✨│ai-help

[viral mason]

ikr?

[eager wadi]

i miss going onto someones computer while theyve gone to take a shit or whatever, swapping one of their apps to a fake virus file that just shuts down the computer after like 30 seconds and watching them panic

[formal wind]

Youre evil

[eager wadi]

i know :D

[cedar sky]

Hi guys, I've found this model https://github.com/Alibaba-NLP/WebAgent?tab=readme-ov-file on hugging face https://huggingface.co/mradermacher/WebSailor-3B-GGUF

Then I've deploy it on huggingface but I do not know how to make it work with internet. In docs said that I can just add an enviroment variables which I've did on huggingface but it still doesent have access to the internet. Can someone help me?

[formal wind]

For kaggle, when I try going to the file browser (The third link) it wants me to long in and I dont know where to sign up

[formal wind]

if you were curious, its this one

[simple ore]

!filebrowser users add "applio" "" --perm.admin

if that does not give you a hint...

[formal wind]

uhhhh

[simple ore]

username: applio
password: (blank)

[verbal nimbus]

guys

deiteris okada wont let me delete uploaded models

Error
unhandledrejection
no error stack
TypeError: Cannot read properties of null (reading 'modelSlots')
TypeError: Cannot read properties of null (reading 'modelSlots')
at i (http://127.0.0.1:18888/index.js:2:1249327)
at Object.updateModelInfo (http://127.0.0.1:18888/index.js:2:1251031)

how can i fix this

[formal wind]

Didnt work btw

[simple ore]

Well, you can change it to a different login/password

I have not tried myself

[formal wind]

I did

it doesnt work

[simple ore]

delete model slots and remake them?

close the app and just do it from windows explorer

[verbal nimbus]

ohh

damn

i forgot about it

[verbal nimbus]

thanks

[wise fiber]

uhh,so my laptop has a igpu,but can i use the rvc models for tts stuff

[latent cypress]

uhh, should i stop it?

"grad norm g = below 300 and going down, i personally noticed if grad g is over 400 the model has always some random problem"

[wise fiber]

basically,yk how u cant really use the realtime voicechangers without a gpu,cann you atleast use the tts services?

[simple ore]

I assume you're using codename's fork.

[latent cypress]

yea

[simple ore]

gradients do not tell you when to stop

it is just a measure to see whether something wrong is happening

like norm g going up and up and up into thousands and then 10 thousands+

with a regular model trainig this would mean the model is utterly fucked up

I said before, norm g going over 1000 is likely means the dataset is wildly different that the pretrain, like high pitch singing dataset using og pretrain

and if it keeps above 3-5k it is not gonna be good

usally 300-500 norm g is fine

[latent cypress]

so basically, I’ll just keep training, and make sure it doesn’t hit that number, right? (cmiiw)

and yea, my dataset’s full of high pitched vocals, and i used the klm 4.9 pretrain

[simple ore]

norms are your secondary concern

[latent cypress]

so what's my main concern? fm?

[simple ore]

yeah

[latent cypress]

ooo okay, thank you

[latent cypress]

there's two fm graph, which one i should take a look?

[simple ore]

it is only 1.5k steps

chill

[latent cypress]

got it

[wise fiber]

bumpy

[knotty moth]

try in another browser or with cleared cookies

[formal wind]

I got it working, I just needed to restart the session sadly

[knotty moth]

https://tenor.com/view/wtf-oof-yikes-no-value-the-critic-gif-10877641404121198102

[formal wind]

https://tenor.com/view/all-that-work-for-nothing-jk-simmons-lenny-turteltaub-bojack-horseman-gone-to-waste-gif-16133839

[wise fiber]

bump again

[kind ridge]

If I have only one minute of a dataset of a voice, and I want to train an RVC model, if I train enough will I get a good result?

[wise fiber]

can someone help :3

[low shard]

Do you still need help? if so, please elaborate

[low shard]

wdym? realtime voice changer or tts? what do you want to do exactly? those are 2 different things, and RVC doesn't mean realtime voice changer, it means Retrieval-based-Voice-Conversion

[wise fiber]

tts but with custom voice models

found in #1175430844685484042

[low shard]

RVC is meant for Speech To Speech, (STS), not TTS

There are different Text To Speech (TTS) AIs:

• GPT So Vits: Great Few Shots (needs a lil training) TTS, its only limited to: english, chinese, Cantonese, japanese & korean, if you wanna check gpt so vits instead, read https://docs.aihub.gg/tts/gpt-sovits/
• 11labs: Easy way to do TTS is https://elevenlabs.io/, 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 other 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 tts audio, 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:

• Ilaria RVC Zero

• Use Applio UI Colab (with google colab T4 free daily limit gpu)

• You could try another tts from our tts index and use the output as an input in rvc

I hope this helps, if you didn't understand something, you can ofc ask

@simple ore also talks about some other tts called chatterbox, but I haven't tried it yet so I can't say much

[wise fiber]

im confused on the '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'

[proven hill]

ilaria rvc zero on pinokio doesnt detect the gpu, im clueless and not on pc, someone know why so i can fix it?

[latent cypress]

the fm graph hasn’t going up at all, is this a problem with the dataset or the batch size i’m using? my dataset is 25 minutes long and i’m using a batch size 8

[low shard]

RVC is for STS, not TTS
So the only way to use it is via using another TTS model as a base to first make the audio, there are a preset of models in edge tts, if your rvc model is an american man, you choose an american man tts model as a base to not make it sound weird

[open pelican]

I Never Be Friends On Friend Request DM In Discord

[craggy bough]

that has nothing to do with this channel

[open pelican]

My Friends Is Gone

Look

[craggy bough]

that has nothing to do with this channel

[simple ore]

what tts language do you need?

[shadow flax]

could someone tell me why it doesnt do anything (when i add any model, its marked red aswell)

[flint vapor]

If is marked red means is selected

[flint vapor]

In what sense it doesn't do anything

[shadow flax]

works now, just not well

im at work now anyways so i cant tinker rn

[open pelican]

How Dare You

[knotty moth]

https://tenor.com/view/cant-see-who-cares-looking-around-linus-tech-tips-linus-ltt-gif-25137214

[shadow flax]

ltt gif is wild

[analog obsidian]

batch 8 is a pretty small batch size so things are very slow, just wait, fm will eventually go up

[latent cypress]

how much should i set the smoothing to see the fm graph?

[simple ore]

with avg_50 charts you dont need to smooth.. maybe 0.5 max

[simple ore]

it is more or less stable, which is surprising

but it is high

[latent cypress]

so nothing to worry about, right?

[analog obsidian]

thats too high

[latent cypress]

the beginning?

