✨│ai-help

[unique rock]

How so? Well I want to try Colab

[low shard]

basically the thing Noobies said

[unique rock]

oh, what did you remove what you showed me in the image?

[simple ore]

--branch is an option to clone a specific branch, you want the main branch, not 3.2.8-bugfix, --single-branch is an option to clone just one branch

[austere stream]

hello, i just finished training and am not really sure which step i should use as the final model. it looks sort of different compared to the other graphs of models ive trained.

dataset is small, like 22 minutes

[simple ore]

if you want main, you need to remove both options

[simple ore]

total being at 48 is somewhat hight

take 55k step model and give it a try

[austere stream]

thank you!

[unique rock]

Sorry, I still don't understand

[simple ore]

remove the highlighted part

[unique rock]

and after that?

[simple ore]

proceed as normal

[unique rock]

and where can I charge the pre train klm 5 and refine gan?

[wanton valley]

i been doing my audios with 44k and i use 40k rate in the collab, do this affect heavily the model? i tried to change the rate to 40 or even 32 but the audio turn slow and low pitched, does that even matter? do i make the audio match his original speed and pitch or i need to redo everything again? sorry for this dumbass questions, i been doing models time from time since 2022 and i never stopped to think about this

[simple ore]

sorry, looks like the colab code needs to be updated too

not my jam, so cant help with that

[simple ore]

are you talking about inference? traning?

[wanton valley]

alright my bad, i didnt explain properly, i mean that i do my audios with audacity and i use 44k by default for everything, the collab usually are 32k, 40k and 48k, and i want to know if this affects the model making it bad

and if i can change the rate simply here, the audio turns slow and low pitched, and i want to know if that matter, or i need to pitch up the audio with their original speed

[simple ore]

it is a sampling rate, export audio from audacity and pick the sampling rate there

do not change the sampling rate in audacity itself, it simply changes the playback speed

by default RVC preprocess resamples the input audio to the desired SR

so dont really need to bother with changing it in audacity

[wanton valley]

https://tenor.com/view/drake-yachty-censored-not-drake-tech-support-gif-8658410013845332703

gad damn it i dont have perms

alright thanks

[unique rock]

oh, i'll wait

[simple ore]

I think it should be usable on UI colab

but noUI needs parameters passed over

[unique rock]

Yes I want to use the UI, since I want to try out the new things it brings.

[hallow thistle]

https://tenor.com/view/think-bro-gif-22403438

[warm oriole]

What setting are we talking about here?

[knotty moth]

changing the rate here doesn't resample it, but actually speeds up/slows down and change the pitch, so changing from 32k to 48k will speed up to 1.5x and raise the pitch to 1.5x

[knotty moth]

for this case, you should resample it to the target sample rate (40k)

[wanton valley]

Anyways i coulndt make the model yet, got a blackout at first try and the second try GPU collab stopped working or something like that so i guedd i try Tomorrow

[low shard]

The pitch

[warm oriole]

okey so do i have to use speech to speech for good result cause tts isnt giving me much of a good result

[low shard]

rvc isn’t made for TTS

it’s speech to speech

applio just uses ms edge to make an audio from tts, then uses it as an inoug in rvc

[fathom ore]

is it possible to train on colab anymore? i used to do it before, any link?

[glad zealot]

Stuff that are in #📰│dev-updates should work

#📰│dev-updates message

[fathom ore]

thanks but i dont see any tutorial links? it looks so different

(Do i use Applio No UI or with UI for training?)

[valid spruce]

Hi guys

I have a question

Does Applio have any way to continue the model after the first training?

Because this was in RVC Disconnected, so I would like to know because I always used this function to create my models.

[simple ore]

you can resume training

[rancid pine]

Which UVR model isn't too heavy on lossy audio? Trying to retain the bitrate instead of clean vocals

[knotty moth]

no it's recommended to use the lossless flac/wav format

[rancid pine]

I'm afraid I phrased that poorly, but some UVR models tend to knock down the quality of the files a notch and I'm trying to find which ones aren't too rigorous with it

Some guy gave me a few tips but the message was deleted

[brittle wing]

Does applio no UI pass new vocoder parameters

[valid spruce]

How can I do this? Could you help me?

[knotty moth]

where's the deleted message? I didn't even see it logged

[rancid pine]

@knotty moth Was from a while ago, around this timestamp

[simple ore]

just start training again, do not run preproces/extract features,

add more epochs, use same batch size

[simple ore]

they are not on the screen, so no

[cobalt oyster]

Guys anybody from tech field I am joining clg next year which course is best

[knotty moth]

I don't recall that post in 2023, perhaps you mean the lossy source and that none of the separation models are made to reconstruct the missing high end frequencies due to the lossy format

[brittle wing]

So KLM 4.0 32Hz refinegan?

[valid spruce]

Okay! ty u

[rancid pine]

Yes

I'd like to reduce its effects as much as possible, hence my question

[uneven relic]

Hi
im trying to mix 2 models on applio but i have an error, do you know why ?

[knotty moth]

it must be two model pth files (the actual model, not the G/D files nor the index files)

[knotty moth]

what effects to remove? unfortunately don't expect good ones other than the dereverb/echo and denoise models

[uneven relic]

Thanks

[knotty moth]

even the audio upscaler/lossy enhancer models are still far from perfect enough for dataset making

[rancid pine]

Separating vocals/instrumentals as well

I'm unsure which ones don't reduce as much of the audio quality

[uneven relic]

this gives me a single pth file by mixing the two pth
What do I do next? I need an INDEX file to make a new voice

[simple ore]

there's no index mixing

[knotty moth]

refer to the recommendations https://docs.aihub.gg/rvc/resources/dataset-isolation/#the-best-models-for-uvr-are

[rancid pine]

Do they ensure the best audio quality?

[uneven relic]

Ok, in fact I'm trying to create a realistic French woman's voice, but there's not much choice in French voices..

[simple ore]

you can use an index from any french model

[uneven relic]

the index file affects what? intonation?

[simple ore]

prononciation

english audio inferred using a french index will sound like french person speaking

with an accent

[uneven relic]

Ok i see

[real field]

does using the original https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI-only local RVC application for conversion have any difference between other forks?

[empty oxide]

I did a clone of Emma Stone’s voice using Mangio RVC and another cloning on a paid site. On the paid site I provided only 1 minute of original audio and cloning was done quickly. On Mangio RVC I used 2 minutes of original audio, and 200 epochs. Still, the cloning on the professional site is better. The voice is more similar, and includes the actress' hoarseness. Why would it be?

[frozen sedge]

ok I have a problem with the software the file called RVC GUI.bat does not launch even after a long wait what should I do???

[visual pelican]

can i ask what link that is? i used to have that link but i forgot it

the link for the website

[pastel oak]

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

[pastel oak]

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

[frozen sedge]

why not is it an alternative?

[low shard]

RVC GUI is outdated asf, you prob watched some yt video on how to make ai covers

what's even ur pc gpu and what u want to do

[low shard]

Mangio RVC is abandoned since 2023

[low shard]

that's the mainline /original one, there's applio which is more user friendly and got some other features, but it's ur choice on what to use

[empty oxide]

Thanks!

[low shard]

are u looking for training models? what's ur pc gpu

[rugged saddle]

-colab

[karmic olive]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• Hina's Mod 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
• Hina's Modified Original W-Okada's Realtime Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab
• 🆕 Music Source Separation Training (Inference), by Jarredou & Makidanye 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

[rugged saddle]

where i make ai cover with web ui?

[empty oxide]

Yes. My GPU is kind of old, NVIDIA GeForce GTX 1650.

[low shard]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• 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 daily, not granted, of GPU

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

[empty oxide]

Great! I will take a look at those options.

[tame mica]

damn

uh ignore what i changed

[unique rock]

How do I put a KLM pre-train in Applio UI?

I have the links for huggingface

[simple ore]

on training tab

[unique rock]

Should I download and upload them like this? Or can I download them directly from Applio?

[simple ore]

or you can put them on the drive and provide a path

/content/drive/MyDrive/G48k.pth

[unique rock]

in the download section it is not possible right?

[unique rock]

Should the folders be separate? I'm already downloading klm 5 32k.pth

[simple ore]

G and D have unique names

[unique rock]

So I have to upload it separately, I thought they were in a separate folder, and it goes directly as I understand it.

[unique rock]

Noobies, where do I upload the route now?

[simple ore]

just use the screen

[pine parrot]

How can I train a ai model with nothing but a MP3 file?

I want a cloud mp3 to rvc if possible

I can pay for this just need it

[unique rock]

It won't let me use refineGAN in Applio UI

[harsh ravine]

hey

where can i find the face swap collab

[knotty moth]

to do your "mp3 to rvc" sorcery, you should learn everything in https://docs.aihub.gg/essentials/how-to-make-voice-models/

[unique rock]

How do I transfer the auto backup to another Drive account and continue retraining in the other account?

[simple ore]

1. make sure you did not accidentally selected an empty space with a mouse cursor in applio window, press Esc a few times

[terse nymph]

I resolved this and forgot to delete the message, my bad

[marble forum]

How do I make AI sing?

[low shard]

What's ur PC GPU

[low shard]

https://discord.com/channels/1159260121998827560/1163571683848900629 , also Colab isn't suggested unless u pay for pro, lightning.ai is better and u can find more info about it on my GitHub,
Also this isn't the right channel to ask in

[rocky zealot]

Hi, is there a way to maintain original style and specifics of the original voice from the song (effects, chorus, etc.) and change only the sound of it to the one form trained model?

