✨│ai-help

[cloud finch]

I split it, same error

I genuinally have no idea what could be causing this

for this specific audio

[glacial pollen]

Then you're having your answer, wave's damaged

or corrupted

[cloud finch]

It plays fine in all programs

It opens

I can edit it

etc

[glacial pollen]

Sure, it plays but it doesn't matter, if there's any sort of corruption, shit can happen

Where did you get it from, what did you do with it, what program u used for exporting etc

[cloud finch]

Recorded it in audacity

[knotty moth]

try re-export the audio using audacity, or another application without including metadata

[glacial pollen]

^ reopen it and export again

[cloud finch]

How do i excport without metadata

[cloud finch]

I tried re exporting

to multiple different formwats

[glacial pollen]

makes no sense

[knotty moth]

uncheck "include markers and other metadata"

[glacial pollen]

funny enough, that never caused issues on my end
( typically export such in rx for archival purposes

[cloud finch]

Is there an audacity equivalent? I didn't record it in audition

or

do you want me to import it into audition

then export without metadata???

[knotty moth]

izotope RX: uncheck "preserve non-audio data"

[cloud finch]

I have no idea what izotope is

Im using Audacity

[knotty moth]

audacity should be able to open files without metadata, so the last thing is to make sure it is a valid stereo/mono file and not corrupted

[cloud finch]

No i mean like

I

recorded the file initially in audacity

so do you want me to export it normally from audacity, put it in audition, then export without metadata?

[glacial pollen]

... whatever you use, doesn't matter, just exclude the metadata

even audacity has it

" edit metadata "

[cloud finch]

how do you get that export screen?

Whenever I hit export it just sends me to the basic save file

[glacial pollen]

export audio / sound depending on your release

or ctrl+shift+e

( and keep it mono. )

??

[cloud finch]

wait nvm

yeah whenever i hit export it doesn't bring up that screen

[glacial pollen]

erhm, then update your audacity

[cloud finch]

i just get this

[cloud finch]

Alrighty

[glacial pollen]

no idea which version u use

[cloud finch]

oh there we go

alright its workinbg im updated now

ok so

Mono

then export without metadata?

I dont see anything metadata related

[glacial pollen]

in that case, just export

[cloud finch]

alr

[glacial pollen]

and retry in applio

[cloud finch]

didn't work

same fuckin error

[knotty moth]

use another working audio file and paste your recording there

[cloud finch]

Paste it, where?

[knotty moth]

ctrl+C your recording, close it then open the working audio, ctrl+A, del, ctrl+V

[glacial pollen]

^

[cloud finch]

Oh, you just want me to combine both into one file?

[glacial pollen]

no

you use file A as a dummy file / " container " for your target audio, File B

or treat it as a bucket, whatever

[cloud finch]

So, you want me to grab my recording, copy it, go to my working audio, and replace it with my recording?

[glacial pollen]

yea

[cloud finch]

I see

sigh

I hate vague errors...

[cloud finch]

It says the audio splits into 1 chunk

[knotty moth]

time to update your applio to 3.2.8 bugfix

[cloud finch]

I did

Thats what I did when i first got the issue

[knotty moth]

no, yours is 3.2.6

[cloud finch]

wtf???

oh

no

the

folder name is 3.2.6

but

I extracted the 3.2.8 bugfix into that folder

I just forgot to change the folder name

[glacial pollen]

Huh, thought I am the only one to write msgs like that

[knotty moth]

don't confuse yourself ah

https://tenor.com/view/spongebob-punch-boxing-boxing-glove-punching-myself-gif-21654165

[cloud finch]

Another thing, when I added something to the file

it split into 4 chunks

and then after 3 chunks converted it gave the CUDNN error

[knotty moth]

how about:

• download the fresh 3.2.8 bugfix
• update your nvidia driver

[cloud finch]

i already got the bug fix

and i already updated my driver a couple hours ago

Alright yeah im pretty sure the main issue here is that the audio splitter isn't cutting it into enough chunks

so when it tries to convert, the files too big

and it gives an error

[glacial pollen]

well then, that must mean the audio's specific enough so the slicer hates it ~ To put it simply

Does it not have any silence parts?

test it on few chunks ( cut em from ur audio ) 3, 5, 8 mins

See what clicks, manually cut it into chunks ur gpu handles and once you get all infers, concatenate 'em and call it a day

[simple ore]

once again, your file has no spot where it can be safely split at a silent spot

so it fails

it could be a file with a noticeable background noise or music playing

[knotty moth]

I bet he had recorded in a noisy environment or bad mic without denoising 💀

[hallow thistle]

W-Okada in #✨│ai-help?

[glass igloo]

Hi. I've written a script for a short story in the witcher universe, and I'm voicing it with voices from the game. I found a repository with almost all the voices for so-vits-svc, but either the models are not trained well enough, or the neuronics itself is outdated and not good enough, often the voice gets computerized or has pronunciation problems. Perhaps someone can tell me if there is a neuron that helps to remove the computerized voice, or in other words, a neuron that kind of makes the voice more natural, so that for example you can generate some replica, and then pass it through this neuron that would improve it, something like “the final touch”? Or maybe there are suitable models for something more advanced, with voices from the game and I just can't find them?

[low shard]

[low shard]

So vits SVC is old

Use RVC v2

[wispy lodge]

RVC guys discover SOLA to chunk audio and achieve infinite output in realtime at small cost, while the same RVC guys use silence to split audio in chunks that's never guaranteed and get OOM during infer. Now I've seen everything

Wasn't it supposed to split chunks simply by time? I think I've seen infer code in rvc mainline, and they pre-calculated estimated vram usage and just chunked the audio by specified amount of time

[low shard]

😭

[knotty moth]

[simple ore]

it is funny that the pipeline itself splits audio into small chunks

but no, it does not do audio splititng before pipeline

[wispy lodge]

Ah, no, it does preprocessing for training, but not for infer

Hecking lmao

Preprocessing interval (and padding) is set in config here based on VRAM like I said
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/blob/7ef19867780cf703841ebafb565a4e47d1ea86ff/configs/config.py#L167

But it's used only here and used for 1-click training, and not sure about padding
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/blob/7ef19867780cf703841ebafb565a4e47d1ea86ff/infer-web.py#L231
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/blob/7ef19867780cf703841ebafb565a4e47d1ea86ff/infer-web.py#L743

Anyway, whatever... It's a huge mess I don't even want to look into

[thorny onyx]

i tried to add model to RVC GUI but it didnt work for me can someone tell how to do it?

[simple ore]

I was not sure what that was about, applio still has (if fp16 chunk = 3.6, if fp32 chunk =3.0

oops.. other way around

per = 3.0 if config.is_half else 3.7

weird

but in config.py preprocess_target_version = "3.7" if precision == "fp16" else "3.0"

[cosmic epoch]

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

[low shard]

Train (make) RVC Models on cloud:

1. Prepare the Dataset
2. Setup RVC:
   Choose a cloud way to use RVC,

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
  • RVCDISCONNECTED (no ui)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio by Vidal (UI)
  • Applio by Shirou (UI, no guide as of right now)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • Applio (UI)
  • Mainline (UI, No guide as of right now)

3. 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.gg which ofc uses RVC

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

You can use either:

• Weights.gg: Easiest Possible Ever Automatic
• Ilaria RVC Zero: Fastest free on cloud
• Applio (ui)

However, u might wanna check ur pc gpu first in case it's good enough

as cloud is only for bad pc

[exotic scroll]

Me too, in weight gg.

I need more prenium training.

[eager cliff]

that was what i meant @low shard , its been like that for almost 2 hours , this just happened suddenly

[hallow thistle]

You have 2 training items, while I have like 30 training items.

[low shard]

weird, have you tried refreshing?

[exotic scroll]

Just, give me more prenium trainings.

