#✨│ai-help

1 messages · Page 243 of 1

simple ore
#

god.. who's using teamspeak in 2025

young dirge
copper phoenix
#

Hello, can anyone tell me which Colab is currently used to make AI Covers and Models?

odd isle
#

is it normal for the output headphones to sound a little better than cable output?

real depot
#

Hello everyone!
I’m looking for help to create a custom Arabic RVC voice model for a gift.
I want to make a voice similar to Gulf Arabic artists like Ouzii / Luigii style, for a personal use (not commercial).

If anyone is available to help me build the model, I would really appreciate it 🩷
I’m ready to pay for the service if needed.
Thank you so much 🙏🏼

charred marten
#

Which voice changers are free?

viral mason
#

is Mel-Roformer-Denoise-Aufr33 better or Mel-Roformer-Denoise-Aufr33-Aggr

viral mason
charred marten
acoustic coral
#

bro i can not find the rvc download anywhere

viral mason
acoustic coral
#

i used to have it it was just called rvc and i rember getting it off git hub, you would import the zip and put in a pre recorded audio and then it would be ai voice so

viral mason
#

are you trying to train a model?

jaunty bloom
#

I wanted to train a voice model, but the link i usually use for the rvc fork is down. Does anyone have another link?

delicate whale
#

Is their a cracked version of Voice Mod?

charred marten
#

Is there a way to fix static or is it just a mic issue?

mighty sinew
#

using applio and recieving this error An error occurred during audio conversion: index -1 is out of bounds for axis 0 with size 0 how to fix?

shell gulch
#

i forgot what is the best ai cover maker rn?

prisma kettle
#

What happened to the separate channels for the RVC help

prisma kettle
prisma kettle
#

I'm not getting any output through the virtual cable doing realtime, anyone know why?

#

I can hear it when I set my monitor to headphones but nothing comes through on discord or in sound control panel

brittle wing
#

hi, does anyone know, how to check my model in training process (weights.com)? because it says that it already created, but i cant use it and even see in the list. I just wanna know what time left

charred marten
#

Is there a way to get rid of like the weird voice cracks?

analog obsidian
#

well different ways actually

#

one is training a better model

#

then the other way is enabling fp32 mode and pray if your current model is going to get any better (if the model was trained in fp16 i don't think enabling fp32 will help that much since the model is already fried inside)

charred marten
analog obsidian
#

i noticed voice cracks happen because the f0 estimator being mid (they're not that great sadly) but also because the model is trying to do a sound it doesn't know how to reproduce

charred marten
analog obsidian
fleet epoch
#

i am using rvc ai cover maker and when im trying to make a cover it at export audio and output information it says error

acoustic coral
viral mason
acoustic coral
#

okay so it was an app. called RVC. i had it on my old computer. And you couldint train voices but you would import zips upload an mp3 click how many. "ceces" i thinnk it was called then it would export it in the voice

viral mason
acoustic coral
#

omfg

#

dude

#

let me do smth rq

#

🥀

#

im gonna make a drawing of what i remeber the app looking like

viscid moss
junior gull
#

Heya, just wondering how do we get ranked up so we can share our RVC and TTS models? Also been doing ton of learning, testing, and ai training and would love to soon share some of the work I have. Making a reliable compact modular system that can spontaneously regenerate your custom AI on almost any platform with training/memory enabled, very lightweight and 100% portable.
Having a blast learning about all this and making stuff and would love the chance to share too!

wicked crow
potent bone
viral ruin
#

getting this on kaggle when starting applio:

knotty moth
fallen crater
#

where can i download onnx models most of them are pth models there

junior gull
digital bear
#

hey i'm currently using the w okada fork but i can't seem to find a way to prevent the voice changer from picking up my laptop's fans and making unwanted noises

simple ore
hard parrot
#

why rvc gui not opening
is it a error?

Running with the system Python.
Nie moPress any key to continue . . .

maiden latch
#

where is a more up-to-date tutorial on how to use colab for w-okoda's realtime voice changer thing?
i dont understand anything and the tutorial im following isint working

digital bear
#

laptop

pastel oak
#

But whats your gpu first of all

maiden latch
#

it's like an amd 580, it's really not the greatest so thats why i wanted to do it online since the app doesnt work too great on my pc

pastel oak
maiden latch
#

was about to ask where to find that thank you

pastel oak
maiden latch
#

tysm

hard parrot
#

why rvc gui not opening
is it a error?

Running with the system Python.
Nie moPress any key to continue . . .

maiden latch
simple ore
maiden latch
#

k

simple ore
#

they may have shifted some options recently

round bear
#

is "Realtime Voice Changer Client" still good?

hard parrot
#

i just installed applio rvc and this shows up
Please run 'run-install.bat' first to set up the environment.
Press any key to continue . . .

#

what i need to do

crude flame
simple ore
#

No, it means they ran run-applio.bat it as admin

#

@hard parrot

digital bear
#

by the way is there a way to make it sound better my models sound nowhere near as good as the samples

maiden latch
analog obsidian
mental pine
#

my twiter is blocked how do i login

#

to my ai hub

#

anyone help me

fleet cedar
#

can anyone help me why is this lagging

image.png
pastel oak
#

Chunk (the 512.0 ms) has to be higher than perf

#

Whats your gpu

fleet cedar
fleet cedar
pastel oak
#

do chunk 200 and extra 2.7

fleet cedar
#

idk bro

pastel oak
#

f0 det rmvpe

fleet cedar
fleet cedar
#

goated

#

it works

round orbit
#

can anybody help me with the error of vol0000?

#

line 1 is working, but sound is not being played. I reinstalled the program and components ~5 times, but no changes.

jaunty anvil
#

so for videos like the president stuff, is elevenlabs the best to do that stuff?

fleet cedar
#

guys when i speak how do i make it loud for others

#

monitor thing doesnt work

languid cliff
jaunty anvil
#

is elevenlabs the best thing for the tts videos or is there a better alternative?

simple ore
jaunty anvil
#

...and i'm not a coder, shit.

simple ore
#

it installs with one command line

jaunty anvil
#

where do i put that command?

simple ore
#

as long as you have python installed, ideally 3.10 or 3.11

jaunty anvil
#

i uh don't have python at all lmao

#

i'll get on that

simple ore
#

dont use 3.12 or 3.13

jaunty anvil
#

got this

image.png
simple ore
#

from command line

#

not from python

jaunty anvil
#

oh

simple ore
#

somewhat more proper way of doing it without using global repository

jaunty anvil
#

this correct?

image.png
simple ore
#

it probably installs cpu version of the torch

#

if you run nvidia gpu you may need to change it to cuda version

jaunty anvil
#

i'm a bit new to all this stuff

simple ore
#

anyway, it can run on CPU as well, that's fine to just test it

jaunty anvil
#

but i am doing the correct command in the right place now

simple ore
#

you are installing it in to the global repository

#

usually not a good idea to do that with multiple project as there are often conflicting versions of the libraries

jaunty anvil
#

any idea how to uninstall it?