[low shard]

wdym

[rocky zealot]

Can you elaborate? I am still new in this 😅

[low shard]

wdym is an abbreviation for what do you mean

i don’t understand the “change only the sound of it to the one form of trained model” part

[rocky zealot]

ok, I want to convert the voice in the song using custom trained model using RVC. But, when I do this I lose the original voice qualities and effects, so I am wondering if there is a way to maintain those when converting.
Lalal.ai can do this (there is an option for that) but they don't support training custom voice.

[low shard]

Does this happen after extracting the vocals or only after doing inference

because lalal.ai seems to only extract voices, not doing ai covers

[crimson depot]

What is the best method to create ai covers locally?

[rocky zealot]

lalal. ai has voice changer: https://www.lalal.ai/voice-changer

I would like to achieve similar effect on RVC with my custom model as the conversion in Lalal.ai with options:
keep original accent
and
keep original tone

[low shard]

what’s ur pc gpu

[crimson depot]

Nvidia GTX 1060 6gb

[low shard]

I think u have to play around with the pitch, which is the tone, and can depend based on original voice and model

[low shard]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• 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 daily, not granted, of GPU

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

[cunning trout]

Hello, may i know how to get the desktop client? The one that popped out is website client(Fork). Sorry if this is dumb question.

[rocky zealot]

ok, thanks. I Will try

[low shard]

I think you’re talking about wokada, this is the wrong channel, RVC is for inference on prerecorded audios and train models, while wokada uses rvc for realtime inference

there’s 2 main versions of wokada
The original made by wok
The modified version made by Deiteris

Wokada uses a web user interface, it’s all normal it opens ur browser

for other things pls ask in #🔍│help-w-okada

[cunning trout]

i see, thank you.

[crimson depot]

applio needs internet locally?

[low shard]

no, only for TTS since it uses ms edge tts api to make an audio, to then use as an input for rvc, since rvc is STS

[crimson depot]

I need it to separate audios voices and make ai unused internet

[low shard]

Applio is for inference (use rvc models) on pre recorded audios and training (making) models

and tts, but that’s only with internet, the other thighs are offline

about separating vocals and instrumentals, you need another program

[crimson depot]

Which?

[low shard]

either local uvr5 https://docs.aihub.gg/rvc/resources/dataset-isolation/#local-uvr or @viscid moss ’s UVR5 UI https://github.com/Eddycrack864/UVR5-UI

[crimson depot]

I am installing this https://github.com/eddycrack864/uvr5-uii, what can I do with this?

[low shard]

separate vocals and instrumentals

[crimson depot]

perfect

[unique rock]

Does anyone know why Applio UI does not generate index?

[simple ore]

show the log

[unique rock]

He stayed like that and didn't train.

[simple ore]

you f'd up

[polar tendon]

It doesn't see the audio

"Preprocess completed in 0.00 seconds on 00:00:00 seconds of audio."

[low shard]

what’s inside ur dataset.zip

what did u use to make it

[polar tendon]

You mean me?

[tame mica]

he was replying to pain's message

[low shard]

Oops wrong ping

[low shard]

I meant @unique rock

Role color confused me

[unique rock]

I uploaded the audio normally, without compression

[knotty moth]

check the structure of the dataset folder and files, or maybe try the "dataset maker" tool

[empty oxide]

I 've created a voice model in weights.com, but it comes with pth and json files. So I can’t run it in applio. Is an index file absolutely necessary to run a model in applio? Can we get an index file from weights.com?

[simple ore]

index is optional, there's no issues running weights.gg model in applio

[empty oxide]

Thanks, thanks. This is strange. I can run in applio models I download here and models I create in applio itself. But the execution of the one from weights.gg gives an error message. I will try to restart things here.

[simple ore]

what error message?

[empty oxide]

I restarted the computer, and now it's working. Thanks

An error icon appeared in applio. I didn't manage to copy the message in cmd.

[tropic plank]

-colab

[karmic olive]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• Hina's Mod 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
• Hina's Modified Original W-Okada's Realtime Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab
• 🆕 Music Source Separation Training (Inference), by Jarredou & Makidanye 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

[tropic plank]

-hf

[karmic olive]

🤗 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

[copper terrace]

got gtx 1650 and i5 10300H
I use gpu for rvc but the voices are not smooth they become robotic or brake down, any suggestion on how to get it smooth

[low shard]

what tut link are u using btw? and that might depend on how the model was trained

[copper terrace]

tut link?

[low shard]

yeah like what tutorial are you following / program using?

[copper terrace]

I tired few different models and try changing the software the setting ( idk any settings just randomly changing ) but its the same

[copper terrace]

vcclient win cuda something

[low shard]

that's not RVC

that's an old version of the original wokada, prob from a youtube tutorial, which is outdated

RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-reocrded audio (ai covers) and train (make) models

Wokada = uses RVC for realtime inference

@copper terrace are you looking for realtime voice changer for calls right?

[copper terrace]

I'm looking for vc and mainly recoding

[low shard]

so, you need both converting audio files and realtime ?

[copper terrace]

yes

[low shard]

then you need both RVC and wokada

[copper terrace]

the recoding in the vcclient is also bad / same as realtime

[low shard]

not just wokada, wokada is made for realtime

[copper terrace]

well... this is confusing lol

[low shard]

RVC = for converting files

Wokada = realtime

Let's start with RVC:
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• 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 daily, not granted, of GPU

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

I will tell you which Wokada to use in #🔍│help-w-okada , the newer wokada has also better quality and performance

[copper terrace]

I have gtx 1650, will rvc run good on it?

[low shard]

it will run good, won't be fast as cloud but you won't have any time limit

[low shard]

I have linked you everything you need for RVC in that message

[copper terrace]

thanks

[low shard]

the model will be the same dw. I will tell u that rq in #🔍│help-w-okada

[copper terrace]

ya got it thanks

[knotty moth]

it's recommended to use the non-realtime rvc as you're not doing live performance and want better quality

[polar tendon]

I was confused that you were replying to me but good thing it was an accident

[brisk canyon]

Whenever I click the start button on the voice changer stuff it says quote "waiting generate pipeline" then it says it fails saying "Pipeline not initialized"

[sudden river]

why does the ai smother a cuss word lol

[valid spruce]

Help

Autobackup Enabled
The tensorboard extension is already loaded. To reload it, use:
%reload_ext
Starting backup loop...
tensorboard
Files are up to date.
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1739398664.153193 3622 cuda_dnn.cc:8310] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1739398664.162762 3622 cuda_blas.cc:1418] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
No wav file found.
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1739398670.200721 3673 cuda_dnn.cc:8310] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1739398670.207588 3673 cuda_blas.cc:1418] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
/usr/local/lib/python3.11/dist-packages/torch/utils/data/dataloader.py:558: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.
warnings.warn(_create_warning_msg(
Starting training...
The parameters of the pretrain model such as the sample rate or architecture do not match the selected model.
An error occurred extracting the index: need at least one array to concatenate
If you are running this code in a virtual environment, make sure you have enough GPU available to generate the Index file.

[simple ore]

the only meaingful error is "The parameters of the pretrain model such as the sample rate or architecture do not match the selected model."

[low shard]

RVC is not Wokada

RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-reocrded audio (ai covers) and train (make) models

Wokada = uses RVC for realtime inference

Share ur PC GPU, and the link to the tutorial you followed in #🔍│help-w-okada

[brisk canyon]

Ah I apologize

[low shard]

It's fine, share the info I said in that channel

[alpine wedge]

hi im trying to find the most realistic rvc voice models but ones i find online is not always the best quality. Can someone help me?

[crude bolt]

most people who dont know what they are doing , i asume you want to change your microphone in other voice. If so, search for help in #🔍│help-w-okada .

Furthermore why am i always being pinged 😂 ?

[tropic kayak]

oky im write area

[stuck harness]

os the client working i have a wait for web server

[ripe whale]

hi everyone boi what here

are there any improvements on rvc v2?

[hallow thistle]

I'm a man. The improvement of RVC v2 over original RVC is most likely the quality.

For W-Okada the realtime audio changer, go to #🔍│help-w-okada. This channel #✨│ai-help here is about RVC and RVC programs.

[ripe whale]

bet ive been using Ilaria RVC for over a year now 😢

[hallow thistle]

Which Ilaria RVC? The one from Hugging face or GitHub?

[knotty moth]

not a good idea to ping a random person and as said above pls go to #🔍│help-w-okada

[hallow thistle]

You know, you should never pinging a random user you didn't know.

[knotty moth]

like if he knows him already

[knotty moth]

the huggingface zerogpu one is still good

[soft stratus]

Is Easygui crashed

[hallow thistle]

On Google Colab?

[soft stratus]

Yeah

[hallow thistle]

Can you send the screenshot of it?

[hallow thistle]

You know, you should never use a user copy version of a Colab notebook, as it might have been outdated compared to the main and original one.

[soft stratus]

And original too

[hallow thistle]

What are you gonna do with RVC? Training a model or doing AI cover? There are some better alternative notebooks to this.

[soft stratus]

Traning

[knotty moth]

if you copy the notebook, you should know how to resolve the package conflicts/errors by yourself

[soft stratus]

Wait a second

[hallow thistle]

Applio the RVC can do that.

[unreal parcel]

has RVC been made more efficient to train locally with a shitty gpu or do you still need a good one or to use google colab

[soft stratus]

The original version

[simple ore]

depeds on gpu

[knotty moth]

6 GB vram as bare minimum, or 8 GB as recommended
GTX 10/16-series gpus may work but RTX 20-series/newer are recommended

[unreal parcel]

ty!

[hallow thistle]

You can train a voice model on GTX 10/16 series GPU. Just don't expect it to finish that fast.