[eager cliff]

yeah

[hallow thistle]

Weights works like normal, even on my slowest laptop. Have you ever tried to refresh this page?

[eager cliff]

i did but nothing new

[low shard]

works all fine for me, did you check if this issue persists on other devices? might be ur WiFi

or where are u from ?

[hallow thistle]

With no premium mode, you can use a premium item to train a voice model.

[exotic scroll]

I only have last 2.

[eager cliff]

id rather not say since i do want private info to stay private , but im sure this just happened today , if the issue was in my wifi , any other site wouldv been also like that but i will check again on other devices

[hallow thistle]

I'm not sure why would you wanna train 69 models there at once. I only train a voice model on Weights as a draft voice model.

[low shard]

id rather not say since i do want private info to stay private
Understood, I meant the staty like "USA" or "Italy" rather than anything more private ofc
I just asked since it seems it was blocked in South Korea and would be weird if it was blocked in other states

[exotic scroll]

https://www.weights.gg/invite?inviteCode=p6bd9

Here's a invite code to earn me more prenium training.

[hallow thistle]

Send your Weights link to #1159290752195633273, not here.

I already have my own Weights account, so unfortunate. I started from the start.

[eager cliff]

alright so it does load on a phone , even recognized my account
but weird thing it didnt on my pc , ig id have to switch to the phone
thanks for help

[exotic scroll]

Oh, i having my phone.

[low shard]

yw, @oak plank u might wanna check on that since weights.gg doesn't seem to load on his PC but does on mobile

[oak plank]

That's weird

It should login, but I will inform Bea about that

[low shard]

maybe he's using mobile data ?

not sure

[knotty moth]

vpn?

[oak plank]

Try clearing cache and loading it again

[low shard]

@eager cliff if u could tell this

[low shard]

idk u should ask charles

[oak plank]

Most of the times it works

[eager cliff]

nope

[eager cliff]

lemme try

[oak plank]

Okayyy

[knotty moth]

either ISP or browser issue

[eager cliff]

prod yeah

[lucid pebble]

RVC

[carmine coral]

what program should I use to use RVC

[low shard]

elaborate more on ur pc gpu and what u need

[carmine coral]

a real time one and I got an RTX3060

[low shard]

for realtime u should use Wokada, especially the fork

lemme explain u in #🔍│help-w-okada

[eager cliff]

alright iv got it fixed , google chrome really sucks sometimes

[glacial pollen]

Use opera gx, easy solution. Chrome is trash

[bleak pasture]

I trained in weight model and the truth is that it was bad, on top of that it only gives the PTH, it doesn't add the index. Training in weight is horrible.

[eager cliff]

thats what i acually did

[glacial pollen]

Oh yea. Gjob, you're a W

[eager cliff]

🙏

[bleak pasture]

So much they said that weight is good, it is literally horrible, updates are needed and in general everything

[glacial pollen]

In what sense?

( asking as I do not use it so, don't know what's wrong / bad etc with it )

@bleak pasture

Might pass the message to some proper people

[bleak pasture]

Everything is wrong, everything does not generate good covers, the voice models are trained poorly, the images are generated poorly, everything is wrong, literally nothing is done well on the page

[glacial pollen]

How much of user's input is needed on weights

for generating covers / using rvc / applio

Cause like, have you excluded possibility you've done something wrong / user's error

[night mortar]

Is there an RVC TTS software i can download that reads documents in my browser? im doing online classes and the select to read tts is absolutely horrible, and i was wondering if there is a rvc line reader i can use for my browser

[low shard]

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 & japanese, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/

Freemium 11labs: An 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

With RVC Models:

RVC is natively for Speech To Speech, but forks such as ilaria rvc mainline & 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
• While Ilaria RVC Mainline here (no guide as of right now)

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

• Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide

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

• if you don't wanna use edge tts, you could try another tts ai from our tts index and use the output as an input in rvc

[night mortar]

does this allow me to select a line and have it read it or does that not exist yet and i have to manually paste it

i want to be able to select text and have it read with rvc directly from browser if possible

[brittle wing]

i need the link for the voice changer

i cant find it

[night mortar]

i think you can find them in #1159513888199540817

whay do you plan to use it for?

i see you are playing roblox but i wouldnt use rvc on there because i got warned when i did that

like the roblox vc auto moderation warned me lmao

[night mortar]

@acoustic scarab is this using my intergrated graphics instead of my gpu

if so how do i change it

oh and it just errored

[acoustic scarab]

can't say for sure with given amount of info

https://tenor.com/view/cat-kitty-tired-sleepy-relaxed-gif-7257672

[night mortar]

bruh

how do i get it to work

[vapid mantle]

Which type of pre-training is the best?

[night mortar]

i just downloaded and ran

[low shard]

There isn't a best one, depends on ur dataset lang and size

[night mortar]

oh uhh i tried one of the ones you sent and i keep getting this error

i uploaded a model

and selected it

and tested it

and it uhh

didint work

[low shard]

Could you exactly link me what u are using

[vapid mantle]

What should be the target_sample_rate? 40k?

[low shard]

the cut off seems to be around 16k, which multiplied by 2, is 32k

[vapid mantle]

I understand it will be 32k

?

[low shard]

ye

[vapid mantle]

thank you

[low shard]

yw

[bleak pasture]

The best one is the default one which is original, it is better to leave it as default which is original.

[vapid mantle]

thx

[bleak pasture]

you're welcome bro

😊👍

[rocky jacinth]

How can i upgrade my Applio? I am using ApplioV3.1.1

[simple ore]

Download the latest compiled release, unzip it into a folder of your choice, like C:\Applio-3.2.8

move your models and audios over from the old folder to new

delete old folder

[rocky jacinth]

Thank you

[hallow thistle]

Bro's still not over about Weights even after a month ago.

I use Weights to train some "draft" voice models to make sure what they could sound like if trained locally. It ain't a big deal.

[rocky jacinth]

I want to use Arabic Vocals to train. Which pre-trained models can do it perfectly?

[brittle wing]

#1235952130855010365

[rocky jacinth]

What does OG pretrain means?

[simple ore]

original pretrain that came with RVC

[knotty moth]

use the default pretrain or try make an arabic pretrain by yourself. using 20+ hours dataset of wide range of speakers and renting some A100s are recommended.

[rocky jacinth]

Thank you

[rocky jacinth]

Where can i find it?

Is it automatically uses OG one when i click start train?

[simple ore]

generally yes

if you see f0G48k.pth name, that's og pretrain

[cloud finch]

The other files I had, had background noise though

and they worked

oh

I did a denoiser on my audio and it worked

[glacial pollen]

yup, so we were on the right track then

splitter had no places it deemed as split-able

[knotty moth]

yeah you should definitely

[jaunty fable]

what are the models made of? vstt stt gtts

[low shard]

Huh? Those are RVC (Retrieval-based-Voice-Conversion) models, which are STS (Speech To Speech)

They are made of a .pth file which is the voice, and an index which is the accent

[buoyant pagoda]

what is the best local rvc

[hybrid fractal]

Hi guys. how much slower training with FP32 is compared to FP16? Does training with FP32 provide enough advantages?

[simple ore]

VRAM is the main concern, speed is not 2x slower, but depends on GPU

if you're finetuning you may get away with just FP16

[exotic scroll]

Although, i already made my rvc models. But i had 2 prenium training.

[simple ore]

may even not explode during training

[dim bough]

#ask Can the voice model here be used for the F5-TTS?

