wise on your GPU

[viral mason]

I meant 63 chunk

[tawny radish]

because ur gpu renders chunks

[fossil sage]

well i somehow manged to do it before in a google collab however someone walked me through a call and holded my hand step by step so ig thats doens't really count

[tawny radish]

dude voicemodels are easier then you think, even if ur doing it yourself its genuinley lightwork

my first voicemodel was even good

the more you do it the more your datasets can improve - also meaning ur voicemodels will sound better

but you dont even need too much expirience to make GOOD voicemodels

id tell you now like male voicemodels are harder then female because female have a more high pitch voice so male ones can be difficult, but not impossible

[fossil sage]

i see

[teal ferry]

hes just trying to stroke his ego

just swap that out n run alltalk again

[fossil sage]

the first one i created my self was a one minute and i couldn't determine whether or not it was good due to not having a inference tts

[tawny radish]

giving general advice means im stroking my ego is crazy 😭

i havent made a voicemodel in a solid 2 months ngl

[tawny radish]

gl though

[fossil sage]

https://www.youtube.com/shorts/znAIPbYftDE

this is the voice i cloned

[tawny radish]

NAH LMAOO

[teal ferry]

no claiming its easy is though

[fossil sage]

i lost the model

[tawny radish]

😭

[teal ferry]

its not

most people make horrific models that vaguely sound like the target

if you call that easy then sure. but its bad

so i would call it a fail

[tawny radish]

when i made my first one it took me like an hour to find the .pth and .index for it

😭

and i overtrained it

so

youll have mistakes but like if ur consistent in learning it youll make good ones in id say a month or two

[teal ferry]

so maybe its not so easy especially dealing with python environments

[fossil sage]

well this going to be a infinite loop hole

[fossil sage]

what does a .index even do

[tawny radish]

i think it changes how strong u want the accent to be

pretty sure it does

[teal ferry]

swap this one in

its gonna work now

[tawny radish]

its actually kinda crucial if u want ur voicemodel to sound more natural or realistic

[fossil sage]

well atleast it was able to detect deepseed

[teal ferry]

E:\xtts\alltalk_tts\alltalk_environment\env\python.exe -m pip uninstall put\the\path\to\deepspeed\whl\here

[fossil sage]

done

[teal ferry]

then either start alltalk or reinstall deepspeed. because we installed new torch deepspeed was invalidated. deepspeed installation uses the version of torch to create itself. so we end up with these two solutions. because you dont need deepspeed i would just start alltalk

[fossil sage]

this sounds pretty good to me

[fossil sage]

[AllTalk Startup]
E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\torch\cuda_init_.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
[AllTalk Model] XTTSv2 Local Loading xttsv2_2.0.2 into cuda
ERROR: Traceback (most recent call last):
File "E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\starlette\routing.py", line 734, in lifespan
async with self.lifespan_context(app) as maybe_state:
File "E:\xtts\alltalk_tts\alltalk_environment\env\Lib\contextlib.py", line 210, in aenter
return await anext(self.gen)
^^^^^^^^^^^^^^^^^^^^^
File "E:\xtts\alltalk_tts\tts_server.py", line 127, in startup_shutdown
await setup()
File "E:\xtts\alltalk_tts\tts_server.py", line 172, in setup
model = await xtts_manual_load_model()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "E:\xtts\alltalk_tts\tts_server.py", line 244, in xtts_manual_load_model
model.load_checkpoint(
File "E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\TTS\tts\models\xtts.py", line 783, in load_checkpoint
self.gpt.init_gpt_for_inference(kv_cache=self.args.kv_cache, use_deepspeed=use_deepspeed)
File "E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\TTS\tts\layers\xtts\gpt.py", line 222, in init_gpt_for_inference
import deepspeed
ModuleNotFoundError: No module named 'deepspeed'

ERROR: Application startup failed. Exiting.
[AllTalk Startup] Warning TTS Subprocess has NOT

[teal ferry]

omg he did

ok install it again

[fossil sage]

[teal ferry]

id assume he had a good reason to exit the loop

but i dont know

[fossil sage]

E:\xtts\alltalk_tts\alltalk_environment\env\python.exe -m pip install C:\Users\yonshuk\Downloads/deepspeed-0.14.0+ce78a63-cp311-cp311-win_amd64.whl

[teal ferry]

yes

[fossil sage]

well imma run it again

and theres a 1 percent chance its gonna work

[teal ferry]

100%

[fossil sage]

[teal ferry]

swap this

[fossil sage]

yep and chatgpt is just telling me to go thorugh a infinite rabbit hole

[teal ferry]

ignore chaptgpt

humans are the superior intelligence

[teal ferry]

where did you get that version of deepspeed

[fossil sage]

https://github.com/erew123/alltalk_tts/releases/tag/DeepSpeed-14.0

from the github you sent me

[teal ferry]

uninstall it

its wrong for wrong version of cuda

[fossil sage]

i have cuda 12.1

and thats the one i installed

[teal ferry]

yeah but it has to work with torch to

o

[fossil sage]

and i already uninstalled it breh

[teal ferry]

E:\xtts\alltalk_tts\alltalk_environment\env\python.exe -m pip install deepspeed

[fossil sage]

alright now whre do i download it from

[teal ferry]

you dont

just input that

were using pypi

[fossil sage]

i just did

[teal ferry]

this may trickle down to a python version issue

ok when its done start alltalk

assuming you got no errors

[fossil sage]

errors usal

at this point im just do the steps back and foward back to

back

[teal ferry]

E:\xtts\alltalk_tts\alltalk_environment\env\python.exe -m pip uninstall deepspeed

then

E:\xtts\alltalk_tts\alltalk_environment\env\python.exe -m pip install deepspeed --no-cache-dir

[fossil sage]

says i need to upgrade cuda

[teal ferry]

yep

[fossil sage]

imma just install 12.8

[teal ferry]

https://developer.nvidia.com/cuda-12-8-0-download-archive?target_os=Windows&target_arch=x86_64&target_version=11&target_type=exe_local

the only thing is you need to remove old env variables otherwise it will always grab the old version.

i beleive its just these three. make sure they say 12.8 or above

[fossil sage]

basiclly the same thing i was gonna download expect its local

[teal ferry]

yeah either one

[fossil sage]

i should have created a system restore point

[teal ferry]

not necessary were not doing anything to core system files. but i mean its not going to hurt if you did

you should just reinstall alltalk after that. otherwise wed have to reinstall torch, deepspeed, and possible edit deepspeed source code again

well maybe not edit it because we install frpm latest

[fossil sage]

alright just installed it

[teal ferry]

i mean it should work like that but maybe deleting 12.1 for redundancy

[fossil sage]

ok

[AllTalk Startup] AllTalk Settings & Documentation: http://127.0.0.1:7851
[AllTalk Startup]
E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\torch\cuda_init_.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
import pynvml # type: ignore[import]
[AllTalk Model] XTTSv2 Local Loading xttsv2_2.0.2 into cuda
ERROR: Traceback (most recent call last):
File "E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\starlette\routing.py", line 734, in lifespan
async with self.lifespan_context(app) as maybe_state:
File "E:\xtts\alltalk_tts\alltalk_environment\env\Lib\contextlib.py", line 210, in aenter
return await anext(self.gen)
^^^^^^^^^^^^^^^^^^^^^
File "E:\xtts\alltalk_tts\tts_server.py", line 127, in startup_shutdown
await setup()
File "E:\xtts\alltalk_tts\tts_server.py", line 172, in setup
model = await xtts_manual_load_model()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "E:\xtts\alltalk_tts\tts_server.py", line 244, in xtts_manual_load_model
model.load_checkpoint(
File "E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\TTS\tts\models\xtts.py", line 783, in load_checkpoint
self.gpt.init_gpt_for_inference(kv_cache=self.args.kv_cache, use_deepspeed=use_deepspeed)
File "E:\xtts\alltalk_tts\alltalk_environment\env\Lib\site-packages\TTS\tts\layers\xtts\gpt.py", line 222, in init_gpt_for_inference
import deepspeed
ModuleNotFoundError: No module named 'deepspeed'

ERROR: Application startup failed. Exiting.