Any GPU that's older than GTX 10/16 GPU, you'd better avoid that for AI.

[soft stratus]

I tried before i use copy verison

[hallow thistle]

That's the same screenshot you sent to me. Since there's a bug going on with EasyGUI notebook, and you have no idea, we have no idea how to solve it, you may try Applio instead.

[soft stratus]

Ok

[rustic oyster]

can someone help me figure out which settings are the best to use

[low shard]

Wrong channel
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-reocrded audio (ai covers) and train (make) models

Wokada = uses RVC for realtime inference

Use #🔍│help-w-okada and explain ur PC GPU and CPU and the link of the tutorial you're following

[low shard]

It depends by model and the voice of the audio, u usually need to play around just with the pitch and index ratio

[low shard]

#📰│dev-updates that colab is broken

[knotty moth]

• pitch: up to +12 for male to female, or -12 for vice versa, otherwise leave it 0
• index rate: if not sure or want to use the model's accent leave it as is

[hollow valley]

yoo guys What is the best program/modulator for changing a male voice to a female voice? It doesn't have to be in real-time. I've used Applio before, but maybe there's something better since I haven't kept up with this topic for a while.

[low shard]

RVC V2 is the best

Applio is a fork (modified version of RVC

So nope there's nothing better, the quality depends on how the model was trained

[crimson depot]

HELP!

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\gradio\queueing.py", line 625, in process_events
response = await route_utils.call_process_api(
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\gradio\route_utils.py", line 322, in call_process_api
output = await app.get_blocks().process_api(
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\gradio\blocks.py", line 2088, in process_api
result = await self.call_function(
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\gradio\blocks.py", line 1635, in call_function
prediction = await anyio.to_thread.run_sync( # type: ignore
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\anyio\to_thread.py", line 56, in run_sync
return await get_async_backend().run_sync_in_worker_thread(
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\anyio_backends_asyncio.py", line 2461, in run_sync_in_worker_thread
return await future
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\anyio_backends_asyncio.py", line 962, in run
result = context.run(func, *args)
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\gradio\utils.py", line 883, in wrapper
response = f(*args, **kwargs)
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\env\lib\site-packages\gradio\utils.py", line 883, in wrapper
response = f(*args, **kwargs)
File "C:\Users\DT Ryan\AppData\Local\UVR5-UI\app.py", line 293, in roformer_separator
raise RuntimeError(f"Roformer separation failed: {e}") from e
RuntimeError: Roformer separation failed: [WinError 2] El sistema no puede encontrar el archivo especificado

Local URV5-UI by @viscid moss

[red violet]

why does using my main microphone input for the https://rentry.co/forkvoicechangerguide#opening-on-windows

decrease my pc audio quality, like during discord call and everything, is there anyway i can prevent this??

and acer purified voice keep saying
"the conference call has started"

everything revert back to normal after i turn it into "none"

weird

[latent cypress]

hi, please use https://discord.com/channels/1159260121998827560/1159290161683767298 channel for voice changer

[red violet]

oh, sorry!

thank you so much neo

[latent cypress]

no problem

[neat lintel]

I'm going to swtich from applio to e2-f5-tts through pinokio, wondering if my setup is good enough for it to handle it, CPU : ryzen 7 4800h, 16 GB RAM, GPU: GTX 1650

[viscid moss]

Que estabas tratando de hacer cuando salio eso? Yo creo que el problema es donde lo instalaste, dejalo en descargas no en esa carpeta de AppData

[dry monolith]

can someone help me

im trying to open w-okada but when i clicked start_http.bat it only opens cmd and then immediately closed

[low shard]

Elaborate

[low shard]

Oh you're following an old YouTube tutorial prolly

You're using the original Wokada which has lower quality and performance than the Wokada Deiteris fork

[dry monolith]

the tutorial is 5 months old...

[low shard]

Yep, that's old

AI progress is fast

[dry monolith]

what's the new one

[low shard]

Also, rvc isn't Wokada

RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-reocrded audio (ai covers) and train (make) models

Wokada = uses RVC for realtime inference

[low shard]

First, tell what u want to do, and ur PC GPU and CPU

[dry monolith]

just doing some AI voice changer, my GPU is RTX2060 and CPU is Ryzen 5 3600

[low shard]

Alright good setup, I'm guessing real-time so we have to talk in #🔍│help-w-okada since RVC ≠ Wokada

[dry monolith]

ok

[near bay]

Hi , i have a problem with generating the index file after training my model , im using Applio through Pinokio , after training the model successfully i can't generate the index file , im stuck at this since a few days , i tried reinstalling Applio and retraining my model but it still don't generate the index , if someone could help me that would be awesome 🙏 im running it localy with a 4080 super and here is the error message in the logs : An error occurred extracting the index: need at least one array to concatenate
If you are running this code in a virtual environment, make sure you have enough GPU available to generate the Index file.

[crimson depot]

No tenía ffmpeg configurado en el PATH. Después configurarlo y volver a intentarlo empezó una descarga que asumo que sea el modelo que elegí para la separación

[viscid moss]

Sip, primero se te descarga el modelo y despues lo usa.

Y funciono? O sigue el problema?

[crimson depot]

Ya funcionó sí. No tenía el ffmpeg en PATH

[viscid moss]

Nice

[simple ore]

you did not train anything... meaning you did not split audio, and all it trained on is two mute files

[near bay]

hey , i did not split audio in the options because i did it already myself , i have my dataset already split and processed so i did not chose any of those options in the applio settings , the model did train however , i have the pth files

[simple ore]

how big are the chunks? it does not load anything over 9s and training with 5s+ chunk is a waste of electricity

[near bay]

431mo , 300 epoch

[simple ore]

i'm not asking that

how many files are in sliced_audios

and what is in model_info.json?

"total_dataset_duration":

[near bay]

3959 files (around 6h of audio)

[simple ore]

probably too big

you dont need 6 hours for finetune

had it worked it would take ~5m/epoch x 300 epoch = 25 hours to train

did it take 25 hours? or did it take 10 minutes?

[near bay]

No it took around 1.5 hour to train with a batch size of 16 , also i did not put the audio files in sliced_audio

[simple ore]

sliced audios is created automatically by preprocess

anyway, your files were too big to fit into the training buckets

so it discarded most if not all of your data

[near bay]

I see , thanks for helping man , so what amount of audio is enough to have a good model ? i also have a metadata.csv in my data set but im not sure its needed with rvc ?

[simple ore]

30-60 minutes

make sure you use the built-in slicer in preprocess

[near bay]

The thing is that the slicer cut the audio not propelry and it cut in middle of sentences

[simple ore]

does not matter, it does overlaps

30-60 minutes requires only batch 4 .. 8

may require some experimenting depending on the dataset

[near bay]

All right , so i cut the big 6 hours audio file to lets say 45 minutes and then just let applio slice the files , it does not need the .csv with the transcripted text right ?

[simple ore]

nope

just the audio

[near bay]

Ok im going to try that , thanks for helping man 🙏

Is it ok to have the 45/60 min audio cut into 2 or 3 audio files or is it better if its a single one ?

[simple ore]

one file is fine

multiple big ones are fine too

[near bay]

Thanks mate

[rare kindle]

i can't find RMVPE help me

[low shard]

Holy shiet you're running an ancient version

Probably from YouTube tutorials

Tell ur PC GPU and what u want to do

[rare kindle]

thx

can u give new version link?

[low shard]

First, tell ur PC GPU and what u want to do

[rare kindle]

4070 and learnig voice

[low shard]

I'm guessing voice training

And ye good pc

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

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

[crystal vector]

Sorry if I'm reviving an old post, but the discussion topic currently interests me. Let's assume I want to use the D/G weights from one of my models as pretrain for another (same vocal, but different LR), but I only have the latest D/G weights saved in my log folder. Is there a risk of overtraining? Or should I save all the D/G weights during the first training to use the D/G weights that correspond to the epoch that seems most qualitative to me as pretrain?

[dusk oyster]

@low shard NVidia RTX 4060 Ti

[low shard]

Alright good enough to make models

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

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

Also take a look at https://docs.aihub.gg/essentials/how-to-make-voice-models/

[simple ore]

usually not a good idea as the model loses generalization

[crystal vector]

Can you elaborate, please?

[simple ore]

you take a pretrain that has male and female speaking and singing, you train it on just male speaking

it retains some ability to sing high notes

but it forgets some of the training

it is more specialized now

then you re-train it on another male voice

this model would likely not be able to sing high notes at all

[crystal vector]

I am aware of the potential issues you mention. However, in my case, the second training would concern the same voice, with a different LR for the training, so the problem of lost training parts doesn't really arise.

[simple ore]

if you resume training, it would use the previously saved LR

all you can do is give it a try and report back

I would not expect anything good coming out of it

[crystal vector]

Not necessarily: hence the approach of using the D and G from the previous training as pretrain.
But perhaps it may not be the right path to follow...

[simple ore]

the more you train the model, the more it forgets the previous capabilities

in the end it would only be able to produce the source dataset and nothing else

[karmic olive]

☁️ Google Colabs

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• Hina's Mod 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
• Hina's Modified Original W-Okada's Realtime Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab
• 🆕 Music Source Separation Training (Inference), by Jarredou & Makidanye Google Colab

[orchid flint]

Does anyone have any experience in training voices with heavy SFX? If I clean these kinds of datasets too much it will ruin the voice. Some robotic characters work well but others sound bad or don’t train well. I have had 1 train very well, 1 train ok, and 3 train very badly but can’t figure out why. I have decently sized datasets for 3 of them (10-20 mins) and large datasets for the 2 others (40-50 mins) yet that doesn’t seem to change how well or poorly they train. For the 2 large datasets models, one trained very well while the other trained very poorly. Both were distorted heavily by effects, yet had very different results. Anyone have any ideas?

