✨│ai-help

[brittle wing]

I'm used to it for training

[crude flame]

It's just a different graph

[brittle wing]

Id it on this colab or only on codenames fork?

[crude flame]

It's only codenames fork and the newest build of applio

[brittle wing]

It's getting worse

[brittle wing]

Yes I understand but

[crude flame]

Down = good

[brittle wing]

Huh

[crude flame]

The graph is going down so that means it's still fine to be trained

[brittle wing]

It's still not overtrained?Why is the line going like this

[crude flame]

Yes it's not overtrained

Graph going down = not overtraining

[brittle wing]

I should resume and keep training?

It trains very fast due to the short dataset

[craggy brook]

...?

[brittle wing]

Started going up after 300!

[crude flame]

To make sure it's ot-ing leave it to train for an extra hour or so

[brittle wing]

Nah it gone up after 350

It made like a "V"

Keeps overtraining I'm done w this one, next model now.

[bold yarrow]

i have no idea how any of this stuff works, i might pay someone to make models for me lol

[tame mica]

#✨│ai-help message

^ good resource to keep

then again do whatever with your models eksdee

[bold yarrow]

hmmmm

im not motivated to start

even tho i wanna create a model

i am so internally conflicted as to whether i want to create a model

cos i wanna make it the cleanest as humanly possible via rvc 2

[ivory holly]

i literally don’t know how to use ai. i have a model i wanna use that i have downloaded, how do i use it?

[tame mica]

-doc

[azure marsh]

Suggestions for @ivory holly

📚 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

[tame mica]

go to ai hub docs n click the "How to make AI Covers" at your left

[ivory holly]

thank you

[uncut sedge]

anyone can help with the rvc on how to use on discord?\

@tame mica

[tame mica]

-realtime

[azure marsh]

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

💻 Local Realtime RVC

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

[tame mica]

go to #🔍│help-w-okada

i dont use realtime

[uncut sedge]

y

is it shit?

[tame mica]

no i just personally dont have any reason to use it

i just like making covers instead :p

[uncut sedge]

ok but how i use in vc?

aka games

[tame mica]

click on the guide

the one with (fork)

[uncut sedge]

which one tho

ok

i installed

the virtual cable thing

but now its not picking up audio

how to fix?

@tame mica

[tame mica]

as i said i dont use realtime so i cant help you

[uncut sedge]

i think i got w-okada

[tame mica]

ye

go to #🔍│help-w-okada

[uncut sedge]

im here

[hallow thistle]

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

[tiny totem]

Annoying but understandable (kinda...) that rvc has a limit for audios ≤0.78s in length. Is there some kind of preference to increasing leanth of an audio or is it just sinply spamming cntrl+v until it's long enough. Maybe variation in pitch? Time stretching it?

I'd rather keep the integrity if the original voice rather than manipulate it too much.

What would happen if I trained I trained a model with just really short audios anyway? I get it's (afaik) not enough to be able to generalize well and Ik TT2 models need variable length audios to be able to make any output length decent

[dusk venture]

Can someone help me please? I wanted to make ai model but I don't know how to do it and I only have a phone

[azure marsh]

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

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

[dusk venture]

I already have a datatest but I don't know how to do it, please help

[tame mica]

-docs

[azure marsh]

Suggestions for @dusk venture

📚 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

[tame mica]

look at "how to make voice models"

[lavish lintel]

Congratulations kar !!!!

Your Samurott is now level 51!

New move!

Your Samurott can now learn Retaliate!

‎

‎

‎

‎

[dusk venture]

Where?

[tame mica]

ai hub docs

[steel forge]

Try it out

I would imagine messing with the voice will make the model really weird. But I do remember an OG Minecraft villager model performing really well with that method of pitching the sample

Idk the best way to get around the .78s limit... Maybe just loop your sample, with minimal silence

[brittle wing]

Is this overtraining?

[simple ore]

imagine being here and asking "is it overtraining"

and then again here

or here

[brittle wing]

It wasn't it started going up after250epochs

[simple ore]

while it is still going on like this

and even worse, using the old charts for that

[modest otter]

do you need a lot of resources to run rvc?