[low shard]

There isn't a best one, all have the same quality

It mostly changes the UI, even if in the latest applio or codename fork there are newer features

What's ur PC GPU

[low shard]

No, those are 2 different type of models, RVC is STS while F5 is TTS

The only way to use it like that is to make a random tts audio with any voice, then convert it to the RVC voice

But you can already just give 30 secs of sample to F5 to make a similar voice, it's just not as good quality as RVC

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 & japanese, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/

Freemium 11labs: An 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

With RVC Models:

RVC is natively for Speech To Speech, but forks such as ilaria rvc mainline & 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
• While Ilaria RVC Mainline here (no guide as of right now)

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

• Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide

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

• if you don't wanna use edge tts, you could try another tts ai from our tts index and use the output as an input in rvc

[dim bough]

In F5, it is very difficult to find an Indonesian model. Applio, on the other hand, has many. Unfortunately, Applio does not support voice emotion and Local LLM. Do you have any suggestions?

[simple ore]

Applio uses Microsoft's voice service (Edge screen reader0

[dim bough]

Is there any way to run Applio with a local LLM?

[simple ore]

no

[velvet gale]

hello, does anyone know why my conversions sound grainy? the original was clear, the model was some of the most popular ones( I tried multiple)

[low shard]

Applico Just uses Edge TTS API to make a TTS Audio, then converts it with RVC

[hybrid fractal]

Hi, my mel spectrogram similarity is stuck at around 40-50%. Could this mean the training is going wrong?

[low shard]

Could the edge tts thing be specified in the info notes btw? People get confused and think RVC is TTS too

[velvet gale]

here's a sample

[low shard]

The quality depends on the voice model

[low shard]

W

I didn't see it in his SS so I thought it wasn't there

[simple ore]

it may be a recent addition

[low shard]

I haven't used applio since September

[simple ore]

after so many questions

[velvet gale]

ah. I have the default pretrain model. and I tried a few models on weights.gg, only the karina one works fine, the rest produces grain like the above (jisoo, jennie, ... all had >20k downloads)

I didn't know if it was the model or some parameter tuning I'm not aware of (big noob)

[low shard]

The default Pretrain model isn't made to be used, it's made only as a base for training your own models

That's how Pretrains work

[low shard]

The quality depends heavily on the model

However applio also has a Clean Audio option in the advanced settings

Maybe that can help

[velvet gale]

when I say default pretrained, it means I didn't do anything to the default pretrained_v2 folder. I just add models to the weights and used it

[low shard]

Yeah dw even if u change Pretrain, it won't matter unless you train your own

[velvet gale]

I'll give applio a try I guess

Thanks @low shard

[low shard]

What are you using rn?

Mainline?

[velvet gale]

the default repo https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI

[low shard]

Applio Is a fork of it, meaning more updated with some things
You aren't forced to switch since it's same quality type except some new features

However I remind you it's better to just search for a better model

[low shard]

Ye we call that Mainline

[hybrid fractal]

Hello, What are some easy ways to spot overfitting by looking at the charts in TensorBoard?

[velvet gale]

yeah I'll try to find other models too. I tried all the most popular models and only karina one worked for me I thought it was something else other than model. Thanks a lot!

[vapid mantle]

What should be the target_sample_rate: 32K?

[low shard]

Yup

[vapid mantle]

thx

[low shard]

Yw

[vapid mantle]

So should the slicer gui settings be like this?

?

[dim bough]

That's right. Both Applio and F5 can run TTS.

Are the models in this group only for Applio?

[low shard]

The slicer GUI?

[vapid mantle]

yup

[low shard]

I don't remember that, do you mean this part https://docs.ai-hub.wtf/rvc/resources/dataset-isolation/#step-4-manual-audio-slicing-optional of the docs?

[vapid mantle]

So it cuts the sound every 10-11 seconds

[low shard]

I'm not sure what program you're running that's why I'm asking what tutorial (link) are u using

[vapid mantle]

I don't know either. I have been using this program for a year. It divides the sound in just 10-11 seconds and I make the model this way. I saw it on a YouTube video

[hybrid fractal]

what causes fm to go up or down ?

[low shard]

@crude flame doesn't RVC automatically slice the audio

[raven notch]

Hello everyone, I have the opportunity to familiarize myself with RVC in Google colab. https://colab.research.google.com/github/w-okada/voice-changer/blob/v.2/w_okada's_Voice_Changer_version_2_x.ipynb.
Everything starts fine and even the standard voices work well, but if I add custom voices, then RVC stops working and throws this error.

RuntimeError: ('Exeption in RVCInferencerv2F0', RuntimeError('CUDNN_BACKEND_TENSOR_DESCRIPTOR cudnnFinalize failed cudnn_status: CUDNN_STATUS_NOT_INITIALIZED'))

• vcclient_dev/server/validation_error_logging_route.py - 33
  2024-12-28 13:49:05,899 - vcclient - validation_error_log - ERROR - Exception: ('Exeption in RVCInferencerv2F0', RuntimeError('CUDNN_BACKEND_TENSOR_DESCRIPTOR cudnnFinalize failed cudnn_status: CUDNN_STATUS_NOT_INITIALIZED')) - vcclient_dev/server/validation_error_logging_route.py - 32
  2024-12-28 13:49:05,901 - vcclient - validation_error_log - ERROR - Exception tb: Traceback (most recent call last):
  File "vcclient_dev/voice_changer/voice_change_manager/inferencer/rvc_inverencer_v2_f0.py", line 92, in infer
  File "vcclient_dev/voice_changer/voice_change_manager/models/rvc/synthesizers/rvcv2_f0.py", line 118, in infer
  File "torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
  return self._call_impl(*args, **kwargs)
  File "torch/nn/modules/module.py", line 1562, in _call_impl
  return forward_call(*args, **kwargs)
  File "vcclient_dev/voice_changer/voice_change_manager/models/rvc/components/TextEncoder768.py", line 49, in forward
  File "torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
  return self._call_impl(*args, **kwargs)
  File "torch/nn/modules/module.py", line 1562, in _call_impl
  return forward_call(*args, **kwargs)
  File "vcclient_dev/voice_changer/voice_change_manager/models/rvc/attentions.py", line 52, in forward
  File "torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
  return self._ca

There can be no problems with the voice file, because I checked locally and everything worked fine.

[simple ore]

use new slicing method

[fierce vector]

i cant use voice model

[vapid mantle]

It's too late now, I've already started making the model

[fierce vector]

how can i use model

[vapid mantle]

You need the RVC GUI

[fierce vector]

what's that

[simple ore]

10sec slices are waaaaaaaay to long

[vapid mantle]

AI helps you use voice models

[low shard]

Wrong channel, use #🔍│help-w-okada

RVC is NOT wokada

[low shard]

Use models how and what's ur PC gpu

[vapid mantle]

Well I have been using the same method for 1 year.

[low shard]

Maybe don't say that else they think it's the GitHub rvc GUI from t1ger that is outdated

[vapid mantle]

Hmm

[low shard]

You're not using that.. right

[hallow thistle]

For W-Okada, go to #🔍│help-w-okada. This #✨│ai-help channel here is about RVC.

[simple ore]

and it is wrong. 10 sec slices require way too much audio to achieve the same as, say 1/3 set with 3s slices

[hallow thistle]

-gui

[azure marsh]

https://cdn.discordapp.com/attachments/1122285248844144733/1203460490475343953/caption.gif?ex=65d12cec&is=65beb7ec&hm=bd2fb8d010006dd7c6e3c1c67d3ae846fd1478e1a3124c544c31b43086fe54aa&

[vapid mantle]

Isn't it too short?

[fierce vector]

i want to send image of whats the problem is, but le me text it, it says u can't upload file of that type

[low shard]

What RVC are u using and what tutorial

And what's ur PC gpu

U need to give info

[simple ore]

each step RVC only uses 0.36s segment out of each sample, so with 10s it uses ~1/30th of the set

[hallow thistle]

https://cdn.discordapp.com/attachments/1159290161683767298/1321108259251425452/image.png

[fierce vector]

how to check

[low shard]

You can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab -> GPU

[hallow thistle]

See my message above. If you're using Windows, press Windows key + R, type taskmgr and click run.