[stiff goblet]

ensemble means the combination of models? i was using MVSEP

[valid spruce]

Hi guys

I need help with Applio

I was going to create a model but his voice came out completely different from the dataset.

I don't know what I did wrong...😕

[edgy tangle]

is this normal?

[orchid flint]

Uh, no

[near bay]

@simple ore Hey man , just passing by to say THANK YOU , i did what you recommend me to do and everything works , i have been able to train my model and continue working on my project i was stuck on for days , big thanks to you , God bless you 🙏

[weary temple]

wait is RVC GUI outdated

all my audio is super distorted

[hallow thistle]

-gui

[karmic olive]

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

[hallow thistle]

Of course, RVC GUI is very outdated. The last update for this one is from October 2023.

[weary temple]

i just followed a tutorial yesterday

lol

so which should i use?

[hallow thistle]

RVC and any other Python program work best locally if there's a GPU. What is your PC GPU?

[weary temple]

a 4070

[hallow thistle]

That's a good one.

-rvc

[karmic olive]

📚 Documentation

AI HUB Docs

https://docs.ai-hub.wtf

🍏 Applio Docs

https://docs.applio.org/

✨ More guides

• How to use RVC Mainline Colab by Cauthess
• AICoverGen Colab Guide by Eddy (Spanish Helper)
• Create a model with RVC disconnected (colab) by Angetyde

[hallow thistle]

Applio the RVC works best of all.

[weary temple]

cool

is it console based or is there a gui?

[hallow thistle]

More like a whole GUI, while there's a console window background showing the status of program.

[knotty moth]

still it doesn't have rmvpe, even mangio had that in 2023

[weary temple]

wait it says the file is 3000kb is that right?

[hallow thistle]

That file size doesn't sound right.

[knotty moth]

typo?

[weary temple]

nope

oh nvm i downloaded the wrong file

i found the actual file

[knotty moth]

which file was it?

[hallow thistle]

The file size of Applio zip should be 3.9GB.

[weary temple]

3.9 GB sounds right

[weary temple]

some error fixing file

or so says the title

[burnt bone]

hello, um sorta new to rvc but I just downloaded https://github.com/litagin02/rvc-tts-webui?tab=readme-ov-file this one and I don't quite know how to launch it.

the activate bat doesn't do anything nor does the acitvate file in the script

do i just need to install something else? i'm not sure how to install the tts web ui

[low shard]

that is outdated

literally archived and doesn’t get an update since 2 years

don’t watch youtube tutorials for this

and tell what’s ur pc gpu and what u want to do

[low shard]

RVC is for STS natively, it’s not TTS, another RVC fork that “does tts” is applio, which does the same thing, using edge tts api to make an audio then use it as an input in rvc

[burnt bone]

my gpu is a Nvidia rtx 3080 and i'd like to do tts or voice to voice if possible with my rig

so i'd be looking for applio then?

[low shard]

STS =Speech to Speech

[low shard]

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

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

yeah

[burnt bone]

awesome thanks i'll check them out!

and delete that file i just made lol

[low shard]

it’s gone

deleted

[low shard]

alr yw

[tidal isle]

Application error: a client-side exception has occurred (see the browser console for more information) .

[burnt bone]

Hi sorry for the ping but i got an error while installing \Miniconda3\Scripts\conda-script.py" shell.cmd.exe activate "C:\AI\applio\Applio\Applio\env" Do i need to redownload miniconda?

[low shard]

elaborate ur pc gpu and what tutorial link u are followijg

[low shard]

did u download the precompiled, unzip it, then run applo.bat?

[burnt bone]

the compiled for windows has been deleted i got the zip and unzipped it and ran the install.bat

[burnt bone]

i got the that zip from here. https://github.com/IAHispano/Applio

[low shard]

u used https://huggingface.co/IAHispano/Applio/resolve/main/Compiled/Windows/ApplioV3.2.8-bugfix.zip And not the source code zip, right?

[burnt bone]

think i used the source code zip

I think i'll download what you sent and let you know how it goes

[low shard]

the release says to use the specific compiled zip

[burnt bone]

I see where I went wrong I thought the windows was deleted and mistook the bug fix as simply that a bug fix. my mistake thanks again i'll let you know if all is good and running following the install guide on the doc

[low shard]

alright yw

[burnt bone]

got it to work now thanks again for all your help!

[low shard]

you’re welcome, for any issues let me know

[obsidian lark]

help pls:

/content/FIX
shell-init: error retrieving current directory: getcwd: cannot access parent directories: No such file or directory
/bin/bash: line 1: .venv/bin/activate: No such file or directory
---------------------------------------------------------------------------
FileNotFoundError                         Traceback (most recent call last)
<ipython-input-8-60c07a872b64> in <cell line: 0>()
     55 # Open a subprocess and capture its output
     56 get_ipython().system('source .venv/bin/activate')
---> 57 process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, universal_newlines=True)
     58 
     59 # Print the output in real-time

1 frames
/usr/lib/python3.11/subprocess.py in _execute_child(self, args, executable, preexec_fn, close_fds, pass_fds, cwd, env, startupinfo, creationflags, shell, p2cread, p2cwrite, c2pread, c2pwrite, errread, errwrite, restore_signals, gid, gids, uid, umask, start_new_session, process_group)
   1953                         err_msg = os.strerror(errno_num)
   1954                     if err_filename is not None:
-> 1955                         raise child_exception_type(errno_num, err_msg, err_filename)
   1956                     else:
   1957                         raise child_exception_type(errno_num, err_msg)

FileNotFoundError: [Errno 2] No such file or directory: '/content/FIX/.venv/bin/python'```

[low shard]

elaborate:

• ur pc gpu
• what u want to do
• what guide tutorial / google colab link u are using

[light pelican]

Hey a 4060 RTX 8GB good enough for training voice models ?

[low shard]

yes

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

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

[light pelican]

So applio is better ?

[obsidian lark]

1. nvidea 1650
2. i want to make an ai cover
3. here's the colab link https://colab.research.google.com/drive/1u1brjK8IZt647UsbZuGYfW29oFM2I4tk?usp=sharing#scrollTo=7SsCCgb9Ycqs
   i clicked the buttons one by one, put the path to an mp3 file in "song input" from my google drive and tried to click generate cover, i got these errors :((

[flint solar]

I’ve used them both and I honestly prefer mainline

I don’t get the applio hype

[low shard]

it gets more updates and it’s more userfriendly, but it’s ur choice

[analog obsidian]

because the author of mainline left the project to work in a new one (gptsovits), while doing this he prevents any dev to add new changes to mainline (1 year since this)
applio is mainline but updated basically

[low shard]

last update was 3 months ago

[analog obsidian]

last applio update was a couple of days ago

[low shard]

ye

[low shard]

more updates and userfrriendly

[analog obsidian]

said update was fixing onnx exporting
which no one uses that anymore bc fork allows amd users to use pth

[low shard]

@viscid moss maybe u should check this

[analog obsidian]

no real code update

[low shard]

also, ur pc gpu is good enough to do inference locally

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• 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 daily, not granted, of GPU

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

But u can also just wait for eddy response if u want

[white bough]

I do not have permission to write on the pre-train models channel, but it sounds to me that KLM5.0 does not have enough English in the data. The results sounds great but they don't sound 100% fluent in English. Anyone else finding this?

[obsidian lark]

oh, thank you so much!!
i also wanted to ask what program i could use to compile vocal and instrumental files?

[simple ore]

use my base then

[white bough]

Yours is great actually. It is in comparison with your RefineGAN model that I made that comment 😉

[viscid moss]

Hmmm lemme try, I update it yesterday and it was working

[low shard]

I think you meant separate

https://docs.aihub.gg/rvc/resources/dataset-isolation/

[low shard]

Weird

[viscid moss]

I think it was a colab weird bug, can u try again? It's working for me

[smoky dove]

Will using UVR give me better audio (I wanna run inference but my recorded audio is so shit

[granite minnow]

I used Anevew Aggressive DeEcho Dereverb to remove reverb, and while the reverb is gone, the echo is still there.

I need to remove the echo. Which option would be better?

UVR's DeEcho – it seems to cut off at 17.5 kHz, so I'm not sure if I should use it
SUCIAL's DeEcho models – There are several of them, but I don’t know which one is the best. Do you have any recommendations?

[eternal oasis]

Hello, any news about the compatibilty with RTX 5080 card ?

[knotty moth]

try manual installation but with torch upgraded to 2.8 nightly

might conflict with some other packages so you'd have to fix it

[neat peak]

Does anyone know how Google Colab's Compute Units are calculated? Like for how many hours can it last on training voice model or for how many Epochs?

[simple ore]

T4 GPU consumes 1.76 compute units per hour

that's from 2 years ago

not sure about now

[low shard]

Google colab GPU time is random

It can be max 4 hours daily tho

[low shard]

It can depend by how much you're using it, how often you use it

Kaggle gives 30 hours weekly of better GPUs instead

[neon hemlock]

How about a GTX 1650

[low shard]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• 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 daily, not granted, of GPU

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

or do you want realtime?

[thin edge]

a GTX 1650 can run Appolio to train model smoothly ?

[low shard]

eh, could train models but will be limited in batch size, dataset lenght and will be slower

I would suggest cloud

[hallow thistle]

You can train a voice mode locally with GeForce GTX 10/16 GPU. Just don't expect it to train that real fast.