simple ore
#

dont worry too much, you can install it properly later

jaunty anvil
#

alright

#

i mean i just restarted the installation like three times already lmao

#

alright it was giving me a lot of errors

#

idk what to do

simple ore
#

worked fine for me

#

I bet that's that wheel creation thing

jaunty anvil
#

wheel creation?

simple ore
jaunty anvil
#

i just don't want the files cramped in my c drive

#

i already installed and cancelled it like 3 times to troubleshoot

simple ore
#

well, by default the global repository is on C

jaunty anvil
#

yeah but i couldn't find it

#

no folder saying global repository

simple ore
#

under c:\users\user\appdata\local\programs\python\python310\libs\site-packages

#

that's the global repository

jaunty anvil
#

python isn't in programs

simple ore
jaunty anvil
simple ore
#

unless you've installed python to somewhere else

jaunty anvil
#

OHH i got it from the microsoft store

simple ore
#

make sure you check [x] add python to path

jaunty anvil
#

i'm just not cut out for this shit 😭

simple ore
#

most of AI project is not for beginners

jaunty anvil
#

so uh

#

i guess i should just do elevenlabs?

simple ore
#

but you can ask chatgpt to explain things

jaunty anvil
#

is that the best for beginners that want to export models here and make tts videos and shit?

simple ore
#

if you install chatterbox is it really simple to use

jaunty anvil
#

alright i'll keep working

coral haven
#

my vcc thing isnt working. help

fleet cedar
#

how do i fix this being loud at the start when i speak then it gets normal

desert linden
#

how do i fix this error:

#

The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
from distutils.util import strtobool

simple ore
#

not an error

tender sand
#

my rvc wont work, my friend is just saying its playing a hello every few seconds and my audio wont go through at all

#

nvm got it to work

desert linden
# simple ore not an error

Timer: 00:00:48/content/voice-changer/server/HVoice.py:3: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
from distutils.util import strtobool
Traceback (most recent call last):
File "/content/voice-changer/server/HVoice.py", line 10, in <module>
from downloader.SampleDownloader import downloadInitialSamples
File "/content/voice-changer/server/downloader/SampleDownloader.py", line 12, in <module>
from voice_changer.RVC.RVCModelSlotGenerator import RVCModelSlotGenerator
File "/content/voice-changer/server/voice_changer/RVC/RVCModelSlotGenerator.py", line 4, in <module>
import torch
ModuleNotFoundError: No module named 'torch'
WARNING:pyngrok.process.ngrok:t=2025-06-02T23:20:37+0000 lvl=warn msg="Stopping forwarder" name=http-41369-e68eec6b-0315-4a59-870c-6d6c66810395 acceptErr="failed to accept connection: Listener closed"
--------- SERVER STOPPED! ---------

#

this is the full error

simple ore
#

seems like you tried to install requirements and it failed. You may want to grab a prebuilt package.

#

but again this looks like colab?

desert linden
#

yeah im using colab

simple ore
#

with realtime rvc?

#

it is dead

desert linden
#

whattttt

simple ore
#

unless someone fixes the install part

desert linden
#

omg ur kidding

#

when did it stop working

#

i used it litch months ago

simple ore
#

show me the url

desert linden
simple ore
#

yeah, outdated af, i aint gonna touch it

desert linden
#

fml

#

i never had issues w it before

#

do u happen to know an updated version?

#

that supports mac?

simple ore
#

well, maybe not oudates, but something is wrong with requrements install perhaps

desert linden
#

how do i fix it

simple ore
#

yeah, f that

#

colab is using python 3.11, this package has only 3.10 wheel

desert linden
#

oh

#

so no way to fix it?

simple ore
#

The last fix for this colab was done 5 month ago

desert linden
#

rip lol

simple ore
#

in that time I had to fix applio colab like 5 times already

desert linden
#

ugh

violet rose
#

what is the best plz ?

fleet cedar
#

whats wrong with this rvc not sending audio

#

to the vc

safe echo
#

hi does anyone know how to stop this happening? i cant delete slots with the edit button as all my slots are coming up as "blank" from previous voices

image.png
#

is there another version? xd

#

Is there a way to use without internet connection/chrome tab?

#

like a built in program?

analog obsidian
safe echo
#

oh, do you have a link/guide for that please?

analog obsidian
#

tutorial is there, just scroll down until you find it

safe echo
#

thank you very much

analog obsidian
#

nvidia only

safe echo
#

gotcha, im on RTX 3070 TI.

analog obsidian
safe echo
#

oh, do i have to compile it myself?

crude flame
#

just download it and run setup.bat

analog obsidian
crude flame
#

then once thats done run start.bat

safe echo
#

im only finding the source code downloads on the "downloads" tab

image.png
analog obsidian
#

download the repo as zip

#

click code > download zip

crude flame
safe echo
#

excited to play around with it 👍

analog obsidian
#

good luck! vonovox is great

#

dev is active in the rvc community

crude flame
#

things change quick

analog obsidian
#

yeah lol

safe echo
#

would you say the performance of vono is better than rvc?

crude flame
#

rvc does not mean realtime voice changer

#

vonovox is faster than w-okada

safe echo
#

oh, what should i use to refer to the web based version?

crude flame
#

w-okada

safe echo
#

gotcha, that would make sense.

crude flame
#

if you talking about the realtime version

safe echo
#

i am yes.

crude flame
#

then yea, w-okada

safe echo
#

Thank you, been using okada for a while, but i feel like my card can be utilised better.

#

unsure how to word that but yeah

analog obsidian
#

wokada is poorly optimized yup, the fork dev tried his best to improve it

#

vonovox is a completely new software, so things are better

safe echo
#

awesome Cool_Doge

analog obsidian
#

in terms of perfomance

safe echo
#

long install time via the bat xD

#

getting there

#

do i use launcher once that says complete?

#

or start

simple ore
#

I think there's now some way to conver the whole pytorch model thing to just a cuda kernel

#

for super fast performance

safe echo
#

oh, vonovox shows my folders empty where my pth files are, thats odd. (oh, vono uses pth not safetensor)

analog obsidian
safe echo
#

where can i stay updated with that please? as all my main voices are on safetensor

analog obsidian
safe echo
#

you've been so helpful, thank you lyery!

safe echo
#

is there any good places to find pth files?

#

usually use weights, but im unsure if its just safetensor there

simple ore
crude flame
safe echo
#

thank you

knotty moth
valid vine
#

I'm about to stab the hugging face website 23 times on the aides of march

#

I want to download deepseek prover V2 671B but there's 163 files and nothing I do works

#

I've tried git, aria, Jdownloader, and the python CLI

simple ore
#

git lfs?

valid vine
#

idk if it doesn't exist or what

image.png
#

I'm on windows btw

#

oh wait I found it

#

how do I download it using that?

simple ore
#

the usual git clone of the repo?

valid vine
#

where does one find that? it's not the link at the top of the website or the one the files are coming from apparently

simple ore
#

wihout /resolve/main ?

valid vine
#

my god finally

image.png
#

tysm

#

I've spent like an hour + on this

spice nacelle
#

does anyone know of any text to speech models that run locally or on web and are not as expensive as eleven labs since I create audio's of over 1hr every other day

red kayak
#

@candid basin

#

Is very likely that it's your datasets issue

#

Care to show me a short sample of the audio u are using

candid basin
candid basin
red kayak
#

Well training is fine tuning so yes you can definitely fine tune

red kayak
candid basin
red kayak
#

Yeah that's a zero shot tts

candid basin
red kayak
candid basin
red kayak
viral ruin
#

anyone has a RVC colab link that works??