[analog obsidian]

everything

[latent cypress]

[analog obsidian]

fm usually have values of 7-10

[knotty moth]

perhaps normalization issue

[analog obsidian]

idk but something is wrong

[knotty moth]

for KLM pretrain, maybe try -23 lufs

[analog obsidian]

slice normalization is better

codename fork doesnt have that

so either download applio main branch or do the slicing in mainline

[knotty moth]

im fine using applio main branch

[analog obsidian]

in the main branch slice norm is named "post" in the preprocess settings

[latent cypress]

so first I sliced the audio, removed the silence, merged it back, then in rx 11 I used the leveler with -14 threshold and normalized it to -1, is it wrong?

[analog obsidian]

yes

just get the regular clean dataset

and slice it using mainline or the new applio branch

[low shard]

i mean codenames fork is experimental, i think noobies already told u that when yall talked about it in #🔥│model-maker-chat and ur model exploded

[knotty moth]

per slice normalization would resemble of the og mainline

[analog obsidian]

thats what im talking about

it gives better results

rvc boss added it for a reason

[latent cypress]

yeah, my bad. i was just still curious about the codename fork, that’s why i decided to keep training there

[analog obsidian]

if you only use the default settings its kinda pointless to use it

mel similarity is mel loss

and you dont need more grad norms graphs

[low shard]

just use applio tbh

[analog obsidian]

validation loss is good tho

applio main branch should also have it

[latent cypress]

okay okay, i'll go back to applio, thanks for the info guys

[analog obsidian]

no problem, good luck

[latent cypress]

oh yeah, if you don’t mind, could you share the step-by-step process? just so i don’t mess it up again. or is it already in the docs?

[analog obsidian]

1. clean the dataset, dont touch the volume, just remove bad sounds
2. (optional) resample using rx11 and truncate silence
3. preprocess in applio, in the advanced settings use "auto" and in the normalization option select "post"
4. pitch extract
5. train

[quasi rover]

anyone can tell me a good voice to AI colab or huggingface repo for models saved on huggingface?

[analog obsidian]

d/adv = going down
mel and kl = going down
fm = going down for a couple of epochs then going up
g/total = this is a weird one, just ignore it
grads norm = both going down, grad g should be between 100-500

[latent cypress]

tysm

https://tenor.com/view/kiyobean-monniethebeanz-beanz-bean-thank-you-gif-345771613568113253

[tepid moss]

Hello! Last time I asked about the advantages and disadvantages of using AMD cards against nvidia ones.

Still keeping with the theme of hardware, I have questions about usage.

I would like to know how the session of selecting the hardware to run works. How do we select on when models should use GPU or CPU and to use RAM or VRAM?

[quasi rover]

which pitch algorithm is better between crepe, rmvpe, rmvpe+, harvest or pm

[open pelican]

My Friends Request DM On Discord Is Gone 😭

@delicate cradle

Take It Look

[hallow thistle]

What is this supposed to mean? Are you looking for a FNF VS mod or something?

[hallow thistle]

rmvpe. I'm not sure about the rmvpe+ one.

[analog obsidian]

too high usually leads to the model having some sort of artifact, maybe a slight robotic sound, bad sibilants, etc

it's all random

it means your generator is struggling

no problem, like noobies said above, what seems to cause the grad g to be higher is when the dataset is too different from the pretrain

[viral mason]

What are you trying to do exactly? Make an ai voice or use one as a voice changer?

[quasi rover]

I fixed already thanks though, found this one https://huggingface.co/spaces/TheStinger/Ilaria_RVC

[viral mason]

Cool 👍

[open pelican]

Yes

[hallow thistle]

There are a lot of Google Colab and Hugging Face spaces that can do RVC, one of them being Ilaria RVC on Hugging Face as you mentioned.

[viral mason]

Try not communicating with him it's like taking to a drowning fish

[open pelican]

I Hate Fixed Movie On Netflix 😡

[hallow thistle]

That's goofy. Imagine looking for a FNF mod in help channel.

[open pelican]

Is New Maeko By Seanleddy167

I Drawing To New Maeko On Ibis Paint X

[viral mason]

Good job buddy

[hallow thistle]

Can't help with that since I don't play Friday Night Funkin. This channel is about about AI programs like RVC and W-Okada, that's one of my best I would help with. For a normal topic not strictly related to AI, it would be better to go to #✦│chat.

[viral mason]

He is going to respond with something completely unrelated to what you said now

[hallow thistle]

Yeah, I'm aware he's just a kid on Discord. Not too deep about that.

[serene cove]

Anyone here with like a working knowledge on ai short form content creation? I've been trying to create shorts all day using chatgpt, Sora, runway etc but I've got nowhere, as it never really created what I was looking for

[patent lance]

It's better Wokada Deiteris Fork or Vonofox actually? About the quality/realism.

[viral mason]

Pretty sure Vonovox is better but I'd wait for the new update

[patent lance]

Oh yes, why?

[viral mason]

Why wha

[patent lance]

Why waiting for the new update?

[hallow thistle]

Do you mean "which one is better between Deiteris fork W-Okada or Vonofox"? Definitely fork W-Okada.

[patent lance]

Exactly, about the quality / realism of the voice.

[viral mason]

It'll allow adding index files so then there can be a true comparison between both voice changers

[patent lance]

Index not really work correctly no?

[viral mason]

It's Vonovox right? I hope I haven't been mispronouncing it

[viral mason]

No it's just there isn't an option yet in the current version

[patent lance]

I never use index with okoda, because it's not very good for me.

[hallow thistle]

If you ask me "why" though. On Vonovox, its UI is kind of professional, not really user-friendly interface as W-Okada. https://github.com/dr87/Vonovox/blob/main/docs/images/ui1.png?raw=true

[patent lance]

I was just asking about the quality of the models 😄

[hallow thistle]

Nah, I interpreted your earlier questions like that.

[patent lance]

I wanted to know which was better, in terms of model quality. 🙊

[viral mason]

Bro isn't answering the question

[patent lance]

In terms of realism etc

[analog obsidian]

both have the same quality bro

[patent lance]

Oki

[latent cypress]

i though it was rvc gui lmao

[viral mason]

[low shard]

Please, stop asking help for that

[flint vapor]

This guy will never stop, more we say it to him more he continue😭

[low shard]

What are you trying to do? AI Covers? E girl trolling/catfishing? TTS? Roleplay?

[patent lance]

Roleplaying on RedM

[low shard]

great as catfishing is bad, what's your pc gpu? vonvox is supported only on nvidia gpus on windows

[patent lance]

RTX 4090 for gaming, RTX 4070 on a second computer for voice changer only

[low shard]

both would be good tbh, the major difference will just be the UI

[simple ore]

run the server on second computer, use ui on the main pc

[patent lance]

Hum I see, index is not really good with real time right?