[thin edge]

I forgot, when selecting the PTH file do I select the closest one to the Overtrain or select the overtrain itself. I read in the appolio guide it says the closest while in the tensorboard guide it uses the overtrain points themselves.

[simple ore]

it is really hard to reach overtraining

[twilit bridge]

Hey does anyone know why RVC is stuttering occasionally when playing Marvel Rivals specifically? This is the only game I've ever had this issue with. The playback will go from sounding very crisp & smooth, to occasionally stuttering a lot or cutting out entirely.

I've ran RVC on max settings in BO6, so I don't think it's a performance issue. I've also made sure that Marvel Rivals is the only game using my audio input (I use Voicemeeter with RVC). I've set my audiodg to High & affinity to CPU 2, but still same issue. This is the only game I have issues with. Does anyone have a possible solution who has encountered this too?

[crude flame]

Try some epochs around the ot point and keep the one that sounds best

[simple ore]

use voice changer fork

limit fps in the game

[analog obsidian]

marvel rivals is a gpu heavy game
nvidia drivers chooses to give all of the gpu resources to the game, leaving almost nothing to the voice changer

[twilit bridge]

I've tried capping from 240 to 144, but it still seems to have the same issue and at the same capacity. (The stutters are not better/worse regardless of which cap I try)

What's voice changer fork though?

[thin edge]

aight thanks

[knotty moth]

• go to #🔍│help-w-okada and get the fork one
• fps cap to 60/120, lower game settings, enable DLSS/lossless scaling (without frame generation)

[analog obsidian]

limit fps to 60, choose 1080p
all graphics to the lowest possible

that game is just very unoptimized

high graphics at 1080p ask for 8gb of vram 😭

[twilit bridge]

Thanks everyone--I'll try each solution 🙂

[crude flame]

Even with this if strange opens a portal you might drop to 40-30 fps

[twilit bridge]

Though, I'm not sure if I need to cap as low as 60. I've played some very heavy games at 140+ with great results (currently running a 4090), but I'll try to figure out if that's the issue--Marvel Rivals is the only one thats been tricky to get this working with

[analog obsidian]

LMAO yeah 😭

[analog obsidian]

your gpu can do all of this, the thing is that the nvidia driver decides to leave all of the 4090 resources to the game

[twilit bridge]

Sweet, so if I can find a way to prevent it from giving all its resources, I can potentially keep the fps higher than 60 (maybe) 😊

[analog obsidian]

you can't, thats how the driver works

[twilit bridge]

Oh.. rip.

But also yes, that game is very poorly optimized right now--every time I close it manually it tells me it has had a GPU crash, and it will tell me my GPU doesn't have 10 GB of RAM to compile shaders 1/4 times I launch the game lol

[analog obsidian]

i love that game but jesus the optimization is ass 😭

[crude flame]

Who's your main?

[analog obsidian]

jeff
just reached plat yesterday

[crude flame]

Jeff such a goober

[analog obsidian]

so silly

he also has the lowest win rate

😭

[crude flame]

How are plat lobbies?

[crude flame]

Rip

[analog obsidian]

games still feel like casual mode

lol

[twilit bridge]

Every rank feels the same for some reason in that game

[crude flame]

Lol, do people comm in plat?

[analog obsidian]

fr

[analog obsidian]

rarely lol

[crude flame]

Gm 2+ is quite different

[crude flame]

Wow

I kinda expect plats to start doing comms

[twilit bridge]

I'm GM2+ and I feel like some games are a coinflip still 😭

[crude flame]

FR

[twilit bridge]

I don't understand the ranking system at all because I'm seeing mistakes people make in Plat all the way in Celestial--consistently too

[crude flame]

It's too generous imo

I don't feel like I deserve gm 2 but here I am

[twilit bridge]

3 weeks ago, I solo q'd from bronze to GM in 1 sitting with only 4 losses. But at the same time, my main was stuck in Diamond 1--I played 16 hours on my main and just lost -14 points in total from the whooole session

[crude flame]

Maybe I got boosted by my godly tank duo

[twilit bridge]

Ranked in this game makes no sense and I'ma lose my mind 😭

[analog obsidian]

ranked is a mess rn

they're glorified casuals

lmao

[crude flame]

Wild

I once went on a 21 game win streak

[low shard]

Btw the right channel is #🔍│help-w-okada @twilit bridge

[crude flame]

Mhm

[twilit bridge]

Ohh, sorry. I wasn't sure where to post--I had posted on there previously but no one had responded. I'll also stop yapping about Rivals here too 😆

[proven hill]

im plat 2 in rct3

[crude flame]

That's heat

[proven hill]

you dont even know rct3 🚬

[crude flame]

Roller coaster tycoon

[proven hill]

based and tycoon pilled

[low shard]

It's fine, Wokada is for realtime, while RVC is for audios and model making

I wonder what happened to the Remote Code Execution exploit to marvel rivals tho

[proven hill]

i ate it

nom

[crude flame]

Didn't even know there was a exploit in it

[analog obsidian]

WHAT

[low shard]

https://www.youtube.com/watch?v=ydQKPBgWKsI
@crude flame

The launcher is written in python 💔

[analog obsidian]

😭 😭 😭 😭 😭 😭 😭

[twilit bridge]

Once you said that, that's the video I was already watching haha

[analog obsidian]

uninstalling rn

[crude flame]

Huh

[low shard]

The attacker has to be on the same network

But like

[crude flame]

Oh so it's useless

[analog obsidian]

oh

[low shard]

The game doesn't even verify that it's connected to the real game server, and the game runs on admin privileges bc anti cheat

From what I know it's only for people on the same network

Still really weird

[analog obsidian]

i hope they don't find a way to make it global

[low shard]

@twilit bridge

[low shard]

Still weird the game has an unsigned dll tho

[analog obsidian]

[hallow thistle]

They gonna hack the game likely they think Marvel's recent movies aren't too good.

[pliant crest]

Hina Mod AICoverGen gives me an error, it used to work just fine

How do I fix "'NoneType' object has no attribute 'setdefault'"

[pastel oak]

Id suggest you use something thats not this outdated. Do you want recommendations

[pliant crest]

yep

is there something similar to hina?

model that separates vocals and allows yt links?

[low shard]

Colab blocks YT downloads, u need to upload the audio manually

iirc all google colab block yt links

bc it goes against yt tos

youtube is owned by google, and google colab is owned by google

soooooo

[pliant crest]

Why tf didnt it happen before

[low shard]

because it's a google moment

[pliant crest]

how tf do you sign up to youtube, I have my fucking google account on both youtube and colab

[low shard]

The only way is that you manually download the video yourself and upload it

when you use google colab you use a remote good pc owned by google

[pliant crest]

oh ok

[low shard]

just google any "youtube mp4 downloader site" or use yt-dlp on github

[pliant crest]

ok thanks

I did it before with previous models but thanks anyways

[soft stratus]

Is google colab fixed

[pastel oak]

I recommend:

Train (make) RVC Models on cloud:

• Applio (ui)
• Mainline (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, no guide as of right now)
  • Applio by Shirou (UI, no guide as of right now)

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)

[low shard]

Most, not all #📰│dev-updates message

[soft stratus]

Thanks

[rose sapphire]

cant set cable output as output

[soft stratus]

When easy gui will be fixed

[low shard]

the creator isn't active in rvc anymore, I already told him since like a month, don't expect that colab to be ever fixed or come back any time soon

just use smt else

[low shard]

wrong channel, rvc isn't for realtime, we already talked in #🔍│help-w-okada

[rose sapphire]

oh shit mb

[soft stratus]

We'll be wait

[inner carbon]

Yo guys I am new with AI and I wanna train my model I have (I got a rtx 4070 ti) is there any recommdation for software to use!

[low shard]

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

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

[inner carbon]

Cool

another question is

like what is training your model

what do you need to train a model

because im a little confused

[low shard]

I think everything you need is in https://docs.aihub.gg/essentials/how-to-make-voice-models/

Our Docs got everything

[stiff goblet]

Which version of Applio ? 3.2.8-Bugfix?

[simple ore]

that's the main branch

not officially released yet

[stiff goblet]

Ok thx

[weary temple]

wait can english models speak other languages

[lime moth]

How GPU intensive is making an RVC model locally?
I'm asking because I am planning on doing a model based off of Michael Jackson in the Bad era
I haven't had much luck with training on colabs
For reference, I have a Gigabyte 3060

[simple ore]

if you use an index, there may be some accent, but other than that it is just a voice model

[simple ore]

VRAM matters, 8GB is good enough

[weary temple]

gotcha, i dunno why its doing this but it sounds completely messed up

especially if you get to like 0:50

[simple ore]

seems like the source audio is quite not what the model was trained to do

[weary temple]

are there seperate music models?

[simple ore]

I did not mean that

this is russian audio with english model

and this is the same without index

so if you're using an english model with an index and you're trying to infer a russian song that sung with a weird voice, the index search may not produce an adequate result

[weary temple]

ah gotcha, not sure how i got this then, it was from a different place but this website did fine with both english and non english songs

i couldn't access that site anymore/it got paywalled so i tried appolio but dunno how id go about getting similar results

[weary temple]

any ways i can work this or do i need a completely different model?

[simple ore]

turn off the index

[weary temple]

do you happen to know where that setting is in applio by any chance?

[simple ore]

or remove

[weary temple]

gotcha

thanks

[lime moth]

Just wanna make sure I don't burn out my gpu
yeah it has 8 gb of vram iirc

[simple ore]

it wont

[weary temple]

dunno why it dies after 30 seconds

[weary temple]

do you know what causes the seizures?