#

for training

jaunty shale
#

I tried it and it's insane how you can make the vocals sound so much better

#

thank you!

hot violet
junior gull
#

Is this W okadas? It looks a little different.

broken urchin
junior gull
broken urchin
junior gull
broken urchin
frigid arrow
#

who can make me a index and pth

graceful jettyBOT
#

🎉 | Jawh leveled up!
ℹ | Level up messages can be disabled for the guild with owo level disabletext

levelup.png
frigid arrow
#

i try everything . shi isnt working

knotty moth
golden walrus
#

Guys. Can i ask if there is any voice changer that can use spin embedder?

crude flame
graceful jettyBOT
#

🎉 | Razer leveled up!
blank | Extra rewards were added for missing levels

levelup.png
crude flame
analog obsidian
golden walrus
# crude flame vonovox

cat_blush btw, do you know if it's okay to pair spin with KLM 4? Or the experimental one that SSS made in pretrain lah.

crude flame
golden walrus
#

Thank you so much

cosmic epoch
#

can someone give me a link to a colab where i can train models (one that isn't applio)?

simple ore
#

good luck with that

median monolith
#

What limitations (if any) does the Weights voice model training feature have regarding audio quality compared to a local or cloud training like "Mainline Collab" or Applio?
Like, whats the max ammount of khs that the Weights training (USING A PREMIUM TRAINING) uses/supports of an audio for example.
Or if it adds any sort of compression to the dataset/final model audio no matter what format and properties it has.
Im only certain of the fact that you cannot upload very heavy audios for the training, wich means you will mostly not be able to use a max quality wav dataset for example.

viral mason
arctic trail
#

how do i even download is there a tutorial

viral mason
arctic trail
viral mason
#

bet I can help u, dm me ^^

viral mason
#

idk any of the complicated stuff I can only provide this kinda info

#

I'm no nerd..

median monolith
median monolith
median monolith
viral mason
storm merlin
#

how can i turn someones normal voice into like an ai singing voice

viral mason
storm merlin
viral mason
daring heath
#

whats the target lufs for rvc dataset

#

-18?

valid vine
#

omfg it doesn't even work???

image.png
simple ore
#

(689 * 1024) / 82.. 3+ hours

#

tf do you expect lol

valid vine
#

yeah no when I say "like this" I mean "100% (163/163)"

#

this has been running for abt 4 hours

#

that 689 isn't the total going to be downloaded, that's the amount already downloaded

#

and it's been going up every once in a while those 4 hours

#

and it's been stuck at 689 for like 30 minutes

#

well more like 40/45 now

valid vine
#

which apparently learning how bad huggingface is was too much to ask

simple ore
#

lfs is how the models are stored there, there's no other way of downloading them, other than clicking off each file manually

valid vine
#

so the way they're stored just doesn't work then

distant turtle
#

-colab

patent trellisBOT
# distant turtle -colab
📒 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

simple ore
#

be grateful it does not give you 5MB/s download speed

#

ohh.. 700GB download is taking more than 4 hours... the horror

valid vine
#

dude

#

I'm not "downloading 700 gb for free" if over half of the 4.3gb dls are 1kb

#

I'm not complaining it's taking long

#

I'm complaining it doesn't work

simple ore
#

chill the f down and wait until it is done

valid vine
#

oh my bad I though percent was out of a hundred

#

so could you enlighten me on what percent means then?

simple ore
#

open one of those 1kb files in notepad

#

then do git lfs pull

valid vine
#

I'm doing stuff with my pool one minute

simple ore
#

lfs downloads the resource pointers 1st, those are 1kb files

#

then it actually pulls the content

sand quartz
#

Does the server have a channel for Lora's for using stable diffusion

simple ore
valid vine
#

so the 1kb files have this in it

image.png
#

but none of them have changed

#

and it's been sitting at 100% for over an hour now

simple ore
#

ctrl-c and git lfs pull again

#

that's the resource pointer

valid vine
#

it's blank

#

not sure how long it's supposed to take before I notice smth

simple ore
#

i mean the text in that 1kb file, it is supposed to be replaced by the actual 4GB content

valid vine
#

yeah I think that's only happened with 4 so far

#

like the one numbered 004 not 4 different ones

simple ore
#

it should be downloading them after the pull

valid vine
#

well they should all be downloaded

#

there's a 641gb "objects" folder in .git

#

I assume it's pulling smth from there

simple ore
#

you can check the status with git lfs ls-files I think

valid vine
#

I assume * is done and - is in progress

image.png
valid vine
#

it was all a fucking waste anyway

#

you need an Nvidia GPU and I have an AMD

#

it doesn't tell you ANYWHERE

#

this one section is the only way it'd be possible to find out

image.png
simple ore
#

Attempts to run deepseek, finds out

valid vine
#

now I'm trying to run the 7B parameter version of math (not prover) and using the exact thing it tells me to in the way it tells me to and it's giving an error based on their code

# 
Traceback (most recent call last):
  File "D:\AIs\Deepseek-Prover\runDeepseek.py", line 6, in <module>
    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, device_map="auto")
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "d:\AIs\Deepseek-Prover\venv\Lib\site-packages\transformers\models\auto\auto_factory.py", line 571, in from_pretrained
    return model_class.from_pretrained(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "d:\AIs\Deepseek-Prover\venv\Lib\site-packages\transformers\modeling_utils.py", line 309, in _wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "d:\AIs\Deepseek-Prover\venv\Lib\site-packages\transformers\modeling_utils.py", line 4508, in from_pretrained
    model = cls(config, *model_args, **model_kwargs)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "d:\AIs\Deepseek-Prover\venv\Lib\site-packages\transformers\models\llama\modeling_llama.py", line 618, in __init__    
    self.model = LlamaModel(config)
                 ^^^^^^^^^^^^^^^^^^
  File "d:\AIs\Deepseek-Prover\venv\Lib\site-packages\transformers\models\llama\modeling_llama.py", line 379, in __init__
    self.post_init()
  File "d:\AIs\Deepseek-Prover\venv\Lib\site-packages\transformers\modeling_utils.py", line 1969, in post_init
    if v not in ALL_PARALLEL_STYLES:
       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: argument of type 'NoneType' is not iterable```
#

this is getting rediculus

simple ore
#

v has no value

valid vine
#

yeah I don't know why

simple ore
#

trace it back

valid vine
#

that's 5 files deep

#

and half the functions just tell me there's no definition in vscode

simple ore
#

usual thing when you IDE did not load definitions

#

if v not in ALL_PARALLEL_STYLES: this is a part of a pretrained model load

valid vine
#

okay so it's hardcoded to be none

simple ore
#

that goes thru attention implementation object

#

no it is not

valid vine
#

"if self._tp_plan is not None and is_torch_greater_or_equal("2.3"):
for _, v in self._tp_plan.items():
if v not in ALL_PARALLEL_STYLES:"