[hybrid fractal]

Is it okay to use japanese hubert while using a pre-trained model?

[fierce vector]

Intel(R) UHD Graphics 620

[analog obsidian]

Yes but remember to inference only using japanese hubert after training

[low shard]

That's integrated graphics, very bad, u can't use models locally but can use cloud

Do u want to use them in realtime for calls or on pre-recorded audios like ai covera

[fierce vector]

yea, real time

[low shard]

Wrong channel then

Rvc isn't for realtime, it's for making models and pre-recorded audios

Go to #🔍│help-w-okada

[fierce vector]

ok

thanks

[hybrid fractal]

thank you!

[low shard]

For japanese models I guess

[crude flame]

Yes

[little ore]

how do i train ai voice model

[latent kettle]

Which GPU Do you have

[shadow prism]

i never understood, why do some people train models and some dont?

[daring moth]

hi can someone help how to put voice model in rvc

[azure marsh]

Hey, ! Sam._! Please use the command !howtoask to increase your chance of getting help by structuring your question in a way others can understand better. Also make sure you're asking in the right help channel:

• General RVC help: #✨│ai-help
• W-Okada / Realtime RVC: #🔍│help-w-okada
• AI image related: #🔍│help-ai-art

[latent kettle]

Some people train models because they need them and they are not available, Some people actually want learn or do experiments because of their creosote to learn something new, Some people also use pre trained models (the models already trained by other users) because of different reasons (they may not have enough time or resources or it can be very complicated for them)

[latent kettle]

Are you using RVC locally or on cloud. And which rvc exactly?

[knotty moth]

i never understood, why do some people bake some cakes and some eat them?

[latent kettle]

Bro... you roasted him 🤣

[hallow thistle]

Roasted like a corn.

[shadow prism]

im not a tech nerd cmon...

[trim hearth]

:blizzard: 👉 :buddha:

:blizzard: 👉 :buddha: à

à:blizzard: 👉 :buddha:

:blizzard: 👉 :buddha: à

:blizzard: 👉 :buddha: à

:blizzard: 👉 :buddha: à

:blizzard: 👉 :buddha: à

[latent kettle]

I answered your question btw

[shadow prism]

i saw it, thanks 😊

[latent kettle]

Don't feel bad or sad for above massages by @knotty moth it's all for fun

[zinc crow]

Ciao @low shard e @proven hill. Vi ricontatto poichè ho provato a fare come mi avete detto (registrare la mia voce, cantando), ma la mia tonalità è molto diversa da quella di Lazza (o di altri artisti), perciò non gli somiglia. Mi esprimo meglio: ho provato a fare "100 MESSAGGI" di Lazza con le mie parole (esempio: il 3° audio, chiamato 100 MASSACRI è identico a Lazza. Link: #1287422682446565578 message), ma purtroppo come detto in precedenza la mia tonalità è molto diversa da quella degli artisti perciò se faccio per esempio Lazza, si gli assomiglia perchè la voce ricorda la sua, ma purtroppo non è proprio uguale, a differenza dell'audio caricato sopra tramite Link.

Avete altre idee? Ad esempio esiste un TTS Singer che posso mettere un audio di riferimento o qualcosa simile? Tra l'altro ho notato che con un TTS semplice viene meglio rispetto alla mia voce (anche se non a tempo, ma questo non è un problema).

Attendo Vostre,
Lorenzo

Scusate il ping, buona serata.

[proven hill]

no, non esiste purtroppo

[low shard]

purtroppoo non credo che esiste

[zinc crow]

Ciao, vi ringrazio per la risposta. Avete altre idee su come migliorare la voce e farla il più simile possibile

Perchè purtroppo ho aspettato (sperando in una possibile risposta), ma purtroppo @civic meteor non risponde.

[low shard]

forse, potresti provare https://www.weights.gg/it/create/song , qua è gratis (se non ricordo male) e puoi mettere una traccia di riferimento, ma non l'ho mai provato onestamente

potresti provare cosi, ma ho un grosso dubbio che non sara simile, però non so cosa altro dirti

[zinc crow]

Grazie. Perchè mi sembra alquanto strano che il link di PlasmaTi viene identico.

[low shard]

dimmi come va con weights.gg, non ho mai provato onestamente ed ho dei dubbi vistoo che le voci italiane sono meno conosciute

[zinc crow]

Massimo 190 caratteri mi dice... Non riesco a fare nemmeno un verso.

[low shard]

ah, mi dispiace

[odd bolt]

hay uno de amador rivas?

el de la que se avecina

[dense drift]

How should i put my model there?

[pastel oak]

https://docs.ai-hub.wtf/rvc/cloud/applio-colab/#1-upload-voice-model

Step Inference: 1. Upload voice model

[dense drift]

Thankss

And does it remove the vocals itself?

[vapid mantle]

@low shard Hello again. Will there be 40K here?

or 32K?

[simple ore]

still 32k

[vapid mantle]

thx

[low shard]

I was busy my bad

[brittle wing]

-colab

[azure marsh]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

More commands: -audio, -hf, -local, -kaggle, -realtime, -rvc, /help

[surreal garden]

I've done voice to voice before but how would you make a parody where you would need to change the original song lyrics

[patent pasture]

udio ai maybe

[dim jewel]

What is more important, Fm or Mel?

[simple ore]

mel > fm > kl... but if fm is all fucky, you wont get good results

mel as in some latest Applio releases

original one in RVC is not good

[low shard]

You need to sing

And convert the voice

[sand sparrow]

does anyone have free time to walk me through how to download the voice changer top to bottom

[low shard]

Realtime voice changer for calls? That's Wokada, not rvc, go to #🔍│help-w-okada

[slate harbor]

the ai cover sounds realy fake and robotic

is there anyway to fix that

[tame mica]

mixing

eq, reverb, compressions

[glacial pollen]

• train better or get a better model

[brittle wing]

how do i import file into okada for ai to cover

[glacial pollen]

w-okada is not for covers 🤦‍♂️

rvc / applio = covers
w-okada = realtime voice changing

#✨│ai-help
#🔍│help-w-okada
Both have their respective channels

[crude flame]

ai hubbers am i right?

[glacial pollen]

oh?

[brittle wing]

im new to this stuff mb😭

[glacial pollen]

's good, dw

[slate harbor]

ok ima train my model

[brittle wing]

so rvc is how i make song covers right?

[slate harbor]

yea

[crude flame]

https://docs.ai-hub.wtf/essentials/how-to-make-ai-cover/ read this

[slate harbor]

can u send the guide to the train thingy

[glacial pollen]

@slate harbor
https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/

[marsh vector]

guys, what's up with Ilaria RVC on huggingface? everytime i try to put a huggingface link for a model it tells me that it's not an huggingface link lol

Ok, it's a problem with her space, because the links are working on other spaces

[proven hill]

there is no problem.

[marsh vector]

it's okay now, but before it was glitchin

[dim jewel]

I had some problems. Turnes out I needed to clean audio of sample and of data. No back vocals, no echo, check your audios for any anomalies that doesn't sound right.

[cerulean flower]

-colab

[azure marsh]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

More commands: -audio, -hf, -local, -kaggle, -realtime, -rvc, /help

[robust bone]

if you have like a pretty good computer, do u think it would be better to do rvc stuff locally

[proven hill]

depends

[robust bone]

r u the person that made llaria rvc

[proven hill]

actually yes lmao

[robust bone]

yooo

its so good

i used it alot

[proven hill]

thank you 😭

[robust bone]

basically i got a pc with a 4070 and a ryzen 7 9700x recently, i dont know much about rvc i kinda just followed the guide for llaria rvc on i think it was huggingface and use pretrained models. I just wanted to know if u could make like, better quality or better sounding covers by running locally, and also the gpu limit on huggingface got annoying. I also wanted to train models that sound like really good but idk how.