[teal ferry]

yeah install deepspeed again

not the whl

[fossil sage]

bro i already did like 3 times

[teal ferry]

i know you have to do it again because all of it hinges on cuda

if you install them with the wrong version of cuda then it wont work

[fossil sage]

great now chatterbox doesn't work

anyways i gotta go to bed

[teal ferry]

youre close, alright

[tawny radish]

not even bad ngl

[simple ore]

you dont need to install cuda toolkit unless you actually developing stuff

[heady plover]

Trying to use the tg develop branch or any branch of w-okada's vc for that matter and all I get is this continuous tapping sound that sounds like the voice. On the Linux vers, does anyone know why this is happening

(3060 12gb)

[edgy minnow]

whats the replacement for okada w?

[viral mason]

could u share a screenshot of what the program looks like

btw @tawny radish is it cool if I add u if I haven't already

you seem chill

o

[tawny radish]

Yeah sure lol

Idm

[viral mason]

did u get it from a yt link

[knotty moth]

!give-media-perms 30m @median quiver

[viral mason]

sent!

[tawny radish]

Put ur extra to 2.7

[viral mason]

first human being ever to get the voice changer from a credible source ^

[tawny radish]

Put ur input as ur mic and ur output as line 1 vb-audiocable

If u havent already download VAC

[viral mason]

these settings work pretty good for my gpu but I have nvidia

[tawny radish]

Ngl if ur thing is still green when u start it u can lower chunks even more

[viral mason]

the lowered delay is neat but still working on getting it lower without it doing anything funky

[knotty moth]

you can lower delay as long the perf stays green

but it may depend on if you're playing a kind of game/another application

[viral mason]

I'm trying to find the line between it still sounding fine with no cutting out or choppiness like that with as little delay as possible

[analog obsidian]

one important thing to keep in mind is that rvc is context based, lower chunks decrease delay yea but also decreases the context of the audio, and if its too low, the model is going to start to have very bad pronunciation

rvc cant predict what are you gonna say, so it actually needs that delay in order to properly say the words

lower extra decreases the delay and but also decreases context of the audio

[onyx badge]

anyone have this issue where after some time VAC lite just stops working

and I ned to reinstall it

[stiff idol]

Hi guys. I'd like to ask how do people fix weird glitch in a voice and what tool to use? I was listening to dataset I'm preparing, but the voice suddenly pitched into a robotic-like voice.

I was looking on some YouTube videos and tried some functions in audacity I know if, but wasn't successful.

[hallow thistle]

In REAPER, there's a built-in VST for tuning pitch of an audio track, named ReaTune. This one also has an option to do autotune-like effect, as in this screenshot. It's simple, but not quite the same level as other free and paid plug-ins.

[hallow thistle]

I have Virtual Audio Cable lite 4.70 installed on my laptop's hard drive, and I never get any error. The cause could either be the program was installed improperly, ran out of memory or still have older version. How is it going?

[golden walrus]

Hmmmm. Guys, can i ask how to improve my model volume? It sounds so small

[spare tree]

guys any tips to lower ai voice changer delay without lowering quality too much?

[spare tree]

guys I cant find my gpu in the gpu dropdown, I can only find "cpu"

[simple ore]

what did you download?

what's your gpu?

[spare tree]

as in voice model or app

[spare tree]

AMD Radeon (TM) Graphics

[simple ore]

what file did you download

iGPU is not a real GPU, so at best you can run the VC on CPU

[short belfry]

hey yall ive just got the w okada voice changer, and when im trying to use it its cutting off every 500ms or so then continuing with the sentence. id love a way to figure this out! im using an AMD GPU, using rmvpe_onnx

[spare tree]

its extremely slow tho

are there any gpus i can download?

[simple ore]

yes, same place where you get extra RAM

[short belfry]

anyone know the solution to this <3

[spare tree]

oh i did just get a new laptop

is it a link to go to

[spare tree]

uhuh

[ruby idol]

hey can someone help me set up MMVCServerSIO

[stiff idol]

Thanks 👍 I'll check that out, I had that sound removed through audacity entirely for now, so I'll get back to it when I get home.

[strange grail]

why was codename fork removed from docs

[thorn narwhal]

how to stop echo

[ruby idol]

how can i fix the delay?

[fervent orbit]

Is there any n8n automation experts that can help me? I need to do an n8n workflow for real estates

[thin anchor]

where to dowload voice changer

i need free one plz 🤑

[viscid topaz]

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Vonovox

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

• Wokada Tg-Develop Fork

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

• Wokada Deiteris Fork

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

⚔️ Wokada Deiteris Fork vs Vonovox

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

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

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

• RVC GUI Mainline Realtime

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

[winter seal]

ts happened while tryting to use model in applio noui thing how do i fixit

[fast yoke]

can you help me coding?

[edgy rapids]

why is my voice changer is so delay?

[tawny radish]

which version are you using?

[bleak smelt]

am i able to use any of these for rvc training? (on runpod) or does anyone know a workable one on there?

[simple ore]

are you trying to use D or G file for inference?

[viral mason]

trying out codename's fork since applio is a shitshow, what is this

should I leave it

[potent bone]

why did it get removed from docs

[viral mason]

why did what get removed?

[potent bone]

codenane fork

[viral mason]

applio or codename fork

idk

[potent bone]

where can i download it from

you know where?

[viral mason]

I have the one that works for kaggle I guess

[thick latch]

Is this any useful to finetune the voice?

[bleak smelt]

does this not work anymore?

[viral mason]

what even is that

[viral mason]

what site is this? I can't really help with the settings but it looks like Vonovox

[thick latch]

It is vonovox, I just clicked advanced settings

[viral mason]

I'm not sure if this is kept up anymore as in it being maintained or updated

I'd ask around maybe Nick or Lista may know

[bleak smelt]

is there anything similar on colab i can use?

[viral mason]

you could check these but other than that I have no real knowledge of the difference gcollabs out there for cover making

-collab

uh

• collab

-colab

[patent trellis]

📒 Google Colab Notebooks

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

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

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

by Jarredou & Makidanye
Google Colab

[viral mason]

I spelled it wrong twice lol

[rain nova]

yo can anyone help with the voice changer download

[winter seal]

idk no maybe

[viral mason]

what gpu do u have

Nvidia, AMD, or Intel

[rain nova]

nvidia

2060rtx

[viral mason]

u can use any of these but I recommend Vonovox as it's the best

-rt

[patent trellis]

💻 Local Realtime RVC

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

• Vonovox

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

• Wokada Tg-Develop Fork

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

• Wokada Deiteris Fork

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

⚔️ Wokada Deiteris Fork vs Vonovox

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

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

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

• RVC GUI Mainline Realtime

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

[viral mason]

read the guide and download the two things, Vonovox and Vac lite

it's the first one

[zinc thunder]

is there a way to setting parameters on Applio when the original voice on a song have trembling (example: hinorobu kageyama) and the AI voice not? every time i try to make a cover, the result is not good, because the "new" voice came like he is trying to make funny noises with his voice

[viral mason]

what audio are you using to make the cover with?