[simple ore]

at 30s?

i have not heard the original

could be the chorus?

[weary temple]

oh yeah youre right

[simple ore]

and you need to de-echo it

[weary temple]

gotcha

[simple ore]

this is not the type of audio that can be inferred

[weary temple]

is it possible to do in applio or should i use a different software?

[solar turret]

which voice changer software is the best according to yall, I used to have one but i forgot

i genuinely have no idea what im looking at in here im so lost

[stiff goblet]

so, i was training my model using refineGAN vocoder, but the result is so robotic, idk why.

i guess it's about pretrain right? i was trying to use the original one. I have checked your refineGAN pretrains. Should i download the 200 epoch version or 300 epoch version? @simple ore

[solar turret]

-rt

[karmic olive]

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

[solar turret]

what even is that,, i feel brain dead

i dont know any of these words i feel like a literal ape someone please help me

[simple ore]

I think 32k 300e is messed up, so perhaps 44k 150 or 32k 200

[stiff goblet]

ok, is this pretrain good for singing?

[simple ore]

yes

there's 10 hours of singing included

[hallow thistle]

How am I supposed to explain to you again? W-Okada is the AI realtime audio/voice changer. RVC is the AI audio changer.

Links from Automaze:

• the first link is a guide for Detris' W-Okada. Download links for NVIDIA and AMD/Intel GPU versions are there, alongside the installation and recommended settings for your GPU. This version of W-Okada runs better than the original version, especially the performance.
• the second link is the original W-Okada. Almost everything is just like the first link. The difference from Detris' fork is that original version eats more performance than that one, and it's still outdated no more update from there.
• I don't know about the third link, but I think it's a realtime mode in RVC GUI, which the program itself is way outdated and has its performance even worse than W-Okada.
• the fourth and final link is about how to use TTS with RVC.

[runic sedge]

Guys! I joined this server for voice models but I don't know how make AI covers with these models. (I watched some videos but some methods or programs don't work.)

[simple ore]

cuz most of the videos are outdated, just like the software they explain

[runic sedge]

And... how to make AI covers now?

[knotty moth]

you should check out this guide https://docs.aihub.gg/essentials/how-to-make-ai-cover/

[runic sedge]

OK
I'll read it

[solar turret]

Thank you mister sir

[carmine siren]

-collab

[carmine siren]

-colab

[karmic olive]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• Hina's Mod 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
• Hina's Modified Original W-Okada's Realtime Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab
• 🆕 Music Source Separation Training (Inference), by Jarredou & Makidanye 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

[carmine siren]

-kaggle

[karmic olive]

📘 Kaggle Notebooks

• Applio Notebook, by Vidal Kaggle
• Applio Notebook, by Shirou Kaggle
• Music Source Separation, by Shirou Kaggle
• UVR5 NO UI, by Eddy Kaggle
• Original W-Okada's Voice Changer, Kaggle
• Modified W-Okada's Voice Changer, Kaggle
• 🆕 UVR5 UI, by Eddy, ArisDev & Nick088 Kaggle
• 🆕 RVC AI Cover Maker UI, by Shirou & ArisDev Kaggle
• 📖 How to use RVC Mainline on Kaggle by Cauthess

Note: Kaggle limits GPU usage to 30 hours per week.

More commands: -colabs, -hf, -local, -rvc, /help

[brittle wing]

-colab

[karmic olive]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• Hina's Mod 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
• Hina's Modified Original W-Okada's Realtime Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab
• 🆕 Music Source Separation Training (Inference), by Jarredou & Makidanye 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

[sturdy oxide]

Hi, what are the ideal models for real-time Russian female voices? Many models crackle for some reason. What should I look at—epochs, dataset length, or something else?

[low shard]

wrong channel

RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-reocrded audio (ai covers) and train (make) models

Wokada = uses RVC for realtime inference

I will ping in #🔍│help-w-okada

[pastel oak]

LMAO i just told him to ask in here 😭

Cause of epochs, dataset etc. questions

[low shard]

Smh

Poor guy 😭

[unique rock]

Installing requirements...
error: File not found: `requirements.txt

Help, I get this error when installing applio UI

[void ravine]

guys does anyone have a google colab generator mine's not working anymore

[knotty moth]

chatgpt could generate some colab for you

[void ravine]

really? i'll try that

[low shard]

what guide link are u following and what's ur pc gpu

[low shard]

it's just writing code

[unique rock]

I use colab, and I only loaded the backup to start the training since in the other account I was disconnected and I can't connect anymore, that installed well, I connected to drive and uploaded the backup and loaded it, but when instantakar the dependencies it gave me that error, and in start applio it said, not found directory files (error 2) something like that

[low shard]

be sure to start it it from the start, and use the same settings and load backup

[unique rock]

I did it in this order:
mount Drive Load backup, and start backup, then install and start

[unique rock]

Can you help me download pre trains faster? I upload them manually but it takes a long time

[low shard]

you need to first do install, then mount drive, then load back up, then start

[low shard]

like custom ones that aren't one of the applio custom pretrains?

bc there are already some custom pretrains u can download in applio

[unique rock]

No, I have some from KLM from this year that I want to use.

[low shard]

KLM 5?

[unique rock]

No, one called KLM (Kpop AI Universe) just came out, by the same creator, but manually it is very slow.

[low shard]

I think the only ways are either u upload them or modify here the code so that it will use google servers to download directly on the machine

[true heron]

hello guys i have a question is it good if i use ai rvc if i want to match between a certain vocalist to confirm if she's featured in a song? i have the files of the supposed vocalist in flac 24bit 96kHZ and the vocals in the hook in 256kbps

i have 25 songs of the supposed vocalist

[unique rock]

So how do I put it?

[low shard]

wdym

[low shard]

modifying the code adding the download links as displayed

[true heron]

i have a chorus of an unknown female vocalist that i want to identify who it is

right

and i have the name of the supposed vocalist and i have 25 songs of the supposed vocalist in 24bit 9khz quality

the chorus is in 256kbps quality

[low shard]

idk if @nocturne mural maybe knows an easier way to add custom pretrains in applio

[unique rock]

to increase the links and put their names as the others are?

[low shard]

also, be sure that that pretrain is hifigan, if it's refine gan, you will have to use the main branch applio instead of the 3.2.8-bugfix

[low shard]

you want rvc to know who sang that song ?

[true heron]

should i put all of the vocal files in 1 giant .wav file?

[true heron]

yes

i want to see if the vocals match or not

its for a search of unidentified media

[low shard]

yeah

[low shard]

rvc doesn't have a database that can identify vocals

it doesn't work like that

rvc can do inference (use models) on pre-recorded audios, and make you train (make) your own models

[true heron]

i can't use rvc to sing over an existing audio?

[low shard]

but if u just give it a random audio, it can't tell you who was the person talking in the audio

[unique rock]

Thanks, if you could give that option in the Download section to download them from huggingface.co it would be great and easier!

[low shard]

I don't maintain that colab, you would have to ask vidal or someone in in iahispano

[unique rock]

Thanks

[low shard]

you said you wanted RVC to identify who was in that song

not to make an ai cover

[low shard]

you're welcome

[true heron]

i misworded it

[knotty moth]

you mean chorus is more than one ppl or harmonies singing together?

[true heron]

no its just 1 person

i want to hear by ear if it matches or not

sorry english is not my first language @low shard

[low shard]

it's fine, I just don't understand your request of help

[true heron]

do you still not understand it?

[knotty moth]

I heard some paper about speaker identification almost a decade ago, but would like to see if you find some decent ones

[true heron]

do i need to compile everything into 1 giant audio file? the supposed female vocalist's vocals?

can i ping?

i'll wait for other help then

[eager bloom]

what are pitch guys ? does it effect the result of the cover ?

[low shard]

if what matches to what

[true heron]

to the 256kbps vocals

[low shard]

yes, it's the tone of the voice

higher = more feminine, higher
lower = more masculine, deeper

[true heron]

and to the rvc vocals

if it sounds the same basically

[low shard]

so, you want rvc to match and compare those 2 different audio files ?

[true heron]

yes

[true heron]

@low shard (sorry for ping it's just that i asked this question twice already)

[low shard]

sorry, but RVC can't compare 2 files to see how they match

[fast scarab]

Hey everyone! What’s the best way to create a voice? Any recommendations?

[low shard]

what's ur pc gpu

[fast scarab]

3070

[compact quest]

Can i make high quality ai covers and which rvc gui should i use, help would be appreciated, here's the specs

The RAM is 3200 Mhz

And my gpu is gtx 1080 8 GB

[low shard]

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

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

[low shard]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• 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 daily, not granted, of GPU

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

the quality of the ai cover will mostly depend by how good is the model you're using though, and to check that you can only listen to samples or test it yourself

[compact quest]

@low shard this one i was looking for

[low shard]

yeah that one is old asf

don't use it at all

[compact quest]

alright

[low shard]

there's better software now

[low shard]

Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• 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 daily, not granted, of GPU

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

I would personally suggest Applio

[compact quest]

i will go for that then

thank you

[low shard]

you're welcome, let me know

[compact quest]

Does apolio really needs to be in C:/

Can't i just put it in my 1 TB hard drive.

[low shard]

it can have issues when used in other external drives

[compact quest]

But it would take all of my storage then, i only have really small amount of storage on my ssd.

Any alternatives that i can put on my hdd?

[brittle wing]

Does Applio support RefineGAN

[low shard]

U can try mainline but it's less user friendly and less features

[low shard]

Only the main branch

[compact quest]

i will just download it, i added some space.

is this the right one?

from here

https://github.com/IAHispano/Applio/releases/tag/3.2.8-bugfix

[brittle wing]

Okay but is it possible to inference refinegan models on the applio colab

[low shard]

Only if you modify the cell code to use the main branch

[brittle wing]

Okay how do I do that?Is it also for training?