[proven hill]

1. quality is the same local or cloud
2. the limit is lame but thats the max i can give to yall without nothing in return
3. for training models you should ask a staffer

[thin stump]

I dm'd you by the way though it's your preference on where you want to train. I can't help with that

[potent helm]

does anyone know what kits uses for the settings to train their models, tested a dataset on it and it came out a lot better than applio's model

[worthy valve]

Hey everyone!
I'm currently looking for an internship in AI/ML and I'm eager to gain hands-on experience. If anyone knows of any opportunities or has leads on guaranteed internships, please let me know. I'm passionate about learning and contributing to AI projects. Any help or guidance would be greatly appreciated!

Feel free to DM me. Thanks in advance!

[lone nest]

Let's say I want to resume training a model on RVC disconnected that has been trained for 100 epoch, I input total epochs as 200, will it be trained for 100 more epochs to a total of 200, or 200 more epochs to a total of 300?

[north scarab]

What does Index search FAILED or disabled failed mean

[simple ore]

it resumes from saved 100 to your selected max epoch

[timber coral]

does not work, neither the application nor discord sees the sound

[low shard]

gonna talk in #🔍│help-w-okada

[pastel oak]

Nothing, just that you left index at 0. Not an error

[chilly creek]

how can i download vocie models from wheigts

[low shard]

To Download a Model from Weights.gg:

1. Login
2. Click the 3 dots at the right of the image of the model
3. Click download
4. Download Anyways
5. Unzip the zip, and you might wanna rename the pth and index since all models on weights are renamed as 'model'

[low shard]

@raw stirrup It's better to ask here rather than in DMS

about your loss fluctuating, maybe wait a bit, it's just the 40 epoch

[raw stirrup]

ok thx

[low shard]

also check the tensorboard

[raw stirrup]

im running the regular rvc version

does that support tensorboard

[low shard]

mainline ?

[raw stirrup]

yeah the og one not mangio or applio

[low shard]

Maybe try checking https://docs.ai-hub.wtf/rvc/resources/training/#tensorboard

[raw stirrup]

its the latest nvidia build but i dont see a tensor board batch

[low shard]

I haven't used mainline locally

[raw stirrup]

ok thanks bro

[low shard]

yw

[raw stirrup]

chat gpt seems to think itsfine so far but its just a bot lol Conclusion
From the logs provided, your model is progressing well. The loss trends are typical of a generator-discriminator setup in mid-training. Assuming no major artifacts in the generated audio, you're in a great position to continue training and refine your model further. Keep monitoring the losses and evaluating samples for the best results.

😄

[pastel oak]

u need to download th tensorboard bat from the link nick sent, put it in the rvc folder, run it in there

[raw stirrup]

cant seem to open link 4sum reasom

[pastel oak]

this is the tensor download link:
https://docs.ai-hub.wtf/rvc/resources/tensorboardfiles/tensorvenv.bat

[raw stirrup]

thanks bro

[brittle wing]

can someone teach me how to use a voice channger

[raw stirrup]

this is the loss graph is it doing well?

[lost pike]

hey guys im a newbie and i was trying to train my voicemodel but i get no-feature-todo instead of all features done in the step 2b

[low shard]

Ye, going down is good

[low shard]

Are all ur files in the dataset.wav?

Try to report them aw wav in audacity

[raw stirrup]

thanks brother 🙂

[low shard]

Rvc is not for realtime voice changer for calls, ask in #🔍│help-w-okada

[lost pike]

ohhh, ty lol it was in mp3 format

[lost pike]

@low shard i still get the same thing :/

[low shard]

You should never train on mp3

It's lossy

Be sure to always get wav sources for ur dataset

[low shard]

Show me how ur dataset is formed

Like a ss

[lost pike]

i cant sent it :(

send*

[low shard]

U are level 1

Level 1 users can send images

[lost pike]

oh

[low shard]

Try restarting discord

[lost pike]

it workds now

[low shard]

It didn't work

You put an invalid path 😭

Double check the path

[crystal gull]

Is RVC Nvidia 1006 still the go-to for realtime voice conversion and training?

[pastel oak]

realtime its https://rentry.co/forkvoicechangerguide
training you can use it, people been using applio lately or forks, check at https://docs.ai-hub.wtf

[crystal gull]

Thanks b 🙂

[crystal fractal]

can someone help me humanize a really short text

i just cant get something in my head

dont even need an AI

[lost pike]

guys i have a rtx 4060 ti but dont know what to set the epoches

[crystal fractal]

idk somewhere ig

[low shard]

https://docs.ai-hub.wtf/rvc/resources/training/#epochs--overtraining

[lost pike]

should i stop the training @low shard

[low shard]

I'm not sure about that error, it's better u ask someone else

[hybrid fractal]

u have to install python

and then you need to add pip to the PATH environment variable.

[surreal delta]

where is big beta 5e in uvr

i have the beta

[carmine hearth]

Recently, some people have reported that when inferring with applio, the beat of a song becomes increasingly out of sync.
Does anyone have any idea what might be causing this?
Sorry if it's hard to understand, I'm using machine translation.

[raw stirrup]

applio is a experimental version of rvc, i recommend rvc mainline or mangio

ive tested all 3 versions mainline just works much better and is more accurate in training models and less bugs

i been doing this stuff before rvc and so-vits began

im from the tacotron 2 deep faking

[carmine hearth]

Thank you for your advice.

[crude flame]

applio is not an experimental version of rvc

at all

[raw stirrup]

its taken from rvc not a new thing

is what i meant

[crude flame]

applio is a fork

not an "experimental version"

and mangio is very outdated

[raw stirrup]

yeah its forked from rvc mainline tho, and is worse in my opinion

[crude flame]

applio is faster tho

same quality

[raw stirrup]

na ive trained 3 models on both versions rvc mainline is much more accurate and infers just as fast

[crude flame]

💀

that is so wrong

applio and mainline are the same

one is faster

[raw stirrup]

? ive used all 3 versions and been doing this since 2020 with extensive testing

[crude flame]

and soon applio is going to be better if refineGAN gets a pretrain

[raw stirrup]

when tacotron2 and talknet weas around

ill test future updates but right now mainline for me is much more stable and accurate

[crude flame]

when was the last time you used applio?

[lavish lintel]

Congratulations Razer by Weights!

Your Torterra is now level 34!

[crude flame]

https://tenor.com/view/allmight-bnha-mha-meme-fist-up-gif-13588517052470008833

[raw stirrup]

2 weeks ago

i even tried the new beta thing they made

was worse in comparison to mainline

[crude flame]

mrf and refine dont have pretrains so ofc they suck

[raw stirrup]

think they either messed with the configs or some other code

[crude flame]

they made it better

thats all

[raw stirrup]

if it was better the models would be more accurate than rvc mainline which isnt the case

i purpose u to try it yourself

[crude flame]

better as in faster

[raw stirrup]

propose*

[crude flame]

i do

ive been using applio since pretty much the start

[raw stirrup]

fatser isnt better if the results dont match up

accuracy is always the goal

[crude flame]

the quality is the same

[raw stirrup]

nope

far from it

[crude flame]

user error then

[raw stirrup]

same datasets on applio and rvc and rvc mainline comes out more precise

every time

ive trained big modles 2pac andrew tate solid snake mainline just is on another level

wish i could agree with u

[crude flame]

if it sounds worse thats just user error

[raw stirrup]

it cant be user error when the dataset is exactly the same and same settings

[crude flame]

well there is no difference other then speed

so

only option

[raw stirrup]

ill provide u with a audio sample from both models 1 trained on applio and 1 trained on rvc mainline and prove ur wrong

[haughty lark]

Can anyone help me? My VB audio doesn't come out with my voice so I can use AI's voice on discord

[azure marsh]

Hey, kakauinworld! Please use the command !howtoask to increase your chance of getting help by structuring your question in a way others can understand better. Also make sure you're asking in the right help channel:

• General RVC help: #✨│ai-help
• W-Okada / Realtime RVC: #🔍│help-w-okada
• AI image related: #🔍│help-ai-art

[raw stirrup]

did u set the mic outputs etc in youre gui

[haughty lark]

Yes, I even selected the microphones in Windows, but the voice doesn't come out

[raw stirrup]

probably best to ask here https://discord.com/channels/1159260121998827560/1159290161683767298

[knotty moth]

did u train the models using applio and try to infer in applio and mainline?