[low shard]

can help with the trained accent, but cause autotune effect

[simple ore]

if you run it on second pc it may be fine performance-wise

[patent lance]

I am french with a french model

My settings

[low shard]

turn disable jit compilation on, and crossfade lenght to 0.15

then ur all done

[patent lance]

Quality will be better?

[patent lance]

When Disable JIT compilation is on, more delay is added right?

[low shard]

actually, it helps with less delay, but takes a few seconds more at startup

[low shard]

crossfade lenght 0.15 helps a bit with quality

[patent lance]

gonna try

[low shard]

you can't do much more than this tho

[patent lance]

I am yellow / red now with disable jit compilation on

So I need to add more chunk, mean more delay right?

[trail spade]

Hey y’all, I’m Masresha (or M). I work as an estimator in construction, but I’ve also done field work, so I’ve seen both sides of the industry.

Been thinking a lot about how outdated some of our processes are, and I’ve got an idea that could really push things forward using AI.

Before I dive into it, I just wanted to ask:
How feasible is it to train or use AI to read structured files like blueprints or technical drawings and generate something visual + interactive from that?

[low shard]

this seemed to have been caused by the crossfade lenght 0.15, yeah, put it to like 60 or 90

[trail spade]

Hi Nick, you seem like you might be able to help me out, anychance?

[patent lance]

It's too much for roleplaying

[patent lance]

off and crossfade 0.15 I am fine

[viral mason]

best bet is just ignore him, he doesn't understand basic human conversation

or anything at all in fact

[latent kettle]

He is still asking for that

[viral mason]

he'll never stop

[latent kettle]

Someone please give him whatever he wants

[viral mason]

he showed me the audio it's literally a chromatic scale

if u dunno what that is just think about pressing the keys on a piano from lowest pitch to highest

very slowly

one at a time

but instead of a piano it's the "voice" of the character

[south fern]

anyone really smart with Voicemeeter?

i need help

nvm i figured it out

[latent cypress]

https://tenor.com/view/anime-unclephuc-spit-aho-girl-gif-8521254120470346556

[tropic glade]

whenever I start a model my pc stars stuttering even when not talking, but nothing is utilized even at 50%

[low shard]

voicemeeter isn't suggested, don't use yt tuts for realtime voice changers

[low shard]

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

Please Elaborate:

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

Please Elaborate:

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

[south fern]

i ended up using two Virtual cables for what i needed to do lol

was trying to get ai voice to work with a TTS to work for a Voice into a game

now i can talk to people as bmo lol

[tropic glade]

AMD Ryzen 7 5700X3D 8-Core Processor
RX 6600 XT
Screenshot: https://i.imgur.com/VD9wOKN.png
I just used the link in guides, and got the first client.

[low shard]

ohh nvm then, thats great

[low shard]

first client
that's just called wokada deiteris fork, not client

also what are you trying to do?:

• AI Covers
• E Girl Trolling/Catfishing
• Roleplay with the realtime voice changer in a game
• TTS
• etc

To be sure you're using the right program

[tropic glade]

Rp

[low shard]

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

Please Elaborate:

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

[low shard]

alright, so roleplay in a game?
set f0 to rmvpe onnx
extra 2.7

on wokada deiteris fork, you can **optionally **use more advanced settings for benefits:

• Advanced Settings -> Force FP32 mode: on (THIS IS OFF BY DEFAULT! Turning this on improves stability. Increases VRAM usage by 200 MB)
• Reduce the delay on Windows via: https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#reduce-more-delay

Set game graphics to the lowest 1080p 60fps, what game are u playing?

[tropic glade]

FiveM

I have a 1080p 240hz monitor

[low shard]

set graphics to the lowest, play on 1060 and 60 fps locked, try the settings i told you, close other programs in the background

[tropic glade]

And if I want to remove noise when a sentence ends or when im breathing which one is best from the three? Echo Sup1 and Sup2?

[low shard]

echo is only for echo issues
sup1 sucks
sup2 would be the best for this case

Also, please be sure to elaborate, and to not use yt tuts, they are outdated

!give-media-perms 1h @tropic glade

so you dont have to use imgur

!give-media-perms 1h @lavish sequoia

so you can send the screenshot of the program

[tropic glade]

you can give the perms permanently, I dont bite 😭

[low shard]

Nah they bite

Jk but those are supposed to be temporarely just for helping

because to get the perms you should be level 10 by chatting, so we make an exception only for helping

[tropic glade]

btw if I have a model can I make it less robotic by changing some parameters, or do I need to retrain it?

[low shard]

I gave you already the best settings possible, you could also try setting crossfade lenght to 0.15, but if it's robotic its best to just retrain it with a better dataset and following the tensorboard

[analog obsidian]

need to retrain it

realtime client settings do not change the quality of a model really

[pastel oak]

@paper lotus have you dealt with the issue yet?

[low shard]

@radiant coyote This server is english only, please ask for help in english

that depends A LOT, you need to elaborate the things I asked you, there are many programs and depends on your hardware and what you want to do

It's very crucial info

Please reply to the stuff I asked so I can help you out

[low shard]

it all depends to the GPU and program and settings you're using, thats why im asking that info

onnxgpu? that's for amd

yep you most likely have an over year old version of original wokada from yt tuts

!give-media-perms 1h @noble kestrel

show a screenshot

you're using an over year old version of original wokada for nvidia.. that version is EXTREMELY OBSOLETE

it has wayy worse delay, wayy worse performance and quality

you deffo used a youtube video tutorial

and you also got vb audio cable

delete the zip, folder and uninstall from windows app settings

[distant idol]

Hey Guys! Not sure if this is the right Channel but i need some help with applio in Kaggle - got some errors 😭
Wanted to train my model locally but my AMD is not supported

[simple ore]

what AMD do you have?

and what errors on Kaggle?

[distant idol]

AMD Ryzen 9 7900X 12-Core

[simple ore]

that's cpu

[distant idol]

Sorry .. AMD Radeon RX 7800 XT

[simple ore]

you have two options - Applio using Zluda, Applio using a beta of ROCm torch

https://github.com/scottt/rocm-TheRock/releases/tag/v6.5.0rc-pytorch-gfx110x

[distant idol]

Which is better or friendlier?

[simple ore]

the zluda method has a guide https://docs.applio.org/applio/getting-started/installation#amd-gpu-support-windows

for rocm you'll need to install python 3.11 and install applio manually, not using run-install

i have not tried rocm method as I dont have 7000 series

[distant idol]

I will try this 🙂

[simple ore]

i assume you're using the latest drivers for AMD

25.5.1+

so there's a small change

[distant idol]

how can i check :)?