[hallow thistle]

Most recent RVC programs don't take a lot of resource to run. They are developed to run better. Unless you got the older version of RVC which indeed takes a lot of resource and performance.

[worldly tundra]

guys is there any way to have something like auto pitch selection for SVC in RVC?

[hallow thistle]

"SVC" voice model cannot be used in RVC.

[worldly tundra]

Sure. I am asking something else.

SVC has an option called "Auto f0 prediction"
Which detects the pitch number for voice conversions automatically

My question is that is there any similar solution for RVC?

Looks like other people are asking about this too

https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/734

[hallow thistle]

GPT-SoVITS or "so-vits-svc" from 2023?

-gui

[azure marsh]

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

[worldly tundra]

so-vits-svc

[hallow thistle]

Nahhh, this voice conversion program is so old. I can't help with that.

[simple ore]

this is one weird method.. it trains f0 predictor... why, i have no idea.. you can just adjust the pitch of the audio you're trying to convert

someone asked for the same thing for Applio

but I'm like what's the fking point

[worldly tundra]

Yup. I try to convert with different pitch values in order to get the right one for my voice

[simple ore]

rmvpe is 'f0 predictor'

[worldly tundra]

But i'm trying to build a product on rvc so other people can convert their voices and users can really understand the process of finding the right pitch value for their voice

[simple ore]

so you just run f0 extraction on the dataset, find min/max/std/mean values that the model does

then you take the audio you're trying to convert, see what it has, adjust

[worldly tundra]

it just got too complicated for me

[hallow thistle]

Running the original RVC GUI in the year 2025 isn't recommended.

[worldly tundra]

any workflow or repository i can use for this?

[hallow thistle]

These audio files are funny.

[simple ore]

original is breaking because the pitch of the audio is waaay too high

but after adjustment the pitch is within range of the model so it does fine, no voice breaking

[worldly tundra]

you try different pitch numbers in order to get the right one for the voice, right?

[knotty moth]

-svc

[hallow thistle]

Yes.

[simple ore]

yeah, I did

[azure marsh]

@worldly tundra Nah, use RVC instead

[hallow thistle]

[simple ore]

I've only half-baked the idea for Applio

[worldly tundra]

alright
but i dont wanna try different numbers
i wanna do similar to that auto f0 from svc
any suggestions?

[simple ore]

i dont have any interest in digging into svc code

[timid olive]

Which model can I use for better training for conversations?

[hallow thistle]

You mean the pretained model?

[timid olive]

I will use it primarily for speech.

[timid olive]

no

[worldly tundra]

i get it
if i can understand the idea it would be great
people on rvc repositories are asking for this feature

[hallow thistle]

Wait, what?

[simple ore]

as I said, run rmvpe on the model dataset to get f0 std and mean values

then adjust audio f0 using thse values, no need to train a f0 predictor

[knotty moth]

there's no point of compatibility to such old svc, and even rvc v1 support is also deprecated
because rvc v2 models are always better

[worldly tundra]

do i have to run rmvpe on all of dataset i used to train the model?

[simple ore]

well, you can just run it on the f0 files rvc created

[hallow thistle]

On Applio, this is where you set the pitch to inference an audio.

[simple ore]

that is not as good as the goofy method I did

but okay

can use that too

there's also formant option as well

[knotty moth]

btw in the latest applio/codename fork, lowering envelope somehow causes popping/artifacts

[worldly tundra]

]this is logs folder for one of the models i trained
where can i find that values

[simple ore]

2a and 2b folders.. one contains float f0 values in .npy other contains coarse

you need to read those float npys, discard zeros, then just do calculate std and mean values

[sturdy cosmos]

i alrd upload some voice model but why mine doesnt working?

[hallow thistle]

I think you're looking for W-Okada. I saw you in #🔍│help-w-okada.

[sturdy cosmos]

what the diff okada and rvc?

[simple ore]

okada is realtime voice changer, rvc is not

[timid olive]

I will train TTS voices and use them in speech texts.

I will use dialogues from sample movie scenes, not so much for songs.