[zinc thunder]

can i post a youtube link with the song? if not, the song is "CARELESS WHISPERS (Japanese Version 1984) - Hideki Saijo". and yes, of course i removed the melody, and only used "acapella" XD

[royal kettle]

Unless you know what youre doing you shouldnt really touch it.

[viral mason]

yo @simple ore sorry for the ping but was wondering if this also did the double batch thing if settings are left like this

[viral mason]

probably not, but u can post the acapella/isolated vocals

make sure u removed the reverb and echo as well

[zinc thunder]

uploading on mega, for example?

yep, i have a version without reberb (mostly of it) that i did through davinci

[viral mason]

what is davinci

[zinc thunder]

oh boy. sorry XD. heavy rain here. sadly im going to leave. but thanks for trying to help! oh, davinci resolve is a tool to make videos, similar to adobre premiere, kden live etc

[viral mason]

ahh ok

btw before u go I'd recommend using this for cleaning audio in the future
https://colab.research.google.com/github/Eddycrack864/UVR5-NO-UI/blob/main/UVR5_NO_UI.ipynb?authuser=1#scrollTo=gmjUWmz8iecd

[stiff idol]

I triedi to ut but can't evne open it because "google logged me out on a different tab"

time for good old CHrome

Does Mel work in UVR app?

oh yeah, it's for UVR 🤦‍♂️ I only don't know where the models are stored

I also didn't see any of the preinstalled models that were supposed to be in UVR, I guess the guide doesn't have the newest info on UVR

[viral mason]

wdym

[stiff idol]

I went back to the main repo of collab

fixed*

[low shard]

Not sure why you're calling applio a shitshow, codename's fork is an applio fork, codename's fork is even more experimental and has caused model exploding before as said by Lyery and Noobies #🔥│model-maker-chat message, #🔥│model-maker-chat message

[viral mason]

due to the broken mess it is normally on kaggle

as of now

[low shard]

That's not an issue related to Applio at all, Ngrok's free tier has lowered the requests per min

You can just use the lightning.ai notebook of applio

[viral mason]

that's so confusing and I had issues with it

so I prefer kaggle

I got the tensorboard to open with gradio but applio won't open in lightning just a blank page

[low shard]

It seems like you could say the same about codename's fork overall since you're asking Noobies about some experimental settings lol

Anyways, I was just warning you that if you mess up with it, especially with settings you don't know, it could just make things worse, I mean your choice tho

[low shard]

Can you show me a screenrecording of the issue? I don't recall having that issues with my notebook, I could check them up if you want

[viral mason]

just brought me to a blank page, it did some loading thing and then nothing

hold on the image has my name in it gotta edit it

[low shard]

yeah that doesn't look like my notebook, it looks like you uploaded manually some notebook file that someone gave you

[viral mason]

the actual notebook I'm not focused on in that ss tho

this is "applio" as it calls it

an empty page

I used this

pretty sure it was the cat guy

I forgot his name

[carmine siren]

--kaggle

-kaggle

[patent trellis]

📘 Kaggle Notebooks

Kaggle is a Cloud (Remote Good PC) Service that offers 30 hours of GPU weekly, but needs a phone number verification

• **Applio Notebook**

by IAHispano
Kaggle

• **Hina Mod Original Wokada**

by Hina
Kaggle

• **Wokada Deiteris Fork**

by Hina & Deiteris
Kaggle

• **UVR5 UI**

by Eddy, ArisDev & Nick088
Kaggle

• **UVR5 NO UI**

by Eddy
Kaggle

• **RVC AI Cover Maker UI**

by Shirou & ArisDev
Kaggle

• **Music Source Separation**

by Shirou
Kaggle

[thorny musk]

what is the best training tool for SD images?

[viral mason]

anyways does it do the x2 batch size thing with this?

[low shard]

yup that's not mine, it's some very simple notebook made by Vidal it seems

Please try mine, I recently updated to automatically use Gradio, so you just need to run the cells and that's all: https://docs.aihub.gg/rvc/cloud/applio-lightning-ai/#studio-setup--installation

[analog obsidian]

if you have 1 gpu just set the batch size to the value you want

theres two gpu there so you know what do to lol

[viral mason]

so there is two gpus there currently? I'm just clarifying because I can be really slow sometime

[analog obsidian]

0 and 1 yes two gpus

[viral mason]

ok ty

[solid pumice]

what settings do i have to apply if my microphone aint it

[low shard]

!howtoask

[patent trellis]

❓ How to Ask for Help

✅ Before You Ask!

1. Check Docs & Guides: Your answer may already be in the AI Hub Docs or the https://discord.com/channels/1159260121998827560/1159513888199540817 channel.
2. Search the https://discord.com/channels/1159260121998827560/1192011222023950368 : Look for existing posts that solve your issue. Do not invade someone else's post.

📝 How to ask?

Tell your:

• Full GPU Name: (e.g., NVIDIA RTX 3060)
• Operating System: (e.g., Windows 11)
• Detailed Description: What were you trying to do and what went wrong?
• Tutorial Used: Link to the guide you were following.
• Screenshot: A picture of the full error message is very helpful.

🚫 Prohibited Topics (We Will NOT Help With These)

To maintain a lega, safe & ethical community, we will NOT provide help for:

• (E girl, as an example) catfishing/trolling, scamming, impersonation.
• NSFW/Porn.
• Any illegal activities.
  Requests for these topics will be ignored and may result in moderation action.

<:matsuripray:1159685390156967936> Community Expectations

• Be Polite & Patient: Our helpers are volunteers. You may ping the Helpers role once.
• English Only: Please keep all conversations in English.

[solid pumice]

thanks

[low shard]

It's a discord bot command to help users elaborate and understand how to ask for help (and what to not ask for help), please read it up so helpers know how to help you

[solid pumice]

ty❤️

[viral mason]

just realized I wasn't actually using this, it was the one u gave me in the guide

got stuff mixed up

but this is the one that gives me that issue after running it

the blank page

nvm it works fine now

😭

can't the same code for lightning ai be ported over to fit Kaggle

@low shard

[stiff idol]

Guys, does anyone have a way to download an exe file from UVR repository? I need this one: https://github.com/Anjok07/ultimatevocalremovergui/releases/download/v5.6/UVR_1_15_25_22_30_BETA_full.exe

so github has a shortlived links which don't alllow me to download it because it strangely fails

[tawny radish]

are u trying to get beta? or just regular UVR

i have it saved somewhere in my pc

[stiff idol]

I'm trying to get the beta with mel Roformer

[low shard]

see

[viral mason]

idk what I did the first time to mess it up lmao

[low shard]

nop, kaggle's built in file browser isn't as good as google colab or lightning.ais one, I asked vidal to just use gradio for Applio and remove filebrowser, but even tho most things can be done by the Applio UI, he wants to keep filebrowser for little adjustments soo

like I could use Gradio for Applio Kaggle, but it would work for only the Applio UI, not the filebrowser that is kind of needed (even tho not a very big necessity) for kaggle

[viral mason]

I mean all that's needed for applio is just applio and tensorboard

so if there's code that allows those two things to load with no issues

[low shard]

#✨│ai-help message

you can check that convo yourself if u wanna

[viral mason]

ngrok causes issues yea but just swap it for gradio or zgrok or anything but ngork

would that work orrrr

[low shard]

yeah I tested it, Gradio does now (it has been recently fixed, as previously it didn't) work on Kaggle, I could use it with Applio UI but I don't think I can pull request a gradio tunnel option without including using a secondary tunnel for the filebrowser as @nocturne mural said

[viral mason]

so in theory it could work but it'd be complicated?