#

" _tp_plan = None"

#

and everything I've done is from the example given

simple ore
#

no

valid vine
#

my bad I thought "here give some examples" was an example

image.png
simple ore
#

config = self._autoset_attn_implementation(config, torch_dtype=dtype, check_device_map=False)

valid vine
#

and that's what I did

simple ore
#

self._tp_plan = self.config.base_model_tp_plan.copy() if self.config.base_model_tp_plan is not None else {}

#

if it is not empty, the loop happens

#

and it throws and exception if it is unsupported style

#

for you it throws one because the value is None

valid vine
#

so I have to go on my own and figure out what a tp plan is and set it for the model to work and they don't say that anywhere

simple ore
#

it should work without much fiddling with the code

#

what's your GPU?

valid vine
#

AMD rx 6750

simple ore
#

using zluda?

valid vine
#

if this was an issue about not having a GPU I'd understand bc it doesn't support any type of cuda or whatever the AMD equivilant was called

valid vine
simple ore
#

it is the magic that lets you run CUDA stuff on AMD GPUs

valid vine
#

does it still work if my GPU doesn't support ROCm?

simple ore
#

lil guide

#

this is for windows

valid vine
#

yeah my actually strong PC is on windows

simple ore
valid vine
#

wait but isn't this all for nothing if it's still giving an error I can't fix?

#

based on what the models are supposed to actually do this probably won't help with the thing I wanted anyway so this all is just kind of a waste of time

opal cobalt
#

@simple ore u seem knowledgeable mind if i shoot u a random question new to this discord but wanted ur opinion on something

hallow thistle
patent trellisBOT
# hallow thistle !howtoask

How To Troubleshoot AIHC_WaitWhat

__**GIVE CONTEXT.**__ 📝
  • Don't simply mention your issue, like "my rvc is not working".
  • Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
  • The more context, the better.
__**BE POLITE.**__ <:matsuripray:1159685390156967936>
  • Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
  • It's okay if you're frustrated, but don't take it into this server.
  • Don't DM without prior consent.
__**BE PRODUCTIVE.**__ 🤝
  • Don't ask for every little instruction. Put your own effort & test things by yourself.
  • Don't ask to ask.
  • Check if your answer is a Google search away/on our guides website.
If you've changed your mind about your message, feel free to delete/edit it.
inland hill
#

My friend cannot change their input settings. Every time they do they get this error message. They just got the software ( fresh install ), version 1.5.3.16a

image.png
hallow thistle
#

There's a better W-Okada than this version.

inland hill
inland hill
#

hoping to avoid compatibility issues whoopsies

hallow thistle
inland hill
#

should we just delete the entire folder the old one was in?

hallow thistle
#

Of course, sure. Delete the old one.

inland hill
#

i just want to make triple sure because i tend to mess up

#

we're downloading htis one right?

hallow thistle
inland hill
#

thank you!

inland hill
#

It's working wonderfully, thanks a lot!

viral ruin
#

anyone got a working colab link for RVC2 (no applio) ?

gleaming wasp
dry jewel
#

sorry to come back so late but i checked and the link doesn't work

#

error 404

dry jewel
#

i'm trying to do this with 'ngrok'

pastel oak
#

Chunk is mainly for latency but if its too low for your gpu to handle it will lose some quality too

pastel oak
dry jewel
#

this guide there doesn't even tell me what site to go to

pastel oak
#

There is a guide on the link nick sent right next to the link you opened

pastel oak
dry jewel
#

i have to use phone numbah?

pastel oak
dry jewel
#

bruh

knotty moth
#

RX 5000 series/newer are recommended

#

if it's only "AMD radeon graphics", more likely it's integrated gpu which is less capable

dry jewel
#

guys

#

what's this setting's purpose

odd shale
# dry jewel

F0 Det is the pitch algorithm you're already using.

dry jewel
odd shale
#

Don't use any of these crepe options.

dry jewel
#

that's the one used for these models

#

but i realized that "_onnx", which was default set, sounds more clear

odd shale
pastel oak
#

No

#

Higher chunk higher delay but also more time to compute the voice, but at some point increasing doesnt improve voice

#

Extra 2.7s , advanced settings: increasing crossfade length helps with clearer voice, turning on fp32 for nvidia gpus too

cosmic epoch
#

can someone give me a link to a colab where i can train models?

viral ruin
#

got this error on mac M1 when i wanna convert. Any ideas how to fix? Tried google already but didn't fix it: AttributeError: 'NoneType' object has no attribute 'tobytes'

#

the audio-path is definitely not wrong

simple ore
viral ruin
#

Traceback (most recent call last):
File "/Users/jlapping/.pyenv/versions/3.10.11/lib/python3.10/site-packages/gradio/routes.py", line 488, in run_predict
output = await app.get_blocks().process_api(
File "/Users/jlapping/.pyenv/versions/3.10.11/lib/python3.10/site-packages/gradio/blocks.py", line 1434, in process_api
data = self.postprocess_data(fn_index, result["prediction"], state)
File "/Users/jlapping/.pyenv/versions/3.10.11/lib/python3.10/site-packages/gradio/blocks.py", line 1335, in postprocess_data
prediction_value = block.postprocess(prediction_value)
File "/Users/jlapping/.pyenv/versions/3.10.11/lib/python3.10/site-packages/gradio/components/audio.py", line 349, in postprocess
file_path = self.audio_to_temp_file(
File "/Users/jlapping/.pyenv/versions/3.10.11/lib/python3.10/site-packages/gradio/components/base.py", line 325, in audio_to_temp_file
temp_dir = Path(self.DEFAULT_TEMP_DIR) / self.hash_bytes(data.tobytes())
AttributeError: 'NoneType' object has no attribute 'tobytes'
2025-06-04 15:10:38 | INFO | httpx | HTTP Request: POST http://localhost:7865/api/predict "HTTP/1.1 500 Internal Server Error"
2025-06-04 15:10:38 | INFO | httpx | HTTP Request: POST http://localhost:7865/reset "HTTP/1.1 200 OK"

simple ore
tough fiber
#

guys anyone can tell me which training and makings rvc models, last time i used RVC1006Nvidia

simple ore
tough fiber
#

thanks for info btw <3

simple ore
tough fiber
#

i wondering if we have good dataset for that voices is it possible to make good models?

simple ore
#

laugher generally fails, screaming requires a dataset with a large dynamic range, generally rvc inference is pretty flat

pastel oak
#

Optional

latent kettle
#

In w-okada ?