[raw stirrup]

no i trained my dataset on both applio and mainline both at 200 epochs and same settings mainline came out more accurate

not just my 2pac model biggie andrew tate etc mainline wins for me

im using latest nvidia mainline build

[crude flame]

did you infer the applio model in applio

[raw stirrup]

yea sending now

[knotty moth]

and did you use default pretrain or another?

[raw stirrup]

default but ive tested multiple like custom on both mainline and applio mainline always comes out on top

[crude flame]

infer using the same audio

[raw stirrup]

okier dokie

these are both 200 epoch

[crude flame]

same batch size?

[raw stirrup]

i always train on batch size 4

always same settings

my rvc mainline model is damn near hard to distinguish from the real 2pac shakur in my opinion very lifelike

[crude flame]

huh well you did something to applio where it does that

not normal

[raw stirrup]

lol ive done nothing to it like i said try it for yourself train all on same settings same dataset and mainline latest build will come out on top

anyone saying otherwise is telling lies

whether its bias or something i dont know

but hey whatever flaots youre boat for me mainline is the best thing to train on

🙂

[hallow thistle]

https://cdn.discordapp.com/emojis/1016021422809817160.webp?size=48

[hallow thistle]

The "AI's voice", you might mean by W-Okada. For virtual line audio, I'd suggest for Virtual Audio Cable (VAC) Lite instead, as this one works better than VB-Cable.

https://rentry.co/ForkVoiceChangerGuide#virtual-audio-cable-mandatory

The only thing W-Okada related to RVC is W-Okada uses RVC voice model to inference, the code and GUI themselves aren't related. It's best to go to #🔍│help-w-okada for more information.

[haughty lark]

It's when the AI ​​audio takes a long time to come out, how to solve it?

[hallow thistle]

It's best to go to #🔍│help-w-okada for more information. This channel here #✨│ai-help is about RVC and certain RVC fork GUIs.

[haughty lark]

okayy

[simple ore]

you likely ran process effects that does normalization of the full source file

while RVC normalizes each individual chunk without asking

that's why there's a difference

but if you take sliced audios from RVC and copy them to Applio's model/sliced audios (and 16k), the result should be the same

again there's a possibility that RVC does yet another normalization for inferred audio

but that's just how it is

[raw stirrup]

@simple ore I ran no effects i trained on both versions with the same dataset at 40k pretrains v2 lol, no effects no tweaks both equal settings same fresh install default settings and rvc comes out more accurate

ive been testing these methods for months on both applio mainline and the mangio, But mainline rvc just trains more accurately

ive tested this on biggie dataset michael jackson andrew tate 2pac freddy mercury

etc

all to the same result

i used to love so-vits before rvc then i favoured rvc, then mangio was a thing and i tried that for a while but it had many bugs then applio was introduced but it never had the accuracy or same impact as mainline rvc i dont know what was changed or altered in the code but the accuracy in model performance is not the same, for me anyway

but you know each to their own whatever works for you guys

[knotty moth]

you still judge the same thing on the latest applio 3.2.8 bugfix? and have you tried some other pretrains like KLM 4.3 on both mainline and applio?

note that the default pretrain was trained from vctk dataset with background noise, and not denoised, so it might be fine for average dataset being not very "clean"

[polar grail]

Joined this server a long long time ago, but was on a laptop that didnt have good enough specs to run anything. On my new pc now, anyone wanna put up with my bs an help me set it up?

[surreal garden]

There's no other way?

[spare geode]

how can u make ur the voice u want to work its not working for some reason like the thing work it just i can't get it to wokr on discordd

[deft turret]

anyone else have an issue of split audio shortening an audio clip?

[glacial pollen]

this is most likely a placebo

I've been training using rvc for over a year and recently switched to applio as my fork's base

it's most likely a coincidence man because there's not a single thing in rvc that makes it better architecture wise

Trust me, it's a matter of randomization, weights init and the nature of stochasticity in a way

Enforce fp32, use deterministic tweak and test them both side by side.
Then, If you want, send some test samples from both and I'll do spectral comparison for you to rule out " differences "

(( oh yeah, also each inference is randomized and different even if by a little. Sometimes better, sometimes worse

[dim jewel]

Hey, guys. Are there noticable difference in training from loss and lossless audio samples?

[low shard]

Don't think so

[low shard]

Lossy files are lower quality

[dim jewel]

But does it matter if I can't tell a hearing difference in original audio sample be it loss or lossless?

[low shard]

Realtime Voice Changer for calls? Wrong channel, tell me ur GPU name in #🔍│help-w-okada ( a screenshot and also the link of the tut u followed)

[spare geode]

thx but i can it working some how lol tho

[low shard]

I don't have a comparison but usually you shouldn't on lossy

[shell holly]

I have a question: is there a difference if we train a model with 1 hour of audio or 5 hours of audio?

[low shard]

iirc there shouldn't be a difference after 1 hour

[valid spruce]

Guys, could someone send me the link to Amplio for training?

Another question! Which is better? KLM 4.3 [x3] or Original?

[raw stirrup]

im just explaining my experience ive been using all versions since they were released i started with tacotron 2 in 2020 then talknet2 so-vits etc and for me rvc mainline outperforms everything based on default setups and default pretrains ive tested custom pretrains on both applio latest version and mainlinme rvc and rvc still trains my models to much better accuracy im not sure why that it is but im just saying thats the case for me so-vits isnt even in the picture anymore im not trying to shit on applio i respect people trying to make better things and improve stuff but for me it hasnt been the case mainline latest nvidia build just works to perfection for me ive trained many models on both applio and mainline mainline wins in accuracy

again u can hear the difference night and day rvc is almost realistic where as applio just sounds off

[knotty moth]

u still haven't answered whether you have tried other non-default pretrains using mainline and applio

[raw stirrup]

i said in the reply i have tried custom pretrains on both versions

[lucid stream]

Does the voice changer **MMVCServerSIO **program lead to a ban in the Fortnite game?

[raw stirrup]

i dont know what you want me to say bro rvc mainline is just more accurate but u dont have to accept that its fine i have 0 reasons to lie i would love for applio to work the best as its newer and more maintained but the reality is maniline for now just works much better

[raw stirrup]

shouldnt do unless youre miss using it

[hallow thistle]

https://cdn.discordapp.com/emojis/1016021422809817160.webp?size=48

[hallow thistle]

For W-Okada, go to #🔍│help-w-okada for more information. This channel here #✨│ai-help is all about RVC and certain RVC GUI forks.

However, I don't think using W-Okada in Fortnite can get you banned. Unless you're using it for something bad.

[hallow thistle]

-gui

[azure marsh]

https://cdn.discordapp.com/attachments/1122285248844144733/1203460490475343953/caption.gif?ex=65d12cec&is=65beb7ec&hm=bd2fb8d010006dd7c6e3c1c67d3ae846fd1478e1a3124c544c31b43086fe54aa&

[hallow thistle]

Training model with better quality, with the same pretrain model and dataset, doesn't have to be just RVC GUI, Mangio, Applio or the Weights. The better quality can happen randomly, even with the same pretrain model and dataset. Unless you've got that good ears to hear the difference.