[simple ore]

adrenaline control panel

if you got 25.5.1+ install latest VC++ Redist, install HIP SDK 6.2.4

[distant idol]

Sorry... Where.. ? XD

[distant idol]

Got 24.12.1

[simple ore]

not a fan of updating? 🙂

okay, then the guide above would work

but and update will break it

[distant idol]

Not a fan of this information 😛 I'm bit quite familiar with all this xD

[distant idol]

Should i keep it at 24?

[simple ore]

up to you

[distant idol]

Server of Windows 10 &11?

[simple ore]

again, make sure you are downloading the right version... 5.7 or 6.1 for old guide, 6.2.4 for new guide

window 10 & 11, obviously

[distant idol]

They Guide you sended says v6.1.2

can't vind the compiled version of Applio v3.2.5

[simple ore]

3.2.9

[distant idol]

Sorry to be so hard 😦

it only says 6.1 not 6.1.2

[simple ore]

that's fine

[simple ore]

dont show your app URLs

anyone can access them

[young halo]

sorry

[simple ore]

your charts are f-d up

fm is going up, g total is too high

how big is the dataset and batch size?

[analog obsidian]

d got too strong, gen looks weak

[distant idol]

Don't have a feeling it's doing something ...

[young halo]

batch size is 4, and as for how big it is, idk if you mean how long is it or the file size

[simple ore]

how many minutes of audio

and what kind of audio is that?

[young halo]

like 11:45

and its FLAC

[simple ore]

i mean... is it like a character voice from some anime that was heavily processed?

[analog obsidian]

zluda training is very slow

[simple ore]

a robot voice

[simple ore]

it is compiling kernels, gonna take like 20 min

[simple ore]

not on 7800xt

[distant idol]

ah now its doing something

[viral mason]

What even is this

[young halo]

it's a kpop idol

[simple ore]

it will compile more, as training has more operations

[analog obsidian]

zluda

[viral mason]

What's zluda do?

[analog obsidian]

training for amd gpus

[viral mason]

Nice

If it's not for Nvidia I probably haven't heard of it most likely

[analog obsidian]

...

[simple ore]

it is CUDA emulator for AMD gpus

to run different AI projects from stable diffusion to tts, to rvc

[viral mason]

What I do wrong 💔

[analog obsidian]

nothing

[viral mason]

[distant idol]

Anyone suggestions about batch size, and how many epochs

[simple ore]

depends on the dataset size

[distant idol]

It's 35 minutes of singing (deverbed and acapellad through uvr5)

[simple ore]

you got 35 minutes, so anything from 4 to 8 is fine, may need to experiment / merge different models maybe

set it to 300e

[young halo]

do you guys think that 300 epochs are fine for my 11 min long dataset

[analog obsidian]

i'd use batch 32 for 35 minutes and train until the model sounds robotic

[proven hill]

stopped the training for you

[analog obsidian]

if i want to tryhard it i'd also train a batch 16 model of the same dataset, then merge it with the batch 32 model

[analog obsidian]

lmaooo

[distant idol]

how can i use my tenserboard (it says no boards to show)

[young halo]

lol

can u expain abt that

[simple ore]

and then you can train batch 4 and batch 8 models, merge them, them merge the merge with the merge of 16 and 32

and see it collapse into a black hole

[analog obsidian]

stable diffusion community has been doing this for years now lmao
every model out there is a merge of 200 models together so shit

[distant idol]

SO far, so great XD

[simple ore]

go make a lunch

or dinner

[distant idol]

It's 22:46 XD

[simple ore]

or to sleep

[young halo]

my model came out really smooth

i'll use kaggle more often

[mellow ermine]

-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

[distant idol]

It's only loading the pretrained models... does it really take this long? it's not even making epoches yet....

[simple ore]

as I said, it has to compile kernels for the new operations it found in training

20 minutes give or take

[viral mason]

do you wanna be an e girl or something like darth vader

[distant idol]

@simple ore Average speed of 2 minutes... is that oke or really slow?

[bronze mirage]

i have a question about this post (and reletaed ones in there): https://discordapp.com/channels/1159260121998827560/1364953969389994004
How does this work?
What is it exacly?
i see very realistic demos send in there and i am able to download a pth file, but i cant put that into my w-okada. so what is it and how is it used?

[simple ore]

2x faster than my old 6700xt

should be faster next release, I think 7800xt supports bf16

[distant idol]

Ah oke! What release :)?

[simple ore]

3.3.0

soon

[simple ore]

uninstall everything you've installed, - vb cable and other stuff

it is outdated guide and outdated software

[low shard]

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

Please Elaborate:

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

[fast scarab]

Hi everyone,
Has anyone successfully added SPIN (Spin Embedder) support to Applio 3.2.9?
If yes, could you share the steps or a patch?
Or is it better to just switch to codename-rvc-fork-3 since it has native SPIN support?

[low shard]

thats outdated asf lmao

oh shit i didnt see u my bad 😭

first, may i ask what you're looking to do? are you trying to do e girl trolling/catfishing? since the tutorial is about ai girl voice and i need to be sure its the right program u want

yeah u can delete everything, and are you trying to do e girl trolling/catfishing too like the other dude since yall were using the same tutorial?

ohh, im sorry for your gender dysphoria, and that's good that you arent trying to catfish because that's not allowed and can lead to illegal stuff

yeah no lol, everyone can make a new tutorial about windows xp, it doesn't mean its new tho

the tutorial uses an over year old version of original wokada, delete the folder and zip

and vb audio cable creates random issues on windows such as crackling, uninstall from windows app settings

after that

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

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

• Vonovox

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

• Original Wokada

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

• RVC GUI Mainline Realtime

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

[low shard]

please read carefully the 1st link, wokada deiteris fork, it has also the vac lite which is better than vb audio cable

there's also vonovox, they are basically the same thing, the only major difference is the User Interface and that vonovox got premium voice ffects like bitcrushing that wokada deiteris fork doesnt have

Dont worry about the dump, we love to help trans or people who just want to feel themselves, we just make sure to not help people who try to use AI for Harm such as tricking others to get money (nitro ekitten 😭) if u get what i mean

you're welcome, you could use either of the first 2, there isn't much difference than the things I explained

ofcourse we promote AI for good things

well, let me know

[simple ore]

you can put it into custom embedders folder

[low ibex]

One message removed from a suspended account.