#

In and out sliders

low shard
covert portal
#

What do you need to download to train locally?

low shard
covert portal
#

nvidia

#

train rvc voice from dataset

low shard
#

nvidia made a lot of gpus

covert portal
#

2070 super

low shard
#

As you got a good PC, you can use RVC locally, you can choose between:

  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
#

I gave you an explaination to the differences and links to the docs

covert portal
#

ok thanks you!

low shard
cosmic epoch
#

can someone give me a link to a colab where i can train models (one that isn't applio)?

viral ruin
#

Let me know when you found something.. I need the same

crude flame
#

gl on finding a working on

warm swift
#

hi

#

someone can help me funding fuerza regida RVC?

viral mason
#

ew money

warm swift
#

?

viral mason
#

don't pay for ai

warm swift
#

i need it for a proyect

viral mason
simple ore
#

finding or funding, that's two very different requests

warm swift
#

o sorry

#

can someone help me?

simple ore
#

a or b?

warm swift
#

A

simple ore
warm swift
#

thank u

viral mason
#

spelling mistakes go hard

junior gull
#

live mods dont review the model submissions do they?
just wondering...i got a very silly and confusing reply for mine.
"sounds distored" "retrain and getr rid of distortion"
.... submited with the details and description and listed online with the same.... DISTORTION is signature to this char's voice pattern, since she always talks that way, and my goal was to faithfully capture that in great detail, and I did. I clearly communicated that on her hugginface listing too, and on my model submission details. Its not any different than the dozens of robotic voices listed here already. Same idea.
But then I got that wierd reply like a real person hadn't even bothered to read anything.

crude flame
#

thats why its a rule that you cant submit a robotic voice or a voice with effects

image.png
junior gull
# crude flame your model was rejected because the voice you submitted had effects and that mak...

but yet there are like over a dozen modles i can immediately see that have such effects and MUCH heavier too all in the model section?
So that doesnt' really make sense? And hers is WAY lighter than that....its a knowon characteristic of this character. It's not added. Its literally in EVERY refrence file to her voice becuase....its her voice. There is no version of her iwthout it....and I wouldn't want one if there was. Against thtat doesnt' make any sense???

crude flame
simple ore
#

Show your skills first with a normal model, then do whatever

crude flame
#

^

junior gull
#

why wasn't that said in the reply then? ~_~ instead of telling me to build my model withotu her main defining feature she is known for lol. could have saved ALOT of confusion.
Alirght then.

crude flame
#

it was

junior gull
#

maybe a languate barrier? that wasn't how it was phrased...it was still complaining about her voice. It never said I had to do a normal BEOFRE hers could be considered. That's a key differnce, and changes the meaning completely. But I understand now thanks to @simple ore clarification.

#

Thank you

crude flame
#

😐

junior gull
#

might be a bit though...the only other one i was already working on was another male voice with a similar effect....based on Zachary Quinto's Invincible character 🙃 Found these both really appealing as voices to use for AI related projects.

sullen lion
#

i'm trying to learn illustrious character lora training

#

the tut i used for pony claims that the settings should be fine for illustrious (and were in fact written with it in mind) but i find my resulting models are always less style influenced

#

any images i gen with them come out looking very booru

#

whats the best place for me to iterate on my settings to try and reenforce style? ive seen plenty of cartoon loras that actually maintain their source style on models like wai-nsfw and im trying to achieve that

#

also worth noting im p sure i trained at 512 nvm i checked i AM on 1024 res, will also say my datasets are unfortunately limited in size, usually around 15-25 total images

warm swift
#

mi spelling mistakes arre hard?

knotty moth
simple ore
#

great quality for regular artsy-fartsy and realistic pics

silent stratus
#

does this look bad to yall

nervyyyyyyyy.PNG
#

4 batch size with a 4 min dataset and that was at like 120 epochs

silent stratus
#

nvm i was just paranoid its fine

royal grove
#

Any good recent okada tutorials?

#

these good?

silent stratus
idle bramble
#

what is the diff between sio and rest protocols for the voice changer? i know what rest is, but does this provide any latency help over sio? if it does, then why isnt it the default? (asking bc some guides say to swap sio to rest)

royal grove
#

anyone know why my shit is just blank white

sand bison
#

Does anyone have the new Google Collab for creating AI models (the RVC v2 disconnected)?

knotty moth
royal grove
#

Any support on how to make them sound less ai>

deep tulip
#

Hey! Where can I find people to train a voice model? I have a dataset, I would be grateful for help

rancid fiber
rancid fiber
#

what would you suggest if you don't have a GPU but still want to train now that Disconnected is gone?

low shard
# rancid fiber what would you suggest if you don't have a GPU but still want to train now that ...

Train (make) RVC Models on cloud:

  1. Prepare the Dataset
  2. Setup RVC:
    Choose a cloud way to use RVC,
  • Google Colabs (max 4 hours of daily T4 16gb gpu not granted for free, not much hours for training, but easy to use, there's a paid tier):
  • Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus, either T4x2 16gb each or P100 16gb, only free):
  • Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly, Free Studios run 24/7 but require restart every 4 hours. There's a paid tier):
  1. Be sure to know about the tensorboard

Google Colab = Easier but risk of getting disconnected
Kaggle = Harder but way more gpu time
If you are looking for the easiest way and for free, is using https://weights.com/ which ofc uses RVC

RVC Inference (use models) on pre-recorded audio on Cloud

You can use either:

low shard
low shard
low shard
low shard
patent trellisBOT
# low shard !howtoask

How To Troubleshoot AIHC_WaitWhat

__**GIVE CONTEXT.**__ 📝
  • Don't simply mention your issue, like "my rvc is not working".
  • Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
  • The more context, the better.
__**BE POLITE.**__ <:matsuripray:1159685390156967936>
  • Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
  • It's okay if you're frustrated, but don't take it into this server.
  • Don't DM without prior consent.
__**BE PRODUCTIVE.**__ 🤝
  • Don't ask for every little instruction. Put your own effort & test things by yourself.
  • Don't ask to ask.
  • Check if your answer is a Google search away/on our guides website.
If you've changed your mind about your message, feel free to delete/edit it.
royal grove
# low shard elaborate

How do I make female voices sound more like human and not ai like the monotone background of the ai

low shard
royal grove
#

RTX 3080

#

cant send pics

#

or files

low shard
low shard
#

also what tutorial link are you following?

royal grove
#

Real time

#

I didn follow any tutorial the only one i used was to actually download wokada

low shard
# royal grove

lemme guess, you found the link via a youtube tutorial?

royal grove
#

yeah

#

this one

low shard
#

video tutorials get outdated easily, and in fact this is an old version of original wokada lmao

#

thats old asf

royal grove
#

oh really

#

damn

low shard
royal grove
#

shit

low shard
#

plus vb audio cable gives issues and can randomly stop working as users told us on windows

#

forget everything you get by video tutorials

royal grove
#

alright

#

what do i do

low shard
#

wokada deiteris fork is way better

royal grove
#

should i delete the vb audio cable

low shard
royal grove
#

how big of a difference is it

#

like massively better

low shard
#

its like trying to run windows xp in 2025, shittiest performance you could get and you're missing out on options to get better quality

royal grove
#

thanks bro

#

ill download this one

#

can i keep the voices when i delete everything?

low shard
# royal grove

that's just a link to the github repo, you need to read the full guide

#

if you dont read it you will just fuck things up lol, dont just click the first thing you see

fair thistle
#

is it supposed to open up in your browser

royal grove
#

which one do i download?

#

they are all AMD

#

@low shard

#

is this not required to download

#

because after downloading the nvidia it opened a version of the voice changer on my browser

#

do i use that one?