[knotty moth]

RVC (STS) or gpt sovits (TTS)

[brittle wing]

I actually know how and I wouldn't pay someone, I usually wait for someone to make a model

[sturdy cosmos]

okay thanks

[brittle wing]

It gone like this after 250 epochs

[hallow thistle]

W-Okada is the "realtime" voice changer. RVC is the audio conversion program.

An RVC program can train a voice model. W-Okada uses RVC voice model to inference, but the GUI and codes majority aren't related to RVC.

[timid olive]

There is no support for Turkish or other languages, and I will clone the voice.

I mean the method, but if there is a better tool, that could work too. My goal is to replace the emotional speech in the movie with another voice, to clone it.

Of course, the language is mixed, Turkish or different languages.

I collected a lot of datasets from ElevenLabs for training, for voice cloning training.

[knotty moth]

if not sure, use RVC, though there's no turkish pretrain I've known so far

[timid olive]

which of these is the best in speech

[hallow thistle]

English.

[timid olive]

how to do pre-training

[knotty moth]

ah, I mean 11labs is still the best as base TTS to convert with an rvc model

[timid olive]

f0_method

[knotty moth]

rmvpe for most cases, or crepe hop length 64 with applio for better quality dataset

[timid olive]

rmvpe songs are good but I don't know which one is good at speaking

[timid olive]

thank you is there a preliminary training model training video

[knotty moth]

-guides

[azure marsh]

Suggestions for @timid 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

[timid olive]

crepe-tiny or crepe ???

[knotty moth]

tiny is just lower quality

[timid olive]

How many times should I do this?

[knotty moth]

let the embedder contentvec which is compatible to original rvc
refer to the guide for the rest and good luck

[formal wind]

Hey Im switching from rvc disconnected to local training and litsa said It works with amd. But the training method and running the steps are super duper slow and Its unbearable. It took like 30 mins just to extract the features. My graphics card is an AMD Radeon RX 6600

[simple ore]

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

[formal wind]

I used that. I thought I did everything right. is there a vid on it. Im slow when it comes to words that are infront of me

[simple ore]

1st inference may be slow because it has to compile stuff

as long as you did everything right, the GPU should be shown under training/advanced settings

[shut nimbus]

why can help me in promlem RVC

[brittle wing]

What colab is this?

[low shard]

!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.

[brittle wing]

Can someone delete some of my old model posts, also how do I remove them from weights.gg???

[hallow thistle]

Contact a moderator/admin here, and they'll fix them.

[knotty moth]

why not delete it by yourself?

[hallow thistle]

[brittle wing]

I can't

[brittle wing]

Who is a moderator

I can't delete them cause they're old models that got synced w weights

[hallow thistle]

Even if you deleted the older and outdated voice models from here, the voice models somehow would still be there on Weights.

[brittle wing]

You don't understand they're posts from my deleted account

[brittle wing]

How do I remove them from weights too

[hallow thistle]

For that Weights part, you can contact a moderator or admin here who's associated with Weights.

[knotty moth]

nvm

[brittle wing]

Is it possible

[hallow thistle]

I don't remember which mod is associated with Weights, but there are Bea and Vijel.

[brittle wing]

I know this ALREADY.I mean something else...

I mean my old models that got backed up by weights, not ones I uploaded now

My two old models.

[knotty moth]

I mean there's no way other than to gain access to ur old account

[brittle wing]

Old account yeah but can someone else remove the posts.

Is what I mean

[knotty moth]

try contact the weights staff

[brittle wing]

How

On the site?

[hallow thistle]

Some are here in this server.

[knotty moth]

I mean if u have deleted the old linked discord account, u could try contact the staff to recover access to the weights account

[brittle wing]

You don't understand.

[hallow thistle]

Try contact Bea or Vijel in their DM if you wanna to delete your old voice model from the site.

[knotty moth]

I don't?

[hallow thistle]

I always understand you, so please stop saying we don't understand.

[knotty moth]

^

[hallow thistle]

[brittle wing]

Well the first time I posted models here in 2022 they got backed up on weights automatically after the og AI HUB server got taken down

[hallow thistle]

Is there anything I said it wrong?