[raw stirrup]

i get what youre saying and that would be the case if i only tested one dataset on all 3 versions but ive been using multiple datasets from different people biggie 2pac andrew tate buklau freddy mercury loads and they always sound more realistic on rvc mainline

i train all models to 200 epoch for a starter and they just train better on rvc cant explain why i didnt code these programs but thats been my experience

i even tried adding the configs from rvc to applio and they still trained differently off sound etc

so weird

[karmic fjord]

french montana model not found?

pleaseeeeeeeeeeeeeeeeeee

[hallow thistle]

French Montana voice model exists on Weights. But it seems like no one made a model of him in #1175430844685484042, sadly.

You can request a voice model of him in #1159289738314919936 or train one by yourself.

[karmic fjord]

doesn't sound like hima at all

KITS AI? or any website free?

[hallow thistle]

Kits.ai is now a scammy site, and has some expensive tiers to use their service. Weights is one of the only free websites alive.

[raw stirrup]

respect to the owners/creators of weights.gg

[daring ermine]

https://tenor.com/view/bowing-thank-you-tom-and-jerry-take-a-bow-chasing-gif-16306244066707271532

[analog vault]

does anyone have most updated guide how to create a dataset and rvc model of my own voice, goal to use it to merge with rvc's so w-okdada would sound really good

[woeful torrent]

How do i in port models with RVC Gui

[raw stirrup]

http://docs.ai-hub.wtf/

[raw stirrup]

just add them to folder weights

inside rvc folder

if youre using the latest rvc mainline its assets/weights

[balmy reef]

Hey is anyone else getting a "build error" when trying to duplicate an RVC space on Huggingface

The space works normally without duplication, but whenever I duplicate it gives me an Exit Code 1 err.

[proven hill]

wich space?

[balmy reef]

For the RVC V2 spaces made by rejekts

[proven hill]

maybe its a space problem (?)

[balmy reef]

That's what I was wondering, I have another space that I recently restarted and it seems to be working fine.

Idk if this helps

This is the err I keep getting on only RVC V2 spaces (Rejekts ed.)

[low shard]

Send link

[balmy reef]

https://huggingface.co/spaces/Clebersla/RVC_V2_Huggingface_Version

[low shard]

That's old, don't use it

What's ur PC GPU

[balmy reef]

I use standard CPU

I used to duplicate space with no issue. Is there a more up-to-date one for RVC?

[low shard]

Oh NVM do it on cloud

[low shard]

• Cloud (remote good pc, easier and faster than ur PC but it's limited):
  • Ilaria RVC Zero: fastest and simplest that you can get for free
  • Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
  • Applio Colab: max 4 hours, not granted, of GPU

Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero

[keen coyote]

is there any ai voices that support things like laughs?

[jaunty mason]

rvchuggin face plz

[low shard]

-hf

[azure marsh]

Suggestions for @jaunty mason

🤗 Huggingface Spaces

• UVR5 UI, by Eddy and Ilaria Huggingface Spaces
• Ilaria RVC Zero, by thestingerx Huggingface Spaces
• RVC⚡ZERO, by r3gm Huggingface Spaces
• Applio, by IA Hispano Huggingface Spaces

More commands: -audio, - colabs, -kaggle, -local, -rvc, /help

[low shard]

the last 3 options

[edgy marsh]

Hey?

Anyone know how to uh

use the files for voices

[proven hill]

depends on what you wanna do

[edgy marsh]

Uh

I wanna do a real time voice changer

and something just for audio recordings

to make voicelines

for combine soldiers

[proven hill]

-rt

[azure marsh]

Interaction has expired, use the command again for a new interaction.

💻 Local Realtime RVC

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[edgy marsh]

Thanks

[proven hill]

for the voicechanger

[edgy marsh]

I was so confused

I'll look into this man

thanks

[proven hill]

@low shard can help you better

[edgy marsh]

Where do I download it?

[proven hill]

theres the download section

[edgy marsh]

ok

do i just click

click uh MMVCSERVERISO

MMVCServerSIO

[hallow thistle]

For W-Okada, you can go to #🔍│help-w-okada. This channel #✨│ai-help is about RVC.

Yes, click that .exe program.

[edgy marsh]

So

I clicked t

now what

[hallow thistle]

https://cdn.discordapp.com/emojis/1016021422809817160.webp?size=48

[edgy marsh]

i have done this

does it not work?

have i been scammed

[hallow thistle]

This means W-Okada successfully launching. It should launch one of your web browsers automatically.

[proven hill]

text comprehension 0

[edgy marsh]

WAIT

YES

DUDE LOOK

[hallow thistle]

Damn.

[proven hill]

yes with viruses and rats

[edgy marsh]

now

i gues

what now

Do I just drop the folder somewhere

[hallow thistle]

I'm not gonna tell you every one by one step.

[edgy marsh]

you dont

show me the guide now

[hallow thistle]

-realtime

[azure marsh]

Interaction has expired, use the command again for a new interaction.

💻 Local Realtime RVC

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[hallow thistle]

First link.

[edgy marsh]

kk

[proven hill]

not even a thanks

just a command

[edgy marsh]

thanks

oh boy

[hallow thistle]

For more information or any issue using W-Okada, you can go to #🔍│help-w-okada. This channel here #✨│ai-help is about RVC and RVC GUI forks.

[knotty moth]

not the tiger18n one

[edgy marsh]

Does it work on Roblox?

[proven hill]

#🔍│help-w-okada

[hallow thistle]

This is like the second time I've repeating this.

[edgy marsh]

I can't hear myself

And it just

is not working

I can't hear myself

I'm trying different settings

this

what is

this

[hallow thistle]

https://cdn.discordapp.com/emojis/1016021422809817160.webp?size=48

Woo. Why did you set the audio system to "server"? The client mode works best for certain scenario.

[cursive vault]

where is the "Unwa's big mel roformer beta 4"?

[hallow thistle]

https://cdn.discordapp.com/emojis/1142118384775987210.webp?size=48&animated=true

[cursive vault]

._.

[hallow thistle]

Have you tried setting like this on W-Okada? This could make a difference.

Again, if you have any issue about using W-Okada, go to #🔍│help-w-okada. This is like the damn third time I've been repeating this to you.

Anyways. It seems like you've downloaded the DirectML version of fork W-Okada, two GPUs appeared on your W-Okada terminal, which can mean your PC either has an integrated or AMD GPU. That GPU part on GUI, you're still setting it to CPU, which means W-Okada will use your CPU to inference instead of your GPU.

Almost every setting on your W-Okada seemed to be set way wrong.

[edgy marsh]

it works

the issue is it's nto workin ingame

in roblox

I ask players

and its

not working

I'm literally in the Help w okada

[brittle wing]

It keeps saying cpu for gpu

[proven hill]

can you show the error?

[brittle wing]

no perms

is there a media chnnael?

[proven hill]

@knotty moth

shouldnt he have perms?

[knotty moth]

what? @oak plank could u check this?

[brittle wing]

It keeps saying cpu in the gpu area

[proven hill]

voice changer?

try #🔍│help-w-okada

[oak plank]

I have given em AI Members role, they should be able to send images here now

I also updated the bot, now it gives AI Members role at level 2

So basically after 10-15 messages

[proven hill]

hello who are you

[oak plank]

I am Vijay

Manager in the server

How about you? :D

[proven hill]

ilaria ex admin

[oak plank]

😮

I seee

[proven hill]

you didnt know?

[oak plank]

Nice to meet you Ilaria!

[proven hill]

nice to meet you!!

[oak plank]

I think I've heard about you, not sure when tho hahaha

[proven hill]

probably negative 😔✊

[oak plank]

I joined in September so yeah lol

[proven hill]

fresh meat!

[oak plank]

Not really, idk I read somewhere you were an ex admin ig

[proven hill]

i assume youre from weights then

[oak plank]

Maybe when I was cleaning up roles

[oak plank]

Yep!