#

nvm i figured it out

#

thanks

low shard
low shard
#

@royal grove dont go in the github, read the guide i sent you

lethal grail
#

where do i put my downloaded models

rancid fiber
simple ore
opal cobalt
#

Just wanted to know anyones thoughts on Renting a Cloud GPU for a few hours specifically for building a model that im unable to build locally but will be able to run locally once built. (havent seen this topic get mentioned what i see is those rent for environment due to specs)
Use Case Qwen2.5 32B -Instruct fp8 Quant GOAL
Extra Context my Specs:
CPU: AMD Ryzen 7 7800X3D (8 cores / 16 threads)
GPU: NVIDIA GeForce RTX 4090
Motherboard: MSI MAG B650 TOMAHAWK WIFI
RAM: 64GB G.Skill Flare X5 Series DDR5

simple ore
#

oops, sorry

#

Q8_0 wont do

opal cobalt
#

thanks im trying to do through tensorRT LLM and go through the building process through trtllm-build but it requires x3 x4 so my thought process is this.
Quantization Optimization Strategy
Testing Protocol (Shoot for the Best, Work Downwards):
Priority 1: FP8 Validation
✅ RTX 4090 Support: Test FP8 tensor operations compatibility
✅ Performance Benchmark: Measure speed vs memory vs quality
✅ Stability Test: Ensure consistent outputs and no crashes
Priority 2: Enhanced AWQ Evaluation
✅ Calibration Quality: Test 1024 vs 512 calibration samples
✅ Block Size Impact: Compare 64 vs 128 block sizes
✅ Mixed Precision: FP8 KV cache + AWQ weights performance
Priority 3: Baseline Confirmation
✅ Standard AWQ: Ensure proven configuration works as expected
✅ Fallback Readiness: Validate backup option performs acceptably

#

goal use cloud gpu specs like A100 80GB to build then it will fit within my 24GB VRAM if that makes sense

simple ore
#

Q5_0 will fit

#

8192 context is a bit small, but it may be fine with it offloaded into RAM

#

Q5_0 model size is 21GB

opal cobalt
#

my apologies for lack of context i aim for 32k context length with this i used Q4_K_M previously but i need improvements because that uses llama.cpp which is like 75%-80% compared to tensorRT
Decision Matrix:
Metric
FP8
Enhanced AWQ
Standard AWQ
RTX 4090 Support
Test Required
✅ Proven
✅ Proven
Expected Speed
🥇 Best
🥈 Better
🥉 Good
Memory Efficiency
🥇 Best
🥈 Better
🥉 Good
Quality
🥇 Best
🥈 Better
🥉 Good
Risk Level
🔶 Medium
🟢 Low
🟢 Minimal

Selection Criteria:
RTX 4090 compatibility (must work flawlessly)
Performance improvement over current Q4_K_M (minimum 6x speedup)
Memory efficiency (must fit in 24GB with overhead)
Output quality (must maintain coherent responses)
Stability (no crashes or artifacts during extended use)

#

Executive Summary
Goal: Build an ultra-optimized Qwen2.5-32B-Instruct model using W4A8_AWQ (but aiming for FP8) quantization via cloud GPU, then deploy locally on RTX 4090 for maximum performance with 32K context support.
Problem: RTX 4090 (24GB) cannot compile TensorRT engines for 32B models due to memory constraints during build process, despite having sufficient memory for runtime.
Solution: Use cloud GPU (A100 80GB) for one-time engine compilation, then deploy locally.

#

Primary Goal (Best-Case Scenario):
Build Qwen2.5-32B with FP8 quantization achieving:
✅ Speed: 70-110 tokens/sec on RTX 4090 (10-20% faster than W4A8_AWQ)
✅ Memory: ~12-16GB runtime usage (more efficient than AWQ)
✅ Context: Full 32K token support
✅ Quality: 98%+ of FP16 performance (floating-point precision advantage)
Secondary Goal (High-Performance Fallback):
Enhanced W4A8_AWQ quantization achieving:
✅ Speed: 65-95 tokens/sec on RTX 4090
✅ Memory: ~14-17GB runtime usage
✅ Context: Full 32K token support
✅ Quality: 96%+ of FP16 performance
Tertiary Goal (Proven Baseline):
Standard W4A8_AWQ quantization achieving:
✅ Speed: 60-90 tokens/sec on RTX 4090
✅ Memory: ~15-18GB runtime usage
✅ Context: Full 32K token support
✅ Quality: 95%+ of FP16 performance
Sorry if is too much context just trying to share relevant details after ideal build is complete i plan to use with anythingllm and setup draft model,embedding model,vector db etc these options will just beat the slow Q4_K_M 32k context speed i was unsatisfied with

full moss
#

how to do text to speech?

#

someone please help?

crystal girder
#

guys why my app crash everytime i tried to use voice ai

latent kettle
full moss
latent kettle
#

Your cpu gpu and ram

full moss
#

i7 4079 geforce 32

latent kettle
full moss
#

17 cpu

#

4070 gpu

#

and 32 ram

latent kettle
#

Ohh I see

#

Your are good to go

full moss
#

yes but how do i use the text to speech

latent kettle
#

You can use kokoro tts, f5 tts

full moss
#

where

latent kettle
#

On your system, install it

full moss
#

the what

#

what do i install

latent kettle
#

Lemme send you guide

#

@full moss

full moss
#

yes

#

but i cant create the voice

#

freemium but its paid to create a voice

latent kettle
full moss
#

no i wanna create an text to speech model

#

i got the singing mp3 already

latent kettle
latent kettle
simple ore
#

I would not suggest that, edge tts in applio is purely for demo purposes. There are better tts available. Edge is just a screen reader for websites after all.

scenic arch
#

is there any benefit to use nvidia broadcasts echo and noise removal as opposed to using okada's builtin echo and sup1&2?

simple ore
scenic arch
simple ore
#

mic -> broadcast app -> voice changer

#

both gonna use rtx cores on gpu, have not tried it personally.. should be fine on a newer gpu

#

version from 5 years ago, should be even better now

scenic arch
#

whats the best tts for rvc also

low shard
# scenic arch whats the best tts for rvc also

There are different Text To Speech (TTS) AIs:

GPT So Vits: RVC isn't as good as GPT So Vits for tts, but gpt so vits (few shot tts, which means needs just a lil training for models) can't use rvc models (and viceversa), and its only limited to: english, chinese, Cantonese, japanese & korean, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/

Freemium 11labs: Easy way to do TTS is https://elevenlabs.io/, you can't use RVC model on this but its a mostly premium easy way for good quality TTS

FishSpeech: FishSpeech is a 0 shot (no explicit training needed) TTS, if you got a good pc you can use it locally else use their site

You can check TTS in our tts index

With RVC Models:

RVC is natively for Speech To Speech, but forks such as Applio have built in tts (using Microsoft Edge TTS to make a generated tts audio, which i suggest you to choose a tts model that is the same gender and language of the rvc model you wanna use, and then convert it with rvc)

If you wanna do tts locally with RVC Voice Models (if you got a good pc):