[simple ore]

if you're using Applio UI Colab, you can change the cell that clones the repo and pick the main branch instead

[low shard]

yes, you just need to remove the blue part

[brittle wing]

Do i have to replace something?

[brittle wing]

Do I have to replace something?

[simple ore]

other than the code above, no

[compact quest]

I need help

i downloaded applio and it's like that

@low shard

Nevermind

i got it to work

[low shard]

oh I didn't see the msg, yes

[low shard]

my bad was eating

[compact quest]

All good

[low shard]

only remove the part I told you

[brittle wing]

I did once but sometimes it fails at dataset preprocessing

[low shard]

show how's ur dataset and tell what u used to make it

[brittle wing]

Why what

Does it matter

[low shard]

ofcourse, I need to check you used the right file formats and how you made it

[brittle wing]

https://drive.google.com/file/d/169rve8O2skU2ODvIz96b4Jim8ge7omsX/view?usp=drivesdk

How I made it? wdym

I used audio separation models

[low shard]

what did you put in the dataset path?

[brittle wing]

The drive folder path why it worked when I didn't clone any repository

[low shard]

it works for me

be sure you put the dataset path without the .zip

[brittle wing]

𝖨 𝗎𝗌𝖾 𝖺𝗉𝗉𝗅𝗂𝗈 𝗇𝗈 𝗎𝗂 & 𝖨 𝗎𝗉𝗅𝗈𝖺𝖽 𝗍𝗁𝖾 𝖺𝗎𝖽𝗂𝗈𝗌 𝗂𝗇 𝖺 𝖿𝗈𝗅𝖽𝖾𝗋 & 𝗂𝗇𝖿𝖾𝗋𝖾𝗇𝖼𝗂𝗇𝗀 𝗂𝗌 𝗐𝗁𝖺𝗍 𝗂𝗆 𝖺𝗌𝗄𝗂𝗇𝗀 𝖺𝖻𝗈𝗎𝗍 𝗇𝗈𝗐

[low shard]

so, you're having that issue still on applio no ui?

or is that issue fixed and you're asking about another one ?

[brittle wing]

𝗂𝗆 𝖺𝗌𝗄𝗂𝗇𝗀 𝗂𝖿 𝗆𝗈𝖽𝖾𝗅𝗌 𝗍𝗋𝖺𝗂𝗇𝖾𝖽 𝗐 𝗋𝖾𝖿𝗂𝗇𝖾𝗀𝖺𝗇 𝗉𝗋𝖾𝗍𝗋𝖺𝗂𝗇𝗌 𝖺𝗋𝖾 𝖼𝗈𝗆𝗉𝖺𝗍𝗂𝖻𝗅𝖾 𝗐𝗂𝗍𝗁 𝖺𝗉𝗉𝗅𝗂𝗈 𝖼𝗈𝗅𝖺𝖻, 𝗂𝗇𝖿𝖾𝗋𝖾𝗇𝖼𝖾

𝖨𝗇𝖿𝖾𝗋𝖾𝗇𝗓𝖺

𝖢𝖺𝗉𝗂𝗌𝖼

𝗌𝗍𝗈 𝖼𝗁𝗂𝖾𝖽𝖾𝗇𝖽𝗈

[low shard]

models trained with refinegan are compatible only with Applio Main Branch

[brittle wing]

𝗁𝗆𝗆 𝗌𝗈 𝖨 𝖼𝖺𝗇𝗍 𝗂𝗇𝖿𝖾𝗋𝖾𝗇𝖼𝖾?

[low shard]

you can only if you remove that part of code i told you earlier, because doing so, will make so you will use the applio main branch

also is that training issue dataset path fixed now that I told you to remove the .zip?

[brittle wing]

𝗂𝗆 𝗇𝗈𝗍 𝗍𝗋𝖺𝗂𝗇𝗂𝗇𝗀 𝖺𝗇𝗒𝗍𝗁𝗂𝗇𝗀 𝗋𝗇 𝖻𝗎𝗍 𝗂 𝗅𝗅 𝖼𝗈𝗆𝖾 𝖻𝖺𝖼𝗄 𝗍𝗈 𝗍𝗋𝖺𝗂𝗇𝗂𝗇𝗀 𝗌𝗈𝗈𝗇

[low shard]

alright

[neat lintel]

@low shard i'm wondering if you or anyone who knows could help me. So i started making some ai songs and they are okay, but i came across this guy and in his songs ai voices are sounding amazing. My question is what kind of software is he using that is making Arthur and others sound so good, they're as close to in game version as you can get. Are they using some RVC that i'm not aware about, are they training their own models or is there a different reason. I'd appreciate any insight that would help. Here is the vid i'm referring to: https://youtu.be/2oE6plJb7f4?si=ALKuaEQN-L8gglWj

[low shard]

It mostly depends on how good the model has been trained

[neat lintel]

That was my initial thought, cause every model i tried wasn't it, i mean it was okay but nothing insanely good, so it could be the model then

Currently i'm using one that is 700 epochs, but i think the data that the model was trained with is more important no?

[low shard]

Epochs don't mean more quality

The dataset length, and mostly important the quality, matters

[neat lintel]

Yeah well thanks a lot. Any idea where i might find some good models, i tried one on weights and it is okay.

[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

[neat lintel]

Thanks a lot for help!

[low shard]

Yw

[neat lintel]

Found this on hugging face, there are 2 .pth files but no index one, what am i supposed to download beside this 2 .pth files

[low shard]

Seems like 2 different epochs of the same model?

Maybe try the 1st

[neat lintel]

I'm downloading both just in case, will try them out, listened to example audio files and they sound extremely good, well that's all from me, thanks for your time, much appreciated.

[fast scarab]

Hey everyone! For those using Weights.gg, do you know how many epochs are used for training vocal models? I couldn’t find the info on their site. Thanks in advance!

[cyan elbow]

Hello, I need some help. Is Applio or Mainline better for AI covers? Which one is better supported by my graphics card (2070 Super)?

[low shard]

They both support your GPU, they both are good, I would suggest Applio for it's more updates and user friendliness

[marble forge]

im trying to train locally but the training isnt showing up in the scalars on tensorboard

[low lark]

Hi guys, I want to have my own voice for the TTS model but I want it to be in Indonesian language, I have dataset and RVC ready, have 8GB RTX 4070 VRAM, can I just train without doing any setup using my language?

[simple ore]

you need tts that can do indonesian language

[low lark]

Applio can do that?

[simple ore]

applio is for voice conversion

there's built in TTS (MIcrosoft Edge screen reader)

it may have indonesian languge

Text -> Microsoft Edge Screen Reader (TTS) -> tts output.wav -> Voice Changer -> output.wav

[low lark]

Thanks for this, I will try it

[brittle wing]

When I download the ai I don't get the voice of a girl or anything, I use amd and I'm a laptop who knows why

[hallow thistle]

!howtoask

[patent trellis]

How To Troubleshoot

__**GIVE CONTEXT.**__ 📝

• Don't simply mention your issue, like "my rvc is not working".
• Describe the step you are on, what you're trying to do, the RVC you're using, a screenshot, etc.
• The more context, the better.

__**BE POLITE.**__ <:matsuripray:1159685390156967936>

• Don't be desperate. You can ping a Helper, but if they ignore, they aren't available/don't know the answer.
• It's okay if you're frustrated, but don't take it into this server.
• Don't DM without prior consent.

__**BE PRODUCTIVE.**__ 🤝

• Don't ask for every little instruction. Put your own effort & test things by yourself.
• Don't ask to ask.
• Check if your answer is a Google search away/on our guides website.

If you've changed your mind about your message, feel free to delete/edit it.

[wind harbor]

Goood day everyone. Please, i do need help connecting my RVC software to realtime on any socail platform.

[hallow thistle]

Not a good day for me, but sure that's my job teaching people how to install RVC/W-Okada, I guess.

If you meant by an RVC program that can convert audio in realtime, you might be looking for W-Okada.

[brittle wing]

My problem is that I download the AI and put my laptop microphone and I'm from AMD and no one listens to me in a deep voice oh girl just with my voxz and I need a help someone who knows that can help me good night

[hallow thistle]

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

If you meant by a realtime mode that found in RVC GUI, this one is too old now, and never use this.

[wind harbor]

I have W-Okada installed on my pc alread and i've been using it to change my voice in realtime over some months now but i haven't been able to set-up a connection for it to work on any social platform. Please do help and i would ruly appreciate.Thank you

[hallow thistle]

Let me tell y'all again, for W-Okada, let's move to #🔍│help-w-okada. #✨│ai-help is all about RVC and any related RVC program.

[wind harbor]

ok lemme go there now...

[low lark]

Hi sir, I'm sorry for jumping in, do you know any RVC that can support Indonesian language?

Ah nvm, I think I will try with RVC base V2, found some Indonesian create some model using it

[hallow thistle]

https://cdn.discordapp.com/emojis/1015884927226613800.webp?size=48

You know it's bad when someone asking me like if I know everything.

[low lark]

Hehe I'm so sorry 🙇‍♂️ I'm just gonna try everything first

[hallow thistle]

I don't know if there's any Indonesian model that can be used for RVC TTS on Applio, but this is all I can find.

There are three English-Indonesian TTS models available on Applio, and that's pretty much it.

Oh, ok, I got it. The language code for Indonesian is id.