[low shard]

If I would have to include filebrowser for the gradio tunnel: yes, because i'd have to use another tunnel for filebrowser since filetunnel is not made with gradio

If i would not use filebrowser (atleast not when using Gradio): technically it would be easier. EDIT: well not as easy, because vidal pointed out that the tensorboard would need a secondary tunnel too

tbh just learn applio lightning.ai, this issue and pull request seems like it will take some time, I'm also checking for other free tunnel alternatives too

cloud isn't as stable as local, it's normal to sometimes switch and modify things

[nocturne mural]

Don't forget about TensorBoard, which you can't visualize in Kaggle.

[low shard]

my bad, you're right

[viral mason]

well Kaggle is dead then

no use left for it

only due to Ngrok

screw them

a reasonable lowering from 4k would at least be like max 1k

[low shard]

not really, you could always pay for ngrok's paid plan

[viral mason]

not whatever tiny number they changed it too

[viral mason]

I'd rather jump off a moving vehicle

I'm getting the true free user experience by never paying these services for their greed

[nocturne mural]

Not entirely. You can use Zrok or Localtunnel for Tensorboard or Filebrowser.

uh

[viral mason]

I dunno how to change code I'm stupid

[low shard]

by statistics, that will end up badly, so not suggested

[low shard]

should I add a Gradio option, then use Localtunnel for the Tensorboard and Filebrowser?

I could commit that right quick

I was actually looking for other possible good free tunnels but that might take a while, doing as said above would fix the issue faster i guess

[nocturne mural]

https://raw.githubusercontent.com/wandaweb/jupyter-webui-tunneling/main/pinggy.py

[viral mason]

trust I'm like a sponge I'll absorb the impact even tho I'm built like a skeleton

[low shard]

pinggy tunnels have a 60 minutes limit, which isn't that good for especially training

[brittle wing]

hello

im using the uvr v5 hugging face ui

which model do i use to get the instrumental to not sound like shit

im seperating an instrumental from vocals

[nocturne mural]

I didn’t understand what you said, but assuming I know what you’re referring to, then yes, it would just be a matter of adding the gradio option to the command, and simply adding a

match method:
  case 'localtunnel':
    #filebrowser
    #tensorboard
  case 'zrok':
    #filebrowser
    #tensorboard

!python app.py --share --listen```

[viral mason]

use gabox FV4

[nocturne mural]

Maybe we could add a timer so we can create another tunnel when the time is up?

[brittle wing]

i ran out of gpu quota while trying 😭

[viral mason]

use thishttps://colab.research.google.com/github/Eddycrack864/UVR5-NO-UI/blob/main/UVR5_NO_UI.ipynb?authuser=1#scrollTo=gmjUWmz8iecd

same thing but on google collab

[dense cave]

anyone familer with the lightweight text to video generaion model ?

[viral mason]

@low shard how do I get this to not be impossible to use

if I open tensorboard at all it fucks everything over as I have no way of returning to the lightning space without it stopping and ruining everything

[simple ore]

that is a very old one

[stiff idol]

so I downloaded a bad one? WHich is the newer one?

[simple ore]

wdym double batch size?

[stiff idol]

seems like the resources aren't up to date or I search wrong

[simple ore]

https://github.com/Eddycrack864/UVR5-UI/releases/tag/v1.8.4

[stiff idol]

Where do you get the info on these things from?

[viral mason]

in kaggle it comes with 2 gpus but I'm not using that anymore I gave up and went to applio lightning ai

[stiff idol]

I guess time to uninstall UVR and install it again

[simple ore]

if you have two gpus under advanced settings, it will use both, so batch size doubles

[ashen patrol]

I thought your pfp was a black and white version of Raven from ten titans go

[stiff idol]

Does the collab notebook not work with local files?

[fleet marsh]

hello, i was just gonna ask, does anyone know if this page is from here?https://www.vocalize.fm/voices

they are using a paid suscription to use ai models, some of those are created by me, i dont want them to use my models and getting paid for it, does anyone know how could i make them take down my models or smth?

[fleet marsh]

i think they are taken from here, because they dont use my personal models

[viral mason]

It's my oc :(

[viral mason]

Wdym local files

[stiff idol]

I wnat to upload a wav, what else

🤣

[viral mason]

Oh lol

Make a folder in your Google drive called Vocales, and make another one with whatever name you want, copy the path of the folder u put the audio you want to clean in (not Vocales)

[stiff idol]

you mean to create a folder in the root of the google drive or the folder in the root of the Collab Notebook folder?

[viral mason]

In your own Google drive

Not in the collab

[stiff idol]

ok 👍 thanks

[fleet marsh]

Atleast does anyone know where should i report this?

[viral mason]

Not sure tbh

[fleet marsh]

Oh ok thanks

[viral mason]

Maybe could sue them?

But idk

[fleet marsh]

Idk how

[viral mason]

Me either

Maybe look it up

[viral mason]

Does this use any of my models?

[simple ore]

you can only infer with small voice models, D/G are training weights, although G has a small model inside, but there's no config,so the application does not know how to set up the model

[analog obsidian]

imagine if applio had the extract small model from the g file feature, like mainline

[fleet marsh]

Idk, go check

They probably do

They got my hollow knight models even when those models are not that famous

Even the one of the knight

[analog obsidian]

isn't weights also technically stealing our models and profit of them?

edit: grey area, read weights TOS

[viral mason]

I see a bunch of Ov2 models which are definitely shit

Dead pretrain

[analog obsidian]

calling them 'our' models is uhh not correct unless we voiced such characters and had the legal rights to create models based off them

[viral mason]

Technically you can make covers for free

Which is the only function of the voice models on weights besides like tts

[analog obsidian]

what weights is doing is also illegal so

edit: grey area

[viral mason]

🤷‍♀️

[analog obsidian]

or i think it's gray area

[viral mason]

At least it tells you it uploads the models you make

When u post them here

[fleet marsh]

Yeah

[viral mason]

This random website I never heard of doesn't

So it's worse

[analog obsidian]

both are the same thing tho

[viral mason]

One looks worse tho

And isn't popular or known

[fleet marsh]

And does without consent

[analog obsidian]

isnt the weights bot doing the same exact thing

[viral mason]

But it at least tells us about it

And we know it does that

[fleet marsh]

What was weights? I don't remember

[analog obsidian]

random corpo that bought this server

[viral mason]

Used to be a good website to do covers and stuff but now it's shit Bec greed

Their greed has literally destroyed them

It was better in 2023

[analog obsidian]

eh i also dislike weights but you gotta understand gpus cost money

[viral mason]

I get that but they did a lot of bad shit especially not telling people about updates randomly

Just running the users over with a sudden change like hitting a deer you couldn't see with a truck

[fleet marsh]

What dont people use applio?

[analog obsidian]

i read their tos, very interesting, so it's not illegal because its the people that are posting the models, not weights themselves

quite smart

[fleet marsh]

But i didnt choose to post My models on the other website, so is that illegal?