  • You can get Applio in our docs

If you don't got a good pc you can do tts with RVC Voice Models on cloud:

tranquil schooner
#

Question about W-Okada voice changer. If I still have an older version, is it not supported anymore or work properly? Because I've been having trouble with my gaming laptop I just got a year ago after graduation that worked just fine until recently last month went I was minding my business on Minecraft and it closed my laptop and never turned back on so Geek Squad looked at it and said its a faulty battery issue they "fixed" but after it returned home to me three days ago and only got 5 minutes of playtime for updating, clearing storage space and some games and it turned back off on me after opening Google so my father had to return it and Geek Squad said they had a feeling he'd be back like they were expecting it and now they're saying it can't be fixed and I'm forced to buy a new laptop after my father used up the protection plan. Anybody know?

tranquil schooner
#

I would like an answer before the laptop arrives in two days...

simple ore
#

Ah, the Geek Scam

#

You either learn how to diagnose and fix your PC or you pay thru the nose for placebo fixes.

tranquil schooner
#

So the voice changer is not responsible? And I knew Geek Squad was a scam from a different pc expert actually opening up the devices to look inside and fix things and replace the motherboard but my father would not listen

brisk plover
#

guys?

#

can you help me out?

#

how can I use okada

#

on discord

tranquil schooner
# brisk plover how can I use okada

So idk if that's a good idea for now until I figure out if my problem with my laptop dying and not turning back on is caused by okada or not...But just saying in case it is

simple ore
viral mason
brisk plover
#

can u help me out?

viral mason
#

Yeaz in dms

brisk plover
#

i already have cable device

viral mason
#

Just gimme a minute since I'm not home

tranquil schooner
tranquil schooner
#

Wrong one but yeah-

latent talon
#

Okay weird things are happening and idk why.

Specs:

Cpu: 13th Gen Intel Core i7-13700K
Ram: 64 GB
Gpu : RTX 4070

So I've tried using the latest W-Okada, and the one from a year ago. The newest tends to break then not work at all, while the old one gets worse over time, basically cuts in and out and fails to make any sound at all.

I'm using it for streaming, and changing my voice in game, so I expect delay, No matter what it comes out to about 4 second delay, then gets choppier and choppier. Any ideas on what to do? I can try and gather more info later today like logs and so on.

Thanks in advance, I'm not the best when it comes to tech so even obvious fixes are welcome

pastel oak
odd valve
#

whats the best rvc for an amd gpu

#

or is okada better?

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)

odd valve
#

want to try out realtime

#

so should i go with the fork?

low shard
odd valve
low shard
plush stream
#

guys please can i ask for some free ai websites except weights gg i need them

odd valve
#

so for my gpu do i use the nividia link?

simple ore
odd valve
simple ore
#

6000 what

#

AMD?

#

or Nvidia RTX 6000?

odd valve
simple ore
#

you download the AMD version of the voice changer from the corresponding link

true heron
#

How do I get started?

latent talon
#

Welp Idk what I have managed to do, but now the old one is borked

swift thunder
#

32 bit float or 64? to train models

20250605_185537.jpg
winter dew
#

how much does mic quality matter? I feel like I can’t get w okada to sound that realistic

#

im using a hyperx solo cast so like about average

simple ore
simple ore
winter dew
#

it still just sounds a bit unnatural ngl I don’t get why

#

you’re right I read that on the fork site I did the crackle fixes as well but it just sounds weird

swift thunder
timid bronze
#

Can someone please point me the right direction on how to get consistent characters in text to images. I am already including as much detail as possible in the prompts. I have tried so many different AI tools already.

simple ore
simple ore
swift thunder
#

At what values do I see the graph better, my friend?

timid bronze
simple ore
simple ore
swift thunder
timid bronze
simple ore
#

if you're using avg_50 charts, they are smooth enough to use 0.5

#

with more epoch the grap itself smoothes out

#

since tensorboard does not really shows every logged value

swift thunder
simple ore
#

for old loss graph 0.987 was necessary because they were so random

swift thunder
#

if I use avg_50

lost ember
#

do yall have any voice changer that i can use the models with?

long obsidian
#

can someone help me when im speaking sometimes the voices make a robotic sound with which setting can i avoid that - using okada

pastel oak
pastel oak
long obsidian
#

i cant fix it tried changing the chunk but it doesnt work

pastel oak
#

Do you have the 2nd nvidia link specifically for rtx 5000

long obsidian
#

let me check

pastel oak
#

Its a separate version

long obsidian
#

deiteris fork?

pastel oak
#

Yes

long obsidian
#

i dont have it

pastel oak
#

Get it

#

Its better than original atm

#

And original cant run rtx 5000 gpus iirc

long obsidian
#

so i need to install voice-changer-windows-amd64-cuda.zip.002
?

#

i read that i need to download both 001 and 002 and then unzip them

long obsidian
#

RuntimeError: CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

@pastel oak it said this

#

what am i supposed to do

#

it works when im using my cpu but doesnt let me use gpu

pastel oak
#

What youre downloading is not for rtx 5000

long obsidian
#

ye i followed the steps

#

oh i see

#

i thought the gpu is better for voice changer

pastel oak
#

I dont think you understand what im trying to say just download the 2nd nvidia link

long obsidian
#

@pastel oak with my gpu do u have idea for what chunk extra and f0 to run it on (last question sry for being annoying)

pastel oak
#

No

pastel oak
#

If the "perf" on the graph the one in green color is a low number like 30, then decrease chunk to around 100 if you want less delay

long obsidian
pastel oak
long obsidian
#

oh okay thanks a lot

#

also how do u delete models bec when i go to edit i dont see it

knotty moth
long obsidian
#

thanksssss

hazy dune
#

help, I installed the voice changer and virtual microphone correctly in the voice changer, I also installed the virtual cable on the microphone correctly in the discord, I start, I say, it doesn't work, but when I turn on the loud video, the voice changer perceives it as a voice, and changes the voice to the video, in general, instead of the microphone sound, the voice changer changes the sound of headphones

#

(im using amd version, im had amd graphics card)

hazy dune
#

and now, it captures both the sound and the microphone, what should I do?

pastel oak
#

Cutting off is you getting too quiet at the end of sentences, keep in. Sens. Threshold further to the left if you moved it to the right

#

Do you mean crackles with distoetion

#

No the in. Sens. Under F0

#

And check crossfade length in advanced settings and bring it to 0.10 if its lower than that

knotty moth
pastel oak
#

Tooltip and guide has brief explanations but roughly explained it constructs the voice more clearly the higher it is but adds delay

obtuse pagoda
#

Whenever i try to use the deiteris fork on kaggle, it manages to process and get the server ready, but when i click on the link it says "this site can't be reached" is this just a network problem or something?

hazy dune
hazy dune
#

there is no access

hazy dune
knotty moth
#

!give-media-perms 30m @hazy dune

snow sphinx
#

Hey all - just joined here so apologies in advance for any repetitive questions.
I'm pretty new to AI, so is there any material / videos that anyone could recommend,
I'm specifically struggling with getting ChatGPT to recall and give me an accurate time.
Anyone had similar issues or any advice on how to resolve?
I'm also interested to understand if / how I can link up my various platforms to create an autonomous set of AI agents who need minimal human supervision or direction?
Thanks in advance

simple ore
#

at best it can repeat something you said from its context

#

or there can be some patches added to the processing, like storing things you told the model to remember in a special context or running queries without using the llm such as "what time is it now?"

knotty moth
knotty moth
#

if the voice changer is being used, the settings are greyed out

worn walrus
#

hey everyone - i just had a question

#

so everyone probably knows how people are using temp student emails to gain free access to veo 3

#

but i had some concerns - what if google finds out? this may be a dumb question but do u guys think they will charge the cards for the full 15 months

#

?

simple ore
#

You are trying to do the most compute demanding tasks one can think of.. for free

worn walrus
#

this guy...