[proven hill]

yea then something happened and baboom

[oak plank]

I am also a server manager there

[oak plank]

[proven hill]

SO IT WAS YOU!!

my role :(

[oak plank]

[proven hill]

how things are going with the server btw?

recently staying here sometimes but idk

[oak plank]

You should have some role btww

Something that tells people you were a member of staff

[oak plank]

In AI Hub or Weights?

[proven hill]

contributor? i remember specifically told to not give me that 💀

[oak plank]

Not that

[proven hill]

both :3

[oak plank]

Something special for admins

[proven hill]

YAY

me wants

btw im also the creator of ilaria rvc

[oak plank]

It's going good, currently I'm focusing on Dreamtavern by Weights activity, it has 4.6k members lol

[oak plank]

Coool

[proven hill]

what is dreamtavern?

[oak plank]

Next, it will be AI Hub's activity, making it active af

[oak plank]

It's a web app where you can chat with AI Characters at https://dreamtavern.ai

So they also have a server

Dreamtavern app is owned by Weights now, so yeah

[proven hill]

good luck!
when i was admin it was pretty active, then most of the usuals left for good and it kinda died

[proven hill]

im gonna check it out!
is something similiar to sillytavern?

[oak plank]

Mhm! We will try our best to make this server even more better ☺️

[oak plank]

Not sure hahaha

I am also new to AI stuff tbh

[proven hill]

double fresh meat!

[oak plank]

I am good with discord bots and web apps tho

[oak plank]

😆

[proven hill]

what are you specialising into?

[oak plank]

web dev mainly, 3d websites, also curious to learn more about AIML

[proven hill]

you also removed my bot, i think to replace with yours(?)

[oak plank]

It was being discussed to remove the bots that are not used anymore, so we removed a bunch of bots 😅

[proven hill]

you should def check out a new LLM that its better than chat gpt 4o

[proven hill]

make sense, i didnt update it anymore

[oak plank]

[proven hill]

let me grab the name of the thing

[crude flame]

claude 3.5 sonnet

[proven hill]

DeepSeek V3

better results on everything

[oak plank]

Thanks for sharing that, Illaria! I will definitely check it out

[proven hill]

let me know if you like it!
its free on their site

[crude flame]

oh wow its cracked

[oak plank]

Suree

[proven hill]

its also open source but you kinda need a bunch of A100 to run it lol

[oak plank]

Hello Razer

[crude flame]

hewo

[oak plank]

How was your day? :D

[knotty moth]

I suppose it's available in hf zerogpu? if so I'll check it later

[crude flame]

is good just suffering in val and looking into DeepSeek V3

[proven hill]

i truly dont know!
im gonna try making a hf port maybe

[oak plank]

Niceeee

[crude flame]

claude still mostly better at coding

[proven hill]

chinese people do be really stay on top regarding ai stuff tbh

[proven hill]

all types?

[oak plank]

I use AI for understanding programming concepts sometimes

because sometimes there is code I have no idea how it works

😭

Especially GLSL

[crude flame]

beats it in 6 benchmarks

out of 8

[proven hill]

how about the one i said?

worse?

[crude flame]

DeepSeek V3 is worse then claude at coding

[proven hill]

rip

[crude flame]

unless this is wrong https://github.com/deepseek-ai/DeepSeek-V3?tab=readme-ov-file

oh wait

am i blind

or dumb

or both

ok flip what i said

i am both

nvm DeepSeek V3 is better

[oak plank]

😭

[crude flame]

this is what val does to you

dont play val kids

[proven hill]

lmfao

im super updated and better i swear

[oak plank]

[oak plank]

Hello Maki, How's it going?

[knotty moth]

new year to me 12 hrs ago

[oak plank]

Happy New Year Maki!

[proven hill]

also, youtube is shit

i dont remember where i was talking about youtube quality today

[lone nest]

Do you know what exactly happens there? Is the 100th epoch retrained or will the model end up with 1 additional epoch?

It seems to read the epoch for D and G files correctly tho so I'm confused why the training is offset by 1

[proven hill]

probably a glitch

[knotty moth]

try the latest applio that should resume from epoch 101 instead of 100 again

[weary elk]

anyone have an idea why in discord for the input devices it says cable output when it should be cable input?

actually tihs happens everywhere

[lone nest]

Other than small dataset and 2^n efficiency, is there any advantage in setting lower batch size than the GPU can handle?

[knotty moth]

batch 8 for most cases, or 4 if short and less diverse (<10 mins)

[latent kettle]

Is there any open source project available like Suno Ai ?

[carmine siren]

-colab

[azure marsh]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

More commands: -audio, -hf, -local, -kaggle, -realtime, -rvc, /help

[lone nest]

Going over 8 makes training faster per epoch, is there any drawback? In case of big dataset

[knotty moth]

easier to overfit and make sure the dataset is diverse enough, though more recommended for training a pretrain

also batch 16 fp32 could take up to 32 GB vram without bottleneck

I'd rather choose batch 8 fp32 better than batch 16 fp16 while taking the same amount of vram

[white rose]

guide

[simple ore]

because RVC-Boss is too lazy to make one line fix for resume

[simple ore]

batch 16, fp32 even without checkpointing

fp32 does not double the whole vram

5GB vram was the base for everything other than fp32 weights, with fp16 it may take total of 8GB, with fp32 it may take total of 11GB

[knotty moth]

prob different hardware, in kaggle T4x2, batch 8x2 causes slower time per step

[wispy lodge]

multigpu configuration may be slower. Though which Applio version?

[low shard]

4 am 😭

[proven hill]

:3

[azure marsh]

Interaction has expired, use the command again for a new interaction.

💻 Local Realtime RVC

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[dry kestrel]

i am rn

...

[brittle wing]

guys.. how do i use find models💀

[low shard]

You can search rvc ai voice models at:

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

if there isnt one, you can:

• #1159289738314919936
• #1191429836321849435
• make it yourself with our docs guides https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/

[earnest musk]

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

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

[craggy brook]

How can I fix this sound so that it sounds like purah from zelda totk. He's very robotic and doesn't have proper facial expressions.

[shell tide]

qERW

\AQ WE

[tame mica]

yass get it king

[craggy brook]

Is it related to this topic?

[sinful lichen]

how do i need to fix weird sounds, because i'm speaking one time its not take my voice, second time its really loud at voice still dont got, my mic is very good

[brittle wing]

hi im having a problem with uploading models ? it doesnt work when i press upload then select the pth and index files, and press upload again it just doesnt do anything at all and stays like that, also the voice cuts off like insane no one would understand

[surreal spruce]

-colab

[azure marsh]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

More commands: -audio, -hf, -local, -kaggle, -realtime, -rvc, /help

[rough oyster]

is applio broken on mac? i have a friend with an m4 mac and conversion isn't working for them

[brittle wing]

whats good foe

[raven burrow]

Hey guys quick question here. What's the fastest way to train in cloud? Paid Colab? Wich gpu? do we have some info about that?

[brittle wing]

@acoustic scarab help this fella out

[acoustic scarab]

https://tenor.com/view/uni-unico-tuxedo-cat-munchkin-cute-gif-13778035118426076010

i dont use applio

• a windows user

[brittle wing]

@hearty idol help this fella out

[hallow thistle]

That's an insulting.

[brittle wing]

what

[rough oyster]

nevermind, it just takes longer than usual

🙏

[regal geode]

realtime voice changer client stops working suddenly

i know it stopped working when the numbers beside vol are not moving when i talk

[storm shard]

is pytorch down?

I formated my pc and can't run RVC anymore

https://github.com/Tiger14n/RVC-GUI?tab=readme-ov-file

I know is really old by now but I got used to it and would like to keep using it

could somebody help me?

[simple ore]

you're installing oudated software with libraries no longer available

[carmine hearth]

https://download.pytorch.org/whl/cu118
Does manual installation also fail?

[storm shard]

damn. is there a newer version or a way to get those libraries?