[viral mason]

Because you can use weights without removing instrumentals and vocals separately just import any song as is and it can make a decent cover

[analog obsidian]

technically speaking, the other site IS doing something illegal because its selling copyrighted material

[fleet marsh]

Bruhhhh

[viral mason]

Only site that does that I think

That I know of

[fleet marsh]

People are dumb, they could do it for free and use Audacity to put it togheter

[viral mason]

Wdym audacity

Ohh

[analog obsidian]

weights in the other hand is very grey area from what i read

[viral mason]

Nvm

That's interesting

[analog obsidian]

also, just saying, you only have the rights of your model only if is either: your voice or have the legal rights of that voice

if not, its not yours

[viral mason]

So if weights gets in trouble we're to blame

[analog obsidian]

weights is grey area so yea they cant get in trouble

unlike the other site which is actually selling the models

[viral mason]

What about the users

[analog obsidian]

according to their tos, the users can actually get in trouble if they upload copyrighted material

[viral mason]

So like every single model in existence

[stiff idol]

Does anyone have a direct link for the repository with this model? mel_band_roformer_denoise_debleed_gabox.ckpt I can't find it on huggingface. UVR downloader acts like a github but worse, it doesn't download the whole file (using curl).

[fossil sage]

@teal ferry im back

[analog obsidian]

From what I understand, they only take it down if the company that owns the rights to the voice files a DMCA claim

[fleet marsh]

I don't think so, i don't think fnf models have copyright, but im not sure

[analog obsidian]

they do have copyright

[viral mason]

Ah ok

[analog obsidian]

but since there's no dmca claim, they're still there

unless scott just fills a dmca

[fleet marsh]

What about the ones that are just synths?

[viral mason]

Like Miku or Teto?

[analog obsidian]

it has to be 100% original made by you

or having the rights of that thing

you don't have the rights? then its not yours

even if its just a piano model lets say

[fleet marsh]

Like zardy or tabi

From fnf mods

[viral mason]

Pretty sure this applies to all fnf models

[fleet marsh]

Oh i ser

[analog obsidian]

weights was smart

but the other site is clearly dumb because they're uploading the models themselves

[fleet marsh]

So we can sue them?

[analog obsidian]

it wont last long before they get sued lol

[analog obsidian]

only if you have the legal rights of the stolen model

[stiff idol]

here is the json file for anyone having a problem downloading through that interface https://huggingface.co/spaces/TheStinger/UVR5_UI/blob/main/assets/models.json

[analog obsidian]

denoise debleed? just use the denoise aggressive

[stiff idol]

I'm not sure what they do.

[viral mason]

I have stopped De-noising my models unless they have a really bad mic

[stiff idol]

I've just come around a document that explains a bit.

[analog obsidian]

well, it removes noise

[stiff idol]

I mean debleed

[analog obsidian]

removes instrumental residuals iirc

[stiff idol]

there's lots of new terms for me

thanks

[fleet marsh]

So only weights can do that or how can i get that?

[analog obsidian]

weights don't have the legal rights of the models either

[fleet marsh]

Who does?

[stiff idol]

it never downloads the whole model, the model has 913MB< what do you guys usually use?

[analog obsidian]

for example, the hatsune miku model, the legal owner of that is crypton, they have the legal rights to sue that site

not the person who made the model

actually the person who made the miku model is also breaking crypton TOS

[analog obsidian]

https://colab.research.google.com/github/jarredou/Music-Source-Separation-Training-Colab-Inference/blob/main/Music_Source_Separation_Training_(Colab_Inference).ipynb

i like the colab

[stiff idol]

I'll use local for now.

[analog obsidian]

1 sec lemme find the links

[fleet marsh]

And what about actual voice actors?

[analog obsidian]

they can too

you own your voice (legally speaking)

xD

[analog obsidian]

https://github.com/Anjok07/ultimatevocalremovergui/releases/download/v5.6/UVR_1_15_25_22_30_BETA_full.exe first download this

[stiff idol]

that's an old one btw

[analog obsidian]

after you install that, install the patch https://github.com/TRvlvr/model_repo/releases/download/uvr_update_patches/UVR_Patch_1_21_25_2_28_BETA_small_rofo.exe

[stiff idol]

1.8.4 is the newest

[analog obsidian]

no?

[stiff idol]

wait

[analog obsidian]

thats the official dl link

if u have something else then is custom

i do not trust forks

[stiff idol]

this is so confusing, I got the version 5.6, then removed it, then got the 5.6.1 (supports former models) btu then removed it for the web one

[analog obsidian]

i'd recommend trying what i sent (install them in order)

it's from the person who made uvr gui

https://huggingface.co/GaboxR67/MelBandRoformers/tree/main/melbandroformers/karaoke

this is gabox's official download link of his mel rofo karaoke

[fleet marsh]

If the page por weights gets sued for a model i made, do i also get sued?

[analog obsidian]

nope, only them since they're the ones selling it

you just made it for fun

[fleet marsh]

Oh ok

[stiff idol]

I'll take a look tomorrow at it.

I wish there was actually a repository with the links.

[analog obsidian]

me too lol

its veery hidden in a server

[stiff idol]

Is there a way to know what is official and what not? actually just checking the author... welp

[analog obsidian]

trvlvr is the author of uvr gui

hes also known as anjok

[stiff idol]

Who is this nomadkaraoke? https://huggingface.co/spaces/TheStinger/UVR5_UI/blob/main/assets/models.json

[analog obsidian]

ai hub moment

xD

thats from illaria and eddy i think

[stiff idol]

I'm downloading the gaboke's model from there

Gabox's*

[analog obsidian]

yea so i always recommend using the official stuff first

forks later

the colab i sent is great tho, gets updated quite often

[analog obsidian]

but locally im using this, it works, that matters lol

[stiff idol]

I haven't checked what is the difference between .onnx and .ckpt models

so they are like safetensors... btu this many formats for the same thing

I heard Okada software converts ckpt to safetensor

[analog obsidian]

the deiteris fork converts them to .safetensors for... some reason, idk why lol

.onnx is an alternative for .pth files, in amd gpu systems is faster

but in nvidia gpu, onnx actually uses more cpu and potentially slower performance/higher delay

[stiff idol]

Is there anything I can read on this?

actually I'm missing something comprehensible, it's either too shallow or an academic paper on a whole algorithm

[analog obsidian]

.safetensors is an alternative for .pth/ckpt to make them safer and prevent them from executing arbitrary code

since you can actually inject code into .pth/ckpt, making them virus

[stiff idol]

can safetensors be trained too? actually never looked into that too much

[analog obsidian]

no idea.. i know in the sd community everyone uses .safetensors because is safer

ive heard is also faster than pth/ckpt

in rvc no one uses it, maybe because applio and w-okada already have a protection against arbitrary code

[stiff idol]

I read that ckpt can be trained with the config and have an executable code that can infect you, same with .pth, safetensors are safe because they omit the code and have only weights

[analog obsidian]

yeah exactly

in applio there's a line named "weights only"... pretty self explanatory what this does

deiteris fork also has it, no idea about regular w-okada

[stiff idol]

I can never look at those error messages in cmd, I see "separation failed" but it's still ongoing. My notebook is very sensitive to any process. I know when it is running and when it is not by sound.

if you stumble upon any repository that has the official links to the models or their creators I'd be glad if you shared them

[analog obsidian]

for the separation models?