[low shard]

replied in https://discord.com/channels/1159260121998827560/1397363566545338521

[cold shoal]

Hello nick can I ask you something

[analog obsidian]

to train with spin you need a pretrain

the original doesn't work with it, unless you finetune 30 hours of data

noobies has some spin pretrains, and also there's the klm one

[weak crater]

does anyone know an unlimited text to speech tool (I would love if it was free and won't mind if it's paid as long as it doesn't run on credits)

[simple ore]

what language?

[weak crater]

english

[simple ore]

do you have a decent gpu?

do you need voice cloning? expressive reading?

[weak crater]

4060 laptop

could handle offline voice generation

but i would love if it got emotions

[simple ore]

https://github.com/resemble-ai/chatterbox

give it a try

[cedar laurel]

how do i download voice changer client demo??

[quiet flint]

anyone know a lightweight few-shot tts engine like gpt-sovlites, but not as computationally intensive? thinking something that can run on my swiftshader

[velvet heron]

how would i go about making my own voice model? like what tools should i use to create it

[unkempt lagoon]

@simple ore @simple ore h.

[knotty moth]

I don't recall anything not much more demanding than zero shot TTS

if you want zero shot one, try chatterbox as shown in the link above

[knotty moth]

please refrain from pinging random person for no reason

[quiet flint]

Is it less computationally intensive than gpt-sovlites

[latent kettle]

Can we use it without input audio

[knotty moth]

even RTX 3050 laptop should be capable of

[raven sage]

holy fk im losing my mind with this i need help
so i found this channel that does dreamcore liminal storytelling things

and i was wondering where they got those voices from? (disregarding their 2 most recent vids and the backrooms one) i assume its AI but it could also not be im not too sure
heres their tiktok
https://www.tiktok.com/@liminal.depart

[quiet flint]

no gpu

what tts do i use that i can call from python and is few-shot

[simple ore]

yes, you'll get random voice

[brave marlin]

hi

[tropic glade]

idk why but no girl model ever sounds good for me 😭 I tried about every setting (mostly tune) but it still sounds fake af

[opaque geyser]

guys

why cant i choose

another model

like it doesnt show no matter how i click

[pastel oak]

This is a general AI server, we need details on what program youre using, for what and your GPU name

Original has bugged "Extra" setting which causes cpu overload for some people (unsure what causes it, some dont have this problem but a lot do), so theyre forced to use lower numbers so lower quality

Not related to quality but fork uses less hardware resources so your gpu has more power for games to play next to it ; or reaching lower delay

Thats usually another reason for people to choose it

[distant idol]

Oke guys i'm angry at my pc

it started up appearently and now only did till 80 epoches

how can i go further?

[distant idol]

Out of all my local training doesn't come any sounds.... i tried e40/e60/e80 @simple ore

[little igloo]

Can someone help me set up audio forwarding from one computer to another via LAN? I've already found how to do this and it seems to have worked, but I don't have a virtual cable in second place to output from voicechanger to the channel
in the virtual cable it says that the driver is installed, but it doesn't work
and voicemod 3 says that it was not possible to download the driver

[quiet axle]

Hi guys, im trying to host my vc on my new laptop but its not working? i already set the ip correctly for my main pc and my new laptop

[quiet axle]

nvm i did it, but when i try to activates it. I barely could use it. rtx 4060 laptop btw

[knotty moth]

then the cloud solutions are viable for you

the huggingface spaces are mostly zero shot

im not sure about GPT sovits for collab, or you could rent some gpu in runpod, vast.ai, etc

[quiet axle]

i dont get it, my 3060 laptop did wayyy better then this

[simple ore]

show the tensorboard

[distant idol]

Tensorboard is not working

Asked that yesterday

[distant idol]

Yesterday it wasnt working aswell

[simple ore]

at least send the events.out.tfevents file

[low shard]

So it's all solved right?

[low shard]

Sure

[low shard]

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

Please Elaborate:

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

[low shard]

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

Please Elaborate:

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

It has performance optimizations, it is true that it depends mostly on the model, but settings like force fp32 mode and crossfade length to 0.15 can help, which original wokada doesn't have

[low shard]

Rvc models aren't perfect, don't except to do any non natural sounds, anyways it's not a big deal, you're using just for roleplay as you said lol

[low shard]

Show a screenshot of the whole settings

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

Please Elaborate:

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

You're welcome

[safe grove]

Guys does any of you how to fix this I'm trying to resume a training for my model >.<

[knotty moth]

the sample rate setting must match the one used in the preprocessing and last training session, and also the pretrain used

[quiet axle]

actually im wondering if theres any software or drivers ishould download first before installing deiteris fork

[hallow thistle]

Virtual Audio Cable lite is needed if you wanna use with Discord. And make sure your GPU driver is stay up to date.

[safe grove]

It matches tho

lol

Ill just retrain using kaggle 🙂

[quiet axle]

thanks,

[safe grove]

Guys if my sample rate is 44.1K is it ok to put 40K since I cant find 44K option 😟

[hallow thistle]

"44000" doesn't exist in any RVC. Are you trying to train a voice model or something?

[safe grove]

voice model

did I analyze this wrong

[knotty moth]

if you think the quality is good enough, you can try going 48k anyway

but note that inference would take less time and memory with model in lower sample rate

[safe grove]

Its ariana's pro tools anyway

[frozen harness]

my goodness it sounds just like her 🫢

[safe grove]

LMAO

thats a sample dataset

😭

[frozen harness]

oh so it actually was her? 😂

[safe grove]

yes

thats her pro tools

session

😭

[frozen harness]

ah 😂 well hopefully your version will sound like that when its done!

[hallow thistle]

That's more like a dataset sample extracted from a song, not the AI converted one.

[frozen harness]

oh 😂 well hopefully the end result will sound super close

[safe grove]

I used 40K sample rate

Im using*

not 48K since it will take long

[knotty moth]

tbh it's also not wrong to go 32k and it would take less resources in both training & inference

so it could be good for voice changer

[safe grove]

okok

let me retrain

Again

hahahahaha

[tough fiber]

guys which one should i use for making rvc model

--rvc

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[latent kettle]

Thank you I'll try

[little igloo]

Can someone help me set up audio forwarding from one computer to another via LAN? I've already found how to do this and it seems to have worked, but I don't have a virtual cable in second place to output from voicechanger to the channel
in the virtual cable it says that the driver is installed, but it doesn't work
and voicemod 3 says that it was not possible to download the driver

[hallow thistle]

Are you using W-Okada? I can help about W-Okada and Virtual Audio Cable, not Voicemod and VB-Cable.

[tough fiber]

i was trying to use applio on my 5070ti gpu but i think theres cuda problem

anyone know how can i fix

[hallow thistle]

Did you follow this step? https://docs.aihub.gg/rvc/local/applio/#nvidia-rtx-5000-series-on-windows-precompiled-fix There's a way to make Applio to run with RTX 50 series GPU especially.