[simple ore]

Ardi Neutral and Gadis Neutral are indonesian voices

the are few more multi-lingual

https://speech.microsoft.com/portal/voicegallery

not all of them are available for Edge TTS

[low lark]

aaah thanks so much @hallow thistle & @simple ore , I really appreciate it !

[knotty moth]

not only that there are also sundanese and javanese models

[fathom tulip]

does anyone know how to reduce reverb while making a cover?

my output vocal had a bit reverb when im set the pitch to low point

[true heron]

well shit my model didn't work well

it now sounds robotic as well

unlucky

i used weights.gg

i guess its not good

[hallow thistle]

W-Okada or RVC?

[true heron]

rvc

i think i know why

but i give up already

if you want to can help me by making for me a model but i doubt anyone would

[edgy tangle]

I think this looks a lot better

[compact quest]

i tried both mirrors

that isn't my internet speed

there no way that's the speed

[low shard]

wtf is wrong with ur wifi

[compact quest]

it's not my wifi

[knotty moth]

try using vpn

[compact quest]

is there any other mirror

[low shard]

I think VPN would slow it tho, even if it could be true that he could be away from where ever is the server

[low shard]

#✦│chat message tbf with such a old gpu I would suggest Cloud

I'm not sure how good it will be running the best models with that gpu locally

[knotty moth]

not always so

[low shard]

I mean depends on the VPN he uses, and most free VPNs aren't that good

[low shard]

yeah the gtx 1080 is kinda old, so that's why I'm saying I'm not much sure how much time will it take to use the latest models

It runs for sure, just saying idk how good the speed will be

[compact quest]

i can't find the cloud one

[low shard]

Cloud would be faster, tho your choice

[low shard]

everything at the right of local

[knotty moth]

it can still run UVR, although may not be faster than 2060/2070

[low shard]

yeah that's for sure

[cyan elbow]

Tysm!

[low shard]

yw

Is that spinel in ur pfp

I remember when steven universe was still running

[compact quest]

I don't see

[low shard]

go above

[cerulean hedge]

best free rvc converter?

[vague glade]

how can i make local applio link public?
i want to let my friend do stuff while my pc runs

[low shard]

what's ur pc gpu

what do u want to do

[low shard]

edit the run-applio.bat and add '--share'

[fast scarab]

Yo everyone,
I’m having an issue with MMVCServerSIO when trying to import a voice model I created. Here’s the error I’m getting:
File "voice_changer/RVC/RVCModelSlotGenerator.py", line 42, in load_model
slotInfo = cls._setInfoByPytorch(modelPath, slotInfo)
File "voice_changer/RVC/RVCModelSlotGenerator.py", line 58, in _setInfoByPytorch
config_len = len(cpt["config"])
^^^^^^^^^^
KeyError: 'config'
I’m on Windows 11 with an RTX 3080, using the latest version of the software. My .pth file is there, but it seems like something is missing in the config.
Here’s what I’ve tried so far:
✅ Checked if a config.json is required (not sure if it’s mandatory or auto-generated)
✅ Tried recreating the config, but not sure if I did it correctly
✅ Tested another model, and that one works fine
If anyone knows where the issue is coming from and how to fix it, I’d really appreciate the help. Thanks in advance! 🙏

[pastel oak]

Wokada/MMVCServerSIO related issues in #🔍│help-w-okada

Are you using the alpha/beta version of wokada?

[small vector]

why the quality of the voice on this app is much worse then when using the applio web app?

I already added all the filters that appear on the front end but it's still much worse then what I get on applio web

I can't send files '-'

[pastel oak]

!give-media-perms 30m @small vector

send now

[fast scarab]

Hey! Thanks for the reply. I don’t think it’s related to the version of Wokada. Other voice models work fine, but the one I created is throwing this ‘KeyError: config’.

[pastel oak]

Wokada and MMVCServerSIO are the same thing

its just a label thing

Nvm i just understood what you meant, you were not questioning that

[small vector]

thanks, this one

[pastel oak]

I am still asking because we are currently recommending the fork wokada instead of the vcclient since it currently runs better. So was checking in if you have the fork or not

[small vector]

I think the missing filter is the one that forces the voice to sound more like the RVC one, using 1 as value

[fast scarab]

I’m using the forked version of Wokada. The issue only happens with my custom model, not with other models. Do you think it’s a formatting issue with my config.json or .pth file?

[small vector]

idk what's the code name for this filter on python, do you know @pastel oak ?

[pastel oak]

With what software did you train your custom model?

Mainline, Applio, Codenames Fork, etc.

[pastel oak]

I dont know what app/software youre using

[small vector]

applio

[pastel oak]

And youre saying running applio local sounds worse than applio colab?

[small vector]

yeah, I developed an automation but the same voice sounds better on applio web version

[fast scarab]

I trained my model using RVC. Here is a screenshot of my model files

[pastel oak]

I dont think I can help, sorry. Wait for someone else or ask in their discord
https://discord.com/invite/urxFjYmYYh

[pastel oak]

That is not a voice model. You uploaded the (D)iscriminator that people would need to use if they want to continue training a model, alongside the G

[small vector]

ok

[pastel oak]

Check your weights folder for the voice model, should be name <YOUR MODEL NAME>_eXXX_sXXX.pth , probably

[brittle wing]

@crude bolt no effect plsss

@crude bolt no effect plsss

[fast scarab]

I checked my files, and here are the ones generated during my model training.
I want to make sure I’m using the right file for MMVCServerSIO. I see multiple .pth files, but I’m not sure which one is the correct voice model.

Can you tell me which file format or naming convention I should look for? I trained my model using RVC. Thanks for the help!

[pastel oak]

If you are using Applio, its in the same folder, like this in the first screenshot

If you are using Mainline, you are in the wrong folder. Go back to the start of the folder where you also find the start to the program. Then go to assets, weights and look for the name in there (2nd screenshot)

[pastel oak]

And btw you can delete all of those D_ files and G files up until the most recent file (the one with highest number) to save space

If you ever plan on continuing to train the model

cause theyre 837mb for just one file 😭

[small vector]

can someone recommend some good pydub/pytorch post processing filters to increase VC quality?

[fast scarab]

Thanks, it works! I was in the wrong folder actually.

[fair ivy]

someoone help

when i import a model like gojo it tells me that i cant upload files of that type

@pastel oak rlly sorry but plsmhelp

[pastel oak]

What program are you using and whats the gojo file youre trying to upload, send screenshots or something

[simple ore]

looking in the wrong place

[pastel oak]

!give-media-perms 30m @fair ivy

[silk vapor]

How do I use https://discord.com/channels/1159260121998827560/1255206706598772796 ? Which one do I select?

RVC?

[simple ore]

you download D and G files and use them as custom pretrain weights

[fair ivy]

Im using voice ai

[fair ivy]

My laptop is dead so my tomorrow

[stiff goblet]

Hi guys. Someone said the loss/g/total lowest point is not true in Tenserboard. He said avarege/loss/g is the right graphic to examine. But in the Colab version of applio, i couldn't find any "avarage" related graphic.

[pastel oak]

Can you tell me who this "someone" is

[pastel oak]

we dont offer support for voice.ai on this server

[simple ore]

those have not been released yet

[fair ivy]

Ok then what app?

[pastel oak]

i can give you link, can you first tell me what gpu you have?

[stiff goblet]

Is it kind of experimental then ? Can I stick with the old graphics if it isn't differing too much?

[stiff goblet]

I guess @simple ore, already replied to me

[fair ivy]

Intel graphics

[pastel oak]

thats not good enough to run the voice changer locally, so you can do colab/kaggle which is online based. its all for free

For Realtime Voice Changing for Calls on Cloud (remote good pc for those who don't have a good one, YOU CANT DO THIS ON MOBILE):

• Google Colabs (4 hours daily of free T4 gpu, easy to use, require only a google account) :
  • How to use Hina's Modified Original W-Okada's Realtime Voice Changer Google Colab (has a Guide)
  • W-Okada's Deiteris Fork Realtime Realtime Voice Changer Google Colab (no guide but explained in the cells)
• Kaggles (30 hours weekly of better GPUs, T4x2 & P100, harder to use, requires an account and a phone number)
  • Hina's Modified Original W-Okada's Realtime Voice Changer Kaggle (no guide but explained in the cells)
  • W-Okada's Deiteris Fork Realtime Voice Changer Kaggle (no guide but explained in the cells)

[simple ore]

old total charts provide a bad picture. they may be useable if your epochs are not too long

[stiff goblet]

49 step for each epoch, 16 minute dataset. Did you mean the total epoch count until overtraining?

[simple ore]

i mean 49s/e is small enough

total loss values are from a random batch from the epoch, but they are short enough, so the smoothed charts are somewhat okay

[stiff goblet]

thank you very much

[brittle wing]

-colab

[karmic olive]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• Hina's Mod 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
• Hina's Modified Original W-Okada's Realtime Voice Changer, Google Colab
• FaceFusion UI, by Nick088 Google Colab
• FaceFusion NO UI, by Nick088 Google Colab
• EasyGUI, by Rejekts Google Colab
• 🆕 Music Source Separation Training (Inference), by Jarredou & Makidanye 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

[formal wind]

Does anyone know what file the index rate settings for the Real time voice changer client would be? Thanks!

[edgy tangle]

is this overtraining?

[brittle wing]

Hello everyone, I recently encountered a problem, I used this neural network about 3 months ago and now I downloaded it again to my PC, nothing has changed, but at the same time my voice lags and resembles a robot (on all models), the delay and so on do not help at all.

[analog obsidian]

rvc is not wokada
use the voice changer help channel 👉 #🔍│help-w-okada