[stiff idol]

yeah

[analog obsidian]

google "audio separation discord"

join the server

[stiff idol]

or dereverb / denoise

[analog obsidian]

everything it's there

the authors, colab links, etc etc

[stiff idol]

gotta go, good night, see u tomorrow

[analog obsidian]

[fossil sage]

whats better Local Eddy's UVR5 UI or google collab uvr

[viral mason]

@low shard in case u didn't see this

I figured it out nvm

[low shard]

I just trained (yes JUST started on CPU bc I don't wanna waste GPU free credits and its a test), the tensorboard seems to load without crashing, try to just wait a bit more

also you can click Open to open it in a new tab

[viral mason]

nah I mean it loads fine but switching back from that particular page screws it up if u wanna go back to see training progress, figured it out tho just clicking the open button, then copy the link to a new tab on my browser

[low shard]

I just tried to switch a few times between Jupyter and Tensorboard without using the open button, it's all fine you just have to wait some seconds between switching, tho using the Open button is more comfortable

Well goodluck staying on lightning.ai

[viral mason]

that's cool, I think the method I used is gonna work fine

I've basically already figured out how to do what I need

[low shard]

let's see to which cloud platform you will switch to when lightning.ai won't be enough for ya in 6 months

[viral mason]

6 months is good enough

[low shard]

It was a joke but I'm just suggesting to get used to adapt or switch to new things since you won't be using the same thing in the next years

are you even gonna use rvc in like 2 years?

i mean who knows what will happen in 2 years lol

[viral mason]

I'll be using it until I either lose interest or it fully is only usable locally

[low shard]

I'm guessing the 1st would happen sooner than the 2nd hopefully

it will always be usable on cloud

there are a lot of renting gpu services

[viral mason]

my tensorboard did a loopdy loop

actually just noticed at the start too it had a stroke

never seen this before

[low shard]

what graph are u checking and wtf

[viral mason]

hopefully neither do, but I'll change at some point or get so traumatized by rvc it'll drive me away from it like everything I dislike has

[viral mason]

not just this one all of them look like that lmao

[low shard]

maybe something got fucked up while you had that random crash, might be better to restart

[viral mason]

nah it's fine

probably all that starting and stopping nonsense from me not understanding how to use it properly

btw can lightning run out of space in the notebook like kaggle?

[fossil sage]

is it just me or this not clear enough for cloning

[karmic copper]

is someone else having problems training a model??
I haven't trained a model for months (means I didn't reach the weekly limit) but It still says I need to upgrade my acc(purchase required) to train my model??

[simple ore]

you've stopped training after an epoch save + some steps, then you resumed, that logged double values

[heady plover]

Still having the same issue from before, this is what is sounds like

it worked fine before nvidia drivers updated and whatnot, and im not sure what else to do beyond using the install script/changing the env version to install newer packages

I'll forward this to forum too but I'm stumped myself.

[viral mason]

What's the best Nvidia GPU I could upgrade to, I want to replace my 1660

[simple ore]

best and not overly expensive would be 5070ti

well, 5060ti is also an option, comparing to 1660 it is a beast

[viral mason]

thx

[viral mason]

https://tenor.com/view/clock-watching-ticking-time-is-ticking-time-is-now-gif-11174641

[fossil sage]

guys im kinda of confused on what he applio built in normalization does

[jaunty minnow]

is there a way to run deiteris okada's as an application rather it opening a google chrome page?

[viral mason]

just use post normalization, it's the best option

idk what it really does

[fossil sage]

So when I train the model that's what I click correct

So far I have a clean 31 minute Audio that I'm ready to train

[analog obsidian]

as you know rvc doesn't train the whole dataset .wav file at once, instead, it slices the dataset into 3s segments with overlap
post normalization applies a normalization to those generated segments

helps the model building frequencies

it's a setting that is enabled by default in mainline, but somehow it's off by default in applio

[fossil sage]

So it's basically like this some of the audio recorded in multiple parts and in those multiple parts the loudness of the audio is different therefore and normalizes the audio for those multiple parts individually rather than normalizing the whole track by itself so it can have a 0DB audio?

I'm confused

[analog obsidian]

there's no point in normalizing the whole audio, rvc doesn't load the whole audio at once

it loads small segments of 3s

[fossil sage]

And within does 3 segments they are normalized individually ?

[analog obsidian]

yes, if you select post norm they get normalized

[fossil sage]

Oh ok I was scared for a sec

[analog obsidian]

once the model loads a slice, instead of "reading" it in a single pass, it will start by reading 0.35s segments until it has gone through the entire file

thats how the model learns the timbre and characteristics of the dataset voice

[fossil sage]

what do you think of this audi

i used Eddycrack864 uvr locally

[analog obsidian]

it's bad, 24k audio, mp3 quality

[fossil sage]

discord max file size limit 100mb

for this server

[analog obsidian]

do u have spek?

[fossil sage]

yes i do

[analog obsidian]

can u load the uncompressed audio so i can see it

just take a screenshot

[fossil sage]

sure

i don't really know whats going on

i read the docs aihub on vocal seperation

[analog obsidian]

oh yea that still looks like 24k to me

everything above 12k is just noise

voice is around 22050

22k/24k

[fossil sage]

idk what that is but ok

[analog obsidian]

idk how rvc will react to a 22k dataset tho

minimum it needs a 32k one

[fossil sage]

💀

the audio is extracted from a show

[analog obsidian]

it should still work, the model in theory will learn the frequency cutoff

[fossil sage]

i exported it 16 bit btw not 32 bit wav

[analog obsidian]

just try

i dont think the model cares much about the missing high frequencies

[fossil sage]

alright i need to read the documentation

[analog obsidian]

rvc never got an official doc lols... o wait it has one, but in chinese

[fossil sage]

im reading the ai hub one

[analog obsidian]

oh ok

[fossil sage]

ye best to read it first before i ask for help

[spare tree]

my rvc is bugging

it wont repeat everything I say and it has a 10 second delay

[knotty moth]

time to upgrade your pc with a decent gpu like RTX 3060 or depending on your budget

[knotty moth]

not with dithering

[viral mason]

what rvc

voice changer? training?

[stiff idol]

Yesterday I tried using RVC's basic MDX models, tey were quite good, they just didn't catch a Chinese instrument. I have a problem with removing it. Noise reduction kinda doesn't work on that.

[spare tree]

realtime voice changer

[viral mason]

whatcha using

can u send a ss

[spare tree]

@viral mason

or do u want me to send the application

[viral mason]

What one is this, and what gpu do u have

Is it deiteris, tg fork?

Did u get it from a yt link?

[spare tree]

yeah

https://github.com/w-okada/voice-changer/blob/v.2/docs_i18n/README_en.md

i got this one

[viral mason]

Definitely outdated

What gpu do ya have

Nvidia, AMD, intel?

[spare tree]

AMD Radeon TDM Graphics i think

AMD

[viral mason]

AMD, cool

U can use either wokada deiteris which is what I use or wokada tg fork which is the same thing but slightly different but I never used it so I recommend deiteris

-rt

[patent trellis]

💻 Local Realtime RVC

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

• Vonovox

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

• Wokada Tg-Develop Fork

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

• Wokada Deiteris Fork

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

⚔️ Wokada Deiteris Fork vs Vonovox

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

⛔ Outdated/Discouraged

These options are not recommended for use.