[tough fiber]

oh i didnt see that one

thanks

[hallow thistle]

Using regular Applio won't gonna work with RTX 50, so make sure to install Applio from start again with the exact but more additional steps.

[little igloo]

#🌏│русский message i use this vc

fork of w okada

can u help me pls

[tough fiber]

In the applio folder, we will open cmd and download the version of pytorch that is compatible with rtx 50. I think this is an extra feature.

[hallow thistle]

Yes, this is literally the Deiteris fork W-Okada. Is your PC GPU an AMD Radeon RX one or something else?

[little igloo]

rtx 4060 ti

[hallow thistle]

Try follow this step. https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#opening-on-multi-pc-setups

[little igloo]

ok

[hallow thistle]

Inside ".env" file would look like this. This is mine.

[tough fiber]

i tried but didnt work same thing. i was using another thing other than applio that was RVC something but i lost that one,

[hallow thistle]

If it didin't work, it can either be you missed a step or there's a problem with your GPU driver.

[tough fiber]

i installed that pytroch version

after i have to click run applio.bat right

[hallow thistle]

If you see Running on local URL: https:// 127.0.0.1:6969 on terminal after double click on run-applio.bat, it means it's all good. Without run-applio.bat, the program won't gonna run.

[tough fiber]

well its quitting after install

[tough fiber]

i cant see that one

this thing annoying omg

C:\Users\user\Desktop\uygulamalar\ApplioV3.2.9>env\python -m pip install torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 --upgrade --index-url https://download.pytorch.org/whl/cu128

i did this and the result

idk it didnt run applio

just exiting

[simple ore]

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

it install updated torch

then you can run applio normally

[hallow thistle]

This command is not where you run Applio the program itself. It's a command that runs a Python program named pip to install packages for Python. After it finishes downloading and installing, it simply ends there.

[simple ore]

.... what?

you run it from applio folder

the screenshot above shows it was done properly

[tough fiber]

yea i just realized my bad

but after i run applio on command line

still cuda error

nothing changed

[hallow thistle]

But he said "it didn't run Applio" after doing that?

[simple ore]

he did not

yes, that's the problem 🙂

[hallow thistle]

No, don't confuse me like that.

[analog obsidian]

try to not intervene if you dont know whats going on

[simple ore]

1. Unzip v3.2.9 2) run env\python ... line to install updated torch 3) run run-applio.bat

[tough fiber]

nah i cant do thing :D

[tough fiber]

tried that this happens all way

[hallow thistle]

I'm sorry? That's my fault? Anyways.

[simple ore]

show the training UI

open Advanced drop-down at the top

[tough fiber]

gpu no is 0 but

i tried before 1

still same nothing change

[simple ore]

0 is correct, you should not touch it

[tough fiber]

when i click train that error happens other ones works good

index etc is okay

RuntimeError: CUDA error: invalid device ordinal
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.

[hallow thistle]

Sometimes, I'm at lost when I didn't help with my own specified screenshot, because text isn't enough to imagine. Now that I have to install Applio again just in case.

[simple ore]

could you open cmd.exe

and type 'set'

[tough fiber]

a sec 2k screen itll be hard to read for u

if ur monitor is laptop

[simple ore]

okay, that's fine

[hallow thistle]

Yes, I can read that. I open it on browser, so it's fine.

[simple ore]

@tough fiber well, something to try: could you close applio terminal, open rvc\train\train.py in notepad, find the line gpus = sys.argv[6] and add a new line after that print(gpus), then re-run training and see what it outputs in the terminal log

[tough fiber]

sec running applio

[tough fiber]

literally says 0

[simple ore]

okay.. and then?

[tough fiber]

yea

works

i guess

[simple ore]

okau, so you messed up something before

[tough fiber]

idk what i did

didnt do anything different lol

[simple ore]

like trying to run training without closing applio

after you updated torch

[tough fiber]

probably u right

[simple ore]

okay, you can remove the print line

[tough fiber]

so should i do something any different or anymore should i click run applio and train models normally

[hallow thistle]

This looks fine now.

[simple ore]

no, you've updated torch, all good now

[tough fiber]

thanks you all u guys best

rtx 50 series is headache most of time

[hallow thistle]

No problem. RTX 50 GPUs are too new, so it can be hard to get that to work.

[tough fiber]

its training quite fast is it normal i didnt train before with 50 series

I think it's normal since the fans are speeding up.

[hallow thistle]

The thing is my laptop is too old and slow, even Applio won't show up.

[tough fiber]

i hope one day god blessing you with an rtx 5090 lol

or a6000

[hallow thistle]

Yes, this is good now. Would you let it continue or stop it there for another training?

[tough fiber]

ill let it continue

just open tensorboard

which graph was important

for overtraining

i just forget

is that less one

[hallow thistle]

I might gonna buy Dell XPS with NVIDIA GeForce RTX 4050. While RTX 4050 mobile is not really ideal for large scale AI training, I just wanted to play Minecraft and run RVC.

[tough fiber]

maan dont buy xx50 ones

that vrams -.-

4060 8gb is okey actually if u can do

even u can play minecraft with shader mod

for rvc vram is quite important if im not wrong 3050,4050 is idk not good with, if u run any game with rvc

[hallow thistle]

The one with NVIDIA GeForce RTX 4070 also exists, but this is more expensive than RTX 4050 one.

[tough fiber]

All I can say is that you shouldn't buy a gpu with less than 8GB vram

yea rtx 5050 is 8gb vram but 4050 (6gb) , 4050 (4 gb)

[sturdy patio]

anyone know why after training a model using a local host tool called rvc beta, it only gives me a .index file only? I don't see any .pth files

[hallow thistle]

This Applio looks new than what I installed before this one.

[eager wadi]

yeah i need to upgrade ive got a 1660 super which has 6gb

[tough fiber]

C:\Users\user\Desktop\uygulamalar\ApplioV3.2.9\logs\xxxx

xxx is ur project name

did u check here

[simple ore]

you're using a completely outdated tool, check assets/weights folder

[tough fiber]

if you can buy rtx 4060 itll be really good idk price difference than 4050 but its really big difference

[sturdy patio]

oh really

whats a good software to use?..

[hallow thistle]

"RVC beta" refers to some older mainline RVC versions. Not really recommended to use them. There's Applio, which is better than them.