#

guys.. its a FREE plan

knotty moth
worn walrus
#

wdym tho

#

it is free

#

for students

#

in college

simple ore
#

are you a student?

#

you are trying to steal a service that is being provided conditionally by google with certain expectations

#

also be aware that google is logging anything and would report you if you attempt to do anything below the board

worn walrus
#

sorry i dont ever recall saying i was going to use it - i was merley asking a question. so pls mind ur own business

lime patrol
#

what to do if the ai itself says TRAIL

low shard
#

what's ur pc gpu? what do u want to do? what tut link are u following?

vital merlin
#

how can i sound like a waifu

#

perfectless

#

i wanna sound like a mommy

lime patrol
vital merlin
#

i use a ryzen 5 5500

#

and a 3060 TI

low shard
#

I hope you didnt use a youtube tutorial, since video tutorials are outdated asf

lime patrol
lime patrol
low shard
#

they use an over year old software

#

-realtime

patent trellisBOT
# low shard -realtime
💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

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

• **RVC GUI Mainline Realtime**

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

low shard
#

read the 1st link

lime patrol
low shard
crude flame
#

its better than deiteris' fork

low shard
crude flame
#

better performance

#

same quality

low shard
crude flame
#

lyery has also messed with it i think

viral mason
sand bison
sand bison
paper bloom
#

is there any way to change the accent on owakada? like even when i speak my first language it sound like i got an accent. is it using like an asian accent ?

simple ore
#

if you increase the index use it would use a blend between the voice model and you, at 1 it is the voice model's... but there's performance drawback from using the index

sullen lion
#

illustrious lora training parameter question

#

why the fuck is this

image.png
#

when i

image.png
#

kohya if you couldnt tell

simple ore
#

what's the source for 1st screenshot?

simple ore
#

in case you have not seen github ticket

viral cape
#

where i download the voice mod?

random latch
#

what kind of files does the beatrice vst take? I'm trying to put a model in but it shows .toml files

#

Also can someone convert an rvc or pth file to toml

sullen lion
#

and i cant tell if it was pushed or nixed

winter dew
simple ore
#

but if your performance is fine, go for it

simple ore
winter dew
#

do you find you could tell if someones using it?

#

im trying to get it to sound realistic but it just doesnt lol

analog obsidian
#

index doesnt make things realistic, its just where the accent of the model is stored, so it kinda makes the model sound more truthful to the original dataset

#

for a more realistic result train a big dataset cat_dance

winter dew
#

makes sense

#

do you recommend any natural sounding female models?

#

searched through the channel for it but maybe you have the secret weapon

random latch
#

can someone help me

simple ore
#

this question.... 50 times a day

random latch
#

im new

crude flame
winter dew
#

so many egirl models and no resources on natural ones

#

makes sense why people ask considering this is why most people go to voice changers

analog obsidian
#

its easy

winter dew
#

yea but why?

#

rvc seems to have been out for a while so id be surprised if no one’s made a good one

analog obsidian
#

i have natural models, it just takes time

winter dew
#

like I don’t really mind going through making one but I feel like someone else would have made something a lot better than I could’ve

winter dew
#

if it’s not a lot of manual work that’s fine for me tbh

sullen lion
analog obsidian
winter dew
#

bc I mean I could try making one that takes longer

analog obsidian
lime crag
#

Do people still use applio for model making? I really haven't heard anyone mention it for some time.i also don't think it's been updated for a few months

simple ore
#

What else is there lol?

#

Sure you can go back to the mainline

lime crag
#

I guess applio is still the best option

winter dew
#

now I get why there’s not many good models lmao

#

do you know any that are actually decent or do most just use what they make

analog obsidian
#

i use mine

winter dew
#

how much time would you say per day id have to put to make it sound good then?

#

if it’s like 20 min a day id consider it lol

analog obsidian
#

i think i was cleaning 40 minute per day

winter dew
#

oh damn

random latch
#

How does one put an rvc model in beatrice vst?

winter dew
#

I think another challenge is actually finding data sets lmao

#

idk where id even start with that tbh

#

wait so the actual length depends on the data set you give it?

#

like a 3 hour data set would be like 3-4 days

analog obsidian
winter dew
#

do you think a lesser data set like 4 hours is enough?

#

im assuming the data sets in the voice models channel were small ones

analog obsidian
#

not robotic but mid

analog obsidian
winter dew
analog obsidian
#

originally i had a 3 hour stream, that after cleaning, truncating etc, got me around 2 hours and 30 minutes??? can't remember

#

ah yes

#

but that person talks a lot in their streams so ye

winter dew
#

I see

#

so id probably need like at least 6 hours of data

#

or somewhere around there

analog obsidian
#

hold on i still have the 2 hour model

analog obsidian
#

so this is the 2 hour model

#

does kinda sound like him??? but not as accurate as the 5 hour one

crude flame
analog obsidian
winter dew
#

I didn’t even know something like this existed

analog obsidian
winter dew
crude flame
winter dew
#

okay

#

razer the problem is though these are like 17 mins each

analog obsidian
#

i would say 2 hours is enough for most models to not sound robotic

winter dew
#

are you saying take multiple videos

crude flame
analog obsidian
#

take multiple streams

#

maybe 3 or 4

winter dew
#

wait yea there’s like 7 hour dumps

analog obsidian
#

yeah u can just take that 7 hour alone

crude flame
#

but these mommy voice are mad monotone so they wont sound as good as a expressive voice

analog obsidian
#

true

winter dew
#

these videos have like breathing and rain and shit

winter dew
analog obsidian
winter dew
#

the monotone gives it away imo

analog obsidian
#

if u wanna train using cvec, no

#

but spin it's better at handling noise

winter dew
#

ill keep that in mind

#

yea ngl I need something that is more expressive

#

legit just a regular woman’s voice is what I need

analog obsidian
#

my workflow is using a noise gate to remove noise, then manually silencing every bad part

winter dew
#

LMAO okay

analog obsidian
#

but ye keep in mind very monotone datasets can't do much

analog obsidian
winter dew
#

yea it sounded like it

#

do you have any recommendations on getting larger data sets of women voices?

crude flame
analog obsidian
#

i only train male voices :D

winter dew
#

maybe like certain voice actors idk really what’s out there

winter dew
crude flame
#

you can also look up speed painting with commentary