• Original Wokada

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

• RVC GUI Mainline Realtime

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

[viral mason]

Third guide is deiteris fork

[spare tree]

are those voice changers

where can i get deiteris

[knotty moth]

many ppl following such youtube tuts are falling to this trap, and the worse thing you havent checked the system requirements that igpus are not capable at all

[viral mason]

Yes they are, the download to deiteris is the guide that is under where it says wokada deiteris

[spare tree]

alr do i uninstall the old one

[knotty moth]

the problem is he has only the radeon laptop igpu

[viral mason]

Yea

[viral mason]

Wdym

[knotty moth]

not really effective so it could only run in cpu mode

[viral mason]

Ah

Yea he might be kinda screwed then

[spare tree]

oh

so i get a new gpu

[fringe snow]

Does anyone know why Kaggle's Applio URL is just taking to the files instead of applio?

nvm now it wants to work as soon as i hit send lol

[viral mason]

If u can afford a new GPU I'd get something from Nvidia

AMD isn't the best for ai realtime

Intel basically doesn't work at all

[carmine siren]

appoilio not working , I am encountering this issue when trying to execute third cell of the notebook in kaggle

[knotty moth]

bare minimum: RTX 2060 or radeon RX 6600
recommended: RTX 3060 12 gb or any faster 8+ gb vram

[low shard]

yeah that’s what I meant, adding a Gradio tunnel and use localtunnel for the tensorboard and filebrowser, will commit it later today

[low shard]

wouldn’t that cause issues for users?

[simple ore]

what does it say?

[humble cobalt]

is there any good alternative ai voice changer thats like w-okada?

[meager lichen]

is rx 6600 enough for ai voice changer

[plush reef]

If you only want to download ready made videos that have been generated by AI, that can be different . most tools are now for creating the videos for you, rather than providing a video library where you can download for free. Some sites, like Pexels, Pixabay, have free stock style videos that look AI-generated, but they are not technically AI-generated on demand. if you want a realistic high quality videos as per you script including enhanced features like AI Natural voices and customs AI avatars , you need to go for the paid ones. There are some tools offers budget friendly plans.

[meager lichen]

yo pls

[carmine siren]

-kagle

--kaggle

--kagle

[hallow thistle]

-kaggle

[patent trellis]

📘 Kaggle Notebooks

Kaggle is a Cloud (Remote Good PC) Service that offers 30 hours of GPU weekly, but needs a phone number verification

• **Applio Notebook**

by IAHispano
Kaggle

• **Hina Mod Original Wokada**

by Hina
Kaggle

• **Wokada Deiteris Fork**

by Hina & Deiteris
Kaggle

• **UVR5 UI**

by Eddy, ArisDev & Nick088
Kaggle

• **UVR5 NO UI**

by Eddy
Kaggle

• **RVC AI Cover Maker UI**

by Shirou & ArisDev
Kaggle

• **Music Source Separation**

by Shirou
Kaggle

[hallow thistle]

You mean this command?

[meager lichen]

is 8 vram enough

guys pls

[hallow thistle]

!howtoask

[patent trellis]

❓ How to Ask for Help

✅ Before You Ask!

1. Check Docs & Guides: Your answer may already be in the AI Hub Docs or the https://discord.com/channels/1159260121998827560/1159513888199540817 channel.
2. Search the https://discord.com/channels/1159260121998827560/1192011222023950368 : Look for existing posts that solve your issue. Do not invade someone else's post.

📝 How to ask?

Tell your:

• Full GPU Name: (e.g., NVIDIA RTX 3060)
• Operating System: (e.g., Windows 11)
• Detailed Description: What were you trying to do and what went wrong?
• Tutorial Used: Link to the guide you were following.
• Screenshot: A picture of the full error message is very helpful.

🚫 Prohibited Topics (We Will NOT Help With These)

To maintain a lega, safe & ethical community, we will NOT provide help for:

• (E girl, as an example) catfishing/trolling, scamming, impersonation.
• NSFW/Porn.
• Any illegal activities.
  Requests for these topics will be ignored and may result in moderation action.

<:matsuripray:1159685390156967936> Community Expectations

• Be Polite & Patient: Our helpers are volunteers. You may ping the Helpers role once.
• English Only: Please keep all conversations in English.

[carmine siren]

i am getting ngrok issues with the applio like its is asking verify with credit card credentials to use it further

[meager lichen]

@hallow thistle uh so same question, rx 6600 8000mb vram is ok right?

im ok to download the voice changer

[deft condor]

guys i downloaded okada and opened the file up, downloaded it. what should i open up afterward?

[hallow thistle]

I'm more familar of GB in RAM/VRAM, so yes, 8GB of VRAM is enough. Deiteris' fork W-Okada "DirectML" will work on AMD Radeon RX 6600.

[hallow thistle]

!howtoask

[patent trellis]

❓ How to Ask for Help

✅ Before You Ask!

1. Check Docs & Guides: Your answer may already be in the AI Hub Docs or the https://discord.com/channels/1159260121998827560/1159513888199540817 channel.
2. Search the https://discord.com/channels/1159260121998827560/1192011222023950368 : Look for existing posts that solve your issue. Do not invade someone else's post.

📝 How to ask?

Tell your:

• Full GPU Name: (e.g., NVIDIA RTX 3060)
• Operating System: (e.g., Windows 11)
• Detailed Description: What were you trying to do and what went wrong?
• Tutorial Used: Link to the guide you were following.
• Screenshot: A picture of the full error message is very helpful.

🚫 Prohibited Topics (We Will NOT Help With These)

To maintain a lega, safe & ethical community, we will NOT provide help for:

• (E girl, as an example) catfishing/trolling, scamming, impersonation.
• NSFW/Porn.
• Any illegal activities.
  Requests for these topics will be ignored and may result in moderation action.

<:matsuripray:1159685390156967936> Community Expectations

• Be Polite & Patient: Our helpers are volunteers. You may ping the Helpers role once.
• English Only: Please keep all conversations in English.

[meager lichen]

ty @hallow thistle

so last question, the deitaris version is better than the original w-okada version right

[deft condor]

im using rtx2050, win11, i downloaded okada and opened MMVCServerSIO, downloaded it. what should i open up afterward? nothing pop up for me. im not following any tutorial

[hallow thistle]

This looks like you have reached the usage limits of your free tier, so ngrok website now asking you to "pay" for it with your credit/debit card to have more usage time.

[hallow thistle]

Which W-Okada? Deiteris or v.1.5.x.x?

[hallow thistle]

Deiteris fork W-Okada, including its DirectML variant, is better than original W-Okada versions.

[deft condor]

deiteris

[cyan pelican]

guys can i use a v2 rvc model on the normal rvc app or will it sound bad

[hallow thistle]

RVC v2 voice models always preferred to use in any RVC program. The original RVC v1 models could work, but their quality won't be superior than RVC v2.

[hallow thistle]

Is the MMVCServerSIO still in your folder? If the program gone after you run it, an antivirus might have interfered it.

[deft condor]

i dont have my firewall nor my antivirus on. if it helps i did download an wokada voice changer before too

[knotty moth]

please explain, did you actually get rate-limited when opening Applio UI, which is a known issue currently?

[nocturne mural]

for now we’ll only use zrok and localtunnel, I’ll check pinggy later.

[viral mason]

Vonovox

[deft condor]

any solution to this guys?

[gleaming rapids]

hey sorry to disturb, what epoch meaning ? and where can i set per exemple 300 epochs

[low shard]

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

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

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

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

[gleaming rapids]

thx

[night dome]

Heyy I have a problem with the AI Voice Changer/or the Virtual Cable, whenever I try to use or test it on discord Its not working. When mic testing on discord it says that "Discord is not detecting any input from your mic" or something like that.

[hallow thistle]

!howtoask

[patent trellis]

❓ How to Ask for Help

✅ Before You Ask!

1. Check Docs & Guides: Your answer may already be in the AI Hub Docs or the https://discord.com/channels/1159260121998827560/1159513888199540817 channel.
2. Search the https://discord.com/channels/1159260121998827560/1192011222023950368 : Look for existing posts that solve your issue. Do not invade someone else's post.