[sturdy patio]

i've been suffering for the past 3 hours

Applio ah I see

thank you guys so much

yall are the best

[tough fiber]

i was using RVC lol i did really good models on them

before i tried applio but i didnt do lol

so i was using always RVC

now applio work so im good with that

[sturdy patio]

wait so

like 1-2 hours of my time

is gone then right

cuz i was using outdated softwares

[tough fiber]

did u check

[hallow thistle]

-rvc

[patent trellis]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[tough fiber]

weights folder

[sturdy patio]

yeah I don't see any .pth

HAHAHAHAHAHAHA

LOL

[tough fiber]

yea looks quite okey but 4070

is there any 4060

[sturdy patio]

[hallow thistle]

Dell XPS laptops are quite rare in my country. There are only ones with integrated Intel GPU, RTX 4050 and RTX 4070. Dell Inspiron laptops also exist, but they don't offer any dedicated GPU from what I looked for. Although some other brands like Alienware, Lenovo and MSI may have 4060.

[tough fiber]

well 4070 is really gonna be good , playing games with running rvc

idk ur budget but

if u not playing cyberpunk etc which is using too much graphic card and processor ull be okey

minecraft, pubg mobile etc It will work properly with rvc. However, when you play games such as Cyberpunk, the ping value in rvc will increase significantly.

I'm talking about situations where you run RVC together with the game*

[hallow thistle]

It is possible run RVC program like Applio with a game, especially a PC with good GPU, although I'm not sure how would you do convert audio on RVC while playing a game at the same time.

[tough fiber]

well

i never tried before :D

its impossible i guess haha :D

well it depents how much vram u give applio

im training model on applio and ill try to play brawlhalla rn

but i cant say gta 5, cyberpunk, elden ring etc runs good lol

[latent kettle]

Is it better than kokoro tts ?

[hallow thistle]

GTA 6

[tough fiber]

man

[viral mason]

What's that

[tough fiber]

i wrote 16 on applio i wish i try 12

using %95 lmao

[latent kettle]

TTS Text to speech

[viral mason]

I meant kokoro, ik what tts is lol

Never heard of it

[hallow thistle]

RVC can be slow sometimes when running with a graphics-demanding game, especially some PCs with entry-level or mid-tier type specs.

[latent kettle]

That's what i told you it's a tts text to speech program. It was listed on ai hub web page

[viral mason]

Ah ok

It was the way u said it that confused me

[latent kettle]

https://docs.aihub.gg/tts/tts-tools/#kokoro-tts

Oops sorry

[viral mason]

You're good 👍

[simple ore]

kokoro has no voice cloning

and it is somewhat bland

[latent kettle]

I don't want voice cloning. I want a good tts for English language.

@simple ore

[simple ore]

it is okay, it mis-speaks some words, but it is mostly fine

[latent kettle]

Any other tts without voice cloning. But should be the best and Natural

[simple ore]

orpheus

https://github.com/isaiahbjork/orpheus-tts-local

you'll need LM studio to run a server process for the LLM

[latent kettle]

It's it good for everything. Like the laughing, yelling etc ?

[hallow thistle]

NVIDIA GeForce RTX 5060 Ti (16 GB), this one has sufficent amount of VRAM where you can actually train a voice model there with less slowness. If Task Manager says "31.6 GB GPU memory", the another 15.6 GB comes from either your PC RAM or an integrated AMD Radeon GPU, while the actual dedicated GPU VRAM is still be "16 GB".

[latent kettle]

It includes that, but how it performs

[tough fiber]

my room getting hotter holly

[hallow thistle]

If your room is hot as a whole, maybe there's no air conditioner there?

[sturdy patio]

what's the best embedder model to use?

[hallow thistle]

What embedder model and which program are you looking for?

[sturdy patio]

applio!

rvc

[simple ore]

unless you're using a pretrain with the same embedder, dont try anything custom

contentvec is default

[latent kettle]

Is it actually good @simple ore

[sturdy patio]

ah i see

thank you

dude what the flip

applio is flipping bomb

[tough fiber]

while we talk

237 epoch

which graph was important

[safe grove]

Can someone tell me how do I read this lol its from my tensorboard

[mellow quail]

question, why my VC getting choppy on steam

[latent kettle]

https://docs.aihub.gg/rvc/resources/training/#usage-guide

[simple ore]

this is just a spectrogram, you can compare the original and generated one

[latent kettle]

But look at average loss instead of total

Average 50

[sullen stream]

when using changer in discord, if someone in the voice channel says something, then this sound is processed and pronounced by my changer. What could be the problem?

[safe grove]

Im training my model in kaggle using the save and run feature…. how can I see my saved models

[simple ore]

i dont know what you mean by save and run

you should be acess files using the file browser

or download the model from kaggle using training ui, there's an option at the bottom of the training tab

[safe grove]

like running the notebook in background

[whole pecan]

I’m randomly getting massive delays on my voice changer while playing gta 5. They’re random and super delayed. My task manager seems fine. What could be the issue?

[sturdy patio]

WAIT THIS IS GOING TO TAKE YEARS

I SET IT UP TO 1000 EPOCHS

IT TRAINS EACH EPOCH 1 BY 1?

[simple ore]

what's your GPU?

[sturdy patio]

amd rtx ryzen 5700x

its been like this for like 1 hour

[simple ore]

that's not a gpu

[sturdy patio]

but its using CPU for sum reason when i followed that tutorial

nvm i was tripping

lol

this one

[simple ore]

how did you start applio?

[sturdy patio]

i did the pre setup

and then i ran the bat

[simple ore]

which .bat

[sturdy patio]

i used the

run-applio.bat

[simple ore]

why

[sturdy patio]

;

is there something wrong I did

[simple ore]

did you read https://docs.applio.org/applio/getting-started/installation#amd-gpu-support-windows ?

[sturdy patio]

ye i did

[simple ore]

see 5.2

[sturdy patio]

[simple ore]

I assume you also missed half of the other things

[sturdy patio]

i did the

torch remove and install tho

[simple ore]

path?

[sturdy patio]

ye

added the path too

with cmd

[simple ore]

okay then close applio, run the proper .bat

[sturdy patio]

wait uh

i gotta change the

0 to 1 right

[simple ore]

depends on the GPU / igpu

try 0, see training tab advanced settings

if it shows the right gpu, you're good

[sturdy patio]

wait for some reason

i dont see the run-applio amd.bat

[simple ore]

4.1

[sturdy patio]

okok

[simple ore]

and, I assume, 4.2 or 4.3 too

[sturdy patio]

dude ngl

thank you

i cant imagine the amount of people like me you have to deal with everyday

danke

[safe grove]

Noobies this is what Im talking about earlier I trained using this

[simple ore]

you got UI?