📝 How to ask?

Tell your:

• Full GPU Name: (e.g., NVIDIA RTX 3060)
• Operating System: (e.g., Windows 11)
• Detailed Description: What were you trying to do and what went wrong?
• Tutorial Used: Link to the guide you were following.
• Screenshot: A picture of the full error message is very helpful.

🚫 Prohibited Topics (We Will NOT Help With These)

To maintain a lega, safe & ethical community, we will NOT provide help for:

• (E girl, as an example) catfishing/trolling, scamming, impersonation.
• NSFW/Porn.
• Any illegal activities.
  Requests for these topics will be ignored and may result in moderation action.

<:matsuripray:1159685390156967936> Community Expectations

• Be Polite & Patient: Our helpers are volunteers. You may ping the Helpers role once.
• English Only: Please keep all conversations in English.

[hallow thistle]

Make sure to read help guidelines before start asking.

[night dome]

Nvidia RTX 3050, Win 11. I was trying to use and test my mic on discord but when I tried to test it it said that "Discord is not detecting any input from your mic"

tutorial used: https://www.youtube.com/watch?v=SxdnGxicJOg&t=417s

I cant send ss

[viral mason]

[glass meteor]

When I say something it lagged in the middle of the sentences, can someone help me? i have RX6800

in the app its laggy but idk if in game it is

[hallow thistle]

Which W-Okada version are you using? And did you follow any tutorial video from YouTube before?

[glass meteor]

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

i follow this

_v.1.5.3.18

https://www.youtube.com/watch?v=Rj8Xbuce1uw i follow this to see if it fixes

[hallow thistle]

The W-Okada version you use is outdated. Try this better W-Okada from this guide link. https://docs.aihub.gg/realtime-voice-changer/local/deiteris-w-okada-fork/#download-amd-intel-and-cpu-on-windows

No, no, better not follow any tutorial video about W-Okada from YouTube for this time.

[viral mason]

is rx amd or nvidia

[fossil sage]

hows te graph

im training 10,000 epoch

when do i stop

[viral mason]

hell no, anything over 600 is overtrained

[fossil sage]

thats what the guide told me to do

it says to stop when there is no longer any improvement

and when it stays flat and makes no progress stop

[analog obsidian]

is that from the ai hub docs?

[fossil sage]

yes

[analog obsidian]

i have told nick to rework the training section because it's so long and wrong

lol

tensorboard doesn't show you when the model is overtrained

[fossil sage]

WHAT

[analog obsidian]

600e is also crazy

i'd recommend maximum 100
but start hearing the epochs from 40e

[fossil sage]

what about the overtraing threshold

i have it enabled

[analog obsidian]

so train 100e, hear the model at 40e

if its sounds good then its done

if not hear 50e, etc etc

[fossil sage]

elborate

[analog obsidian]

rvc finetuning overtrains fast

thats just how the arch works

[analog obsidian]

try listening to 45e, and 60e

[fossil sage]

does it look like its being over trained right now

[analog obsidian]

anything past 105 is most likely overtrained

[fossil sage]

so i should stop the training correct ?

[analog obsidian]

yeah

[fossil sage]

ok

[analog obsidian]

and in future trainings just ignore the tensorboard lmao, just train 100e max
99% of the time the 'good' epoch will be 40e, but sometimes can be 50e, 60e, etc

the tensorboard becomes useful when you train without a pretrain, while using a dataset of 55 hours

[fossil sage]

oh ok

when the guide going to be updated

[analog obsidian]

i forgot to say also, don't use a batch size below 8, anything below that makes the model learn painfully slow

[analog obsidian]

idk

increase chunk size and leave extra at 2.7

decrease game graphics, limit fps to 60

1080p

[fossil sage]

i put it at 8 exactly

[analog obsidian]

stop the conversion

[analog obsidian]

oh, is your index value at 0?

it's not 100% possible to erase a model's native accent tho, they will always have some leftover from the dataset

especially if the model is overtrained

sorry english is not my first language, but you want the model to use it's native accent or yours?

increasing the index blends more of the model's accent into the result

[viral mason]

drink water

[analog obsidian]

if your goal is to make the model sound more accurate to the dataset, index can help

index 0 the model is going to use your accent instead

but it's not 100% perfect

yup! actually thats very good for the model

it'll have an easier job while doing the conversion

[fossil sage]

the model sounds decent

the probelm is that im not able to mimic his accent

[analog obsidian]

try setting index to 0.7

[stiff idol]

Guys, does anyone have a normal guide on removing echo and re-verb? And I can tell you I tried the denoise, de-echo and other such functions form various plugins, but nothing as I thought it'd. So I'm probablyj ust bad at it and want to get better. I extracted vocals through a model. (mel and MDX ...VF actually gets me the same result). So the only thing I want is to remove an instrument the model doesn't know at all (some kind of Chinese instrument) + echo and reverb that I hear with in the voice.

and tutorials are more about "how to lower the audio" instead of "removing the echo" (I also tried the Audacity tools denoise, noise gate and the other typical ones)

[fossil sage]

better but its ok

[analog obsidian]

use uvr de echo model

[fossil sage]

why does it have some artifacts doh

[analog obsidian]

decrease index to 0.5

[stiff idol]

Which one do you recommend? I really tried playing with the audacity and other plugins, but nothing really worked.

[fossil sage]

how do i do that

[analog obsidian]

uvr de echo is the best imo

[analog obsidian]

go to the inference tab, click advanced settings
set feature index ratio to 0.5 or 0 to fully disable it

[stiff idol]

UVR denoise light?

sorry

read wrong xD

[analog obsidian]

uvr de echo

[stiff idol]

there's lots of models in VR architecture

[fossil sage]

alr thanks

[stiff idol]

there's many de-echos, could you be more specific about the model?

[analog obsidian]

show ss

screenshot

[analog obsidian]

try de echo aggressive

[stiff idol]

ok 👍

[fossil sage]

@teal ferry SOUNDS PRETTY GOOD

this before

[analog obsidian]

in the same advanced settings also enable split audio

[fossil sage]

what does it do

[analog obsidian]

instead of converting the whole reference audio at once it will split it into small segments, decreasing vram usage

after it converts the chunks, it will merge them into one audio file

[fossil sage]

oh ok

now i need to read the appilo documentation

[analog obsidian]

and i personally noticed my results sound better with split audio on tbh

[fossil sage]

do you use tts or no

[analog obsidian]

nop

[fossil sage]

this is why i use tts im not able to mimic his accent and the way he talks

[teal ferry]

does it

what model is it

[fossil sage]

i just finished making it like 10 minutes ago

[analog obsidian]

yea it cant mimic the accent if the inference audio doesn't sound like him

[fossil sage]

exactly

which is why you need something like xtts

[fossil sage]

yonshuk dexter morgan model_90e_9900s anything above 90

[analog obsidian]

yeah rvc doesnt learn emotions or accent since its learning spectograms/mel

[teal ferry]

rvc?

[stiff idol]

man, thank you 🙏 it worked, is there any list of a good models and the ones that can get rid of Chinese instruments and such?

[teal ferry]

if its just copying what the reference audio sounds like then yeah it will sound fine. theres no real use case for that though beyond real time live inference. or if youre going to want it to convert your own speech. or using it for like a secondary filter

for what you want to do you want to use it with tts

[analog obsidian]

hmmm have you tried gabox fv4?

but if your vocals have harmonies you have to use becruily's mel karaoke

for me they both work fine

[stiff idol]

Nope, I haven't tried those 2. Could you provide the links?