[safe grove]

yuppiee

applio ui link and the tensor board

link

[simple ore]

training tab:

at the bottom

[safe grove]

okok

[sturdy patio]

wait its saying to now run run-install.bat first

but i dont see that

[safe grove]

Not done training tho haha

[simple ore]

where does it say you need to run it?

[sturdy patio]

in the run applio-amd

[simple ore]

full image

dont cut the top

[simple ore]

are you trying to run it as admin?

[sturdy patio]

nope not at all

[simple ore]

show the folder contents

[sturdy patio]

hey boss nvm

i found it

it worked

[simple ore]

do not run

[sturdy patio]

oh

wait now its saying its missing the torch wtf

😭

[simple ore]

lets start from the start

[sturdy patio]

okokok

[simple ore]

download https://huggingface.co/IAHispano/Applio/resolve/main/Compiled/Windows/ApplioV3.2.9.zip

unzip to C:\Applio

[sturdy patio]

ok done

and unzipped

[simple ore]

then follow starting from step 3.2

[sturdy patio]

okok

wait so i open cmd with admin

then i locate the path of appolo

[simple ore]

no

[sturdy patio]

and start there right

oh

[simple ore]

where does it say open it as admin?

[sturdy patio]

oh but i normally run cmd as admin to locate a path and run cmds there

okokok

ill try that

[simple ore]

Do not run anything as admin, unless it is specificaly required

[sturdy patio]

ah

okok let me try that

why are you so cute wtf thank you so much

[simple ore]

unless you want 540 viruses instealled on your PC

[sturdy patio]

LOL

my pc atleast has a couple rats and crypto miners on it 100% guaranteed

[latent kettle]

What is a QC in Ai hub

[sturdy patio]

uhm wtf

LMFAO

IMA ACTUALLY KMS

shit dumped everything here 😭

[viral mason]

interesting servers you're in

[sturdy patio]

OH WTF

WHY DID IT SS THE SERVERS

😭

[viral mason]

lol

[sturdy patio]

😭

[safe grove]

oh

[sturdy patio]

ok im reinstalling

the torch shawties

[simple ore]

you need to unzip properly

[sturdy patio]

okko i fixed it

Add C:\Program Files\AMD\ROCm\6.1\bin to your system's Path environment variable. this one uh

how do i this boss?

okay i did it nvm

Error loading "C:\ApplioV3.2.9\env\lib\site-packages\torch\lib\caffe2_nvrtc.dll" or one of its dependencies.

now it says this

[latent kettle]

@viral mason do you know what is QC ? What it do about how

Or @simple ore do you know about that

[simple ore]

they were people evaluating models for people seeling model master role

[sturdy patio]

daddy noobies help a brother out 😭

[simple ore]

follow the guide

step by fucking step

do not skip

do not ignore

[sturdy patio]

I DID DADDY

[simple ore]

what did you do

[sturdy patio]

i went into the

[simple ore]

show the screenshot

[sturdy patio]

SystemPropertiesAdvanced

environment variables

then added it

ok wait

wait do i need to restart my pc

ugh lemme just try that

fuck my life

[simple ore]

no

you can just start it in a new window

show me what you did in env variables

[velvet heron]

amd radeon 6900xt
windows 10
make an ai voice model
no tutorial (that's what im asking for)
no program (that's what im asking for)

[simple ore]

https://docs.applio.org/applio/getting-started/installation#amd-gpu-support-windows

applio download: https://huggingface.co/IAHispano/Applio/resolve/main/Compiled/Windows/ApplioV3.2.9.zip

[sturdy patio]

okok

so like

i added it right

the amd 6.2 thing

WAIT THE AMD THING RAN

IT RAN

IT FINALLY RAN

[simple ore]

... does the instruction say you need to install hip sdk 6.2?

[sturdy patio]

yeah

wait how do i check if my gpu is working?

[simple ore]

open training tab, expand advanced settings at the top

[sturdy patio]

aboutta throw my whole pc outta this shit

[simple ore]

what is your adrenaline driver version?

[sturdy patio]

its

25.6.1

[sturdy patio]

boutta suck my dick

and jump out my window

thanks

ok inshallah

praying this works

im aboutta go insane

ok i removed it

and now it has another error message

LMFAO

Error loading "C:\ApplioV3.2.9\env\lib\site-packages\torch\lib\cublas64_11.dll" or one of its dependencies.

[simple ore]

without quotes

[sturdy patio]

wdym

[simple ore]

just this

[sturdy patio]

it says this

[simple ore]

okay, so you did not run patch-zluda-6.2?

[sturdy patio]

currently it looks like this

i did run that

[simple ore]

did you add path to 6.2/bin ?

[sturdy patio]

uh

yeah

[simple ore]

not the Path

[sturdy patio]

nooo then i didnt

can you help me out

[simple ore]

either one of these

it literally says 'Path'

how do you miss it?

[sturdy patio]

oh

thats so haram wtf

wait so

C:\Program Files\AMD\ROCm\6.2\bin

i add this

as path right right?

ok i did it

i editted that and made a new path

[simple ore]

just add a new entry

[sturdy patio]

okok

ok i did it

wallah if this doesnt work

im not feeding my cat snacks tonight

she's getting fed today

💪

thanks noobies

you can touch my boobies

wtf i should be a rapper

[viral mason]

what

[sturdy patio]

dont worry

she gon eat well

[viral mason]

I'm scared

[sturdy patio]

hey bosses

uhm

this baddie is stuck on 1/925

for the psat 5 minutes

[bronze mirage]

Someone please?

[sturdy patio]

okay its done but uh

the model finished training in 1 minute

something has to be wrong

ok yeah i only hvae index wtf

[simple ore]

I'm sure you fucked up the preprocess

[sturdy patio]

oh my fucking god

[simple ore]

since you need to point it to a folder with the audio, and not at C:\

[sturdy patio]

wtf

ok

ima just

do it all again

C:\Users\Knock\OneDrive\Documents\sex

this one doesnt work then right

oh my fuck my life

[simple ore]

there is Seoul's experimental pretrain done with spin embedder

to use it, you need to download the embedder and save it into custom embedders folder

and you need a set of mute files for this embedder

you preprocess the set, extract features using custom spin, then you edit the filelist.txt in the model's folder and change the path for mute files to mute_spin

then you train the model as usual, for inference you'll need to use a custom spin embedder as well

[bronze mirage]

Thats alot of words i dont understand xd. Ill see alm this when im home from vacation, thx already

[simple ore]

but that time there will be applio release that does most of these things automatically

[sturdy patio]

is there a problem

[simple ore]

see the log

[sturdy patio]

compilation in progress

for so long

[simple ore]

all you had to do is read the instructions

[sturdy patio]

ye it said