[brittle wing]

But I didn't have an account on the site back then I didn't register

[brittle wing]

Nothing wrong

[hallow thistle]

https://cdn.discordapp.com/emojis/1015964749470650400.webp?size=48

[brittle wing]

Now I uploaded other models with MY weights account

The old ones are overtrained and outdated and nosiy

But I can't access and delete them

They weren't uploaded by me, they were synced w the og server.

[hallow thistle]

Have you read my eariler message? There are some people here who work for Weights.

[brittle wing]

Before it got taken down.

[brittle wing]

Uhm ping me one.

[knotty moth]

once again, it doesn't belong to ur current weights account right?

[brittle wing]

Absolutely no, if it was, I wouldn't be asking.

Can someone tell me an admin or moderator?

Cause I asked a moderator on the k-pop ai server and they deleted my old posts.

[hallow thistle]

Hello @jade cosmos and @oak plank. If you know where to manage voice models on Weights, please help this user to delete their old and unwanted voice models from there.

[brittle wing]

THANKS

[knotty moth]

more likely to regain access to the old account in order to enable managing the models, otherwise it would be an inactive stray account that risks to be eventually hacked

[past beacon]

hey y‘all, it wont let me upload an audio file on huggingface, like i‘m adding it but nothing happens

[willow perch]

Hii. How do I continue training a model that I already trained? Im training it on kaggle

do I need to add the D_xxxx.pth/G_xxxx.pth and from there train it?

[simple ore]

D+G weights, + dataset files

[willow perch]

So basically re do everything with same name and everything but add the d+g files, right??

sorry if I dont get it, first time continue training a already trained model lol

[brittle wing]

What batch size for 1:30 minutes of data?
I chose 8 and I think it was a mistake doing so

[formal wind]

@simple ore I did all the steps again and the gpu is there. The feature extraction is normal speed but the training is like 30 mins per epoch.

[simple ore]

how big is the dataset?

with my 6700xt 45min set was taking ~4m/epoch

[formal wind]

27 minutes

Im goin insane man 😭

[simple ore]

i mean.. did you see 'Compiling..." in the log?

[formal wind]

It did it before the first epoch then 30 mins later the first epoch finished. Then compiling. Then same process

I gave it a bit to see if it just had to load but nope

[simple ore]

it should stop compiling, it only does it once per operation

[formal wind]

Idk if im doing something wrong or not

I probably did

I tend to muck up or fail when Im not watching a video tutorial 😭

[willow perch]

@simple ore Once I have d+g set up, do I create a new index file aswell or with the other one is ok?

sorry for the ping

[simple ore]

original is fine

[willow perch]

thankss

[formal wind]

So is there a fix Noob or am I just lost

[simple ore]

run again

[formal wind]

do you train with it? If so can you send a ss of the settings you use. Like aside from the epochs and batch size obviously

[simple ore]

of course I did

see task manager/performance, if it uses shared memory it will be slow af

this is fine

full dedicated + more shared = bad

[formal wind]

How tf do I check that

[formal wind]

nvm got it

erm..

Welp.

[simple ore]

looks fine

[formal wind]

Oh so what about the shared memory being 16 and dedicated being 8

does that not matter. Is it just the graph that matters

[simple ore]

just the graph and actually used memory

shared is always half of RAM

[formal wind]

Well ok so atleast thats good news. But what am I supposed to do to fix the issue. Whats re-running it supposed to do?

[simple ore]

to make sure everything is compiled and your speed is based on the compiled code

[formal wind]

But ive ran it 2 or 3 times. Whats the 3rd or 4th going to do. This hurts my brain.

Do I change this to 1 or leave it at 0

Sorry for all the questions 😭

[simple ore]

this looks fine

[formal wind]

Ill run it again and see what happens

Mannn its still doin it

[simple ore]

you got hip 5.7, right?

and replaced libraries using the optimized set?

[formal wind]

Yeah

[quartz turtle]

hi!

What is the best pitch extraction algorithm for a song with high pitches that doesn't sound robotic?

[primal galleon]

how can i download it and start it

[low shard]

what u want to download

what's ur pc gpu

[primal galleon]

rvc

[low shard]

what do u want to do

[primal galleon]

i want to put my voice or any voice to a song

[low shard]

ok so AI Cover,
RVC is the right program

what's ur pc gpu

[primal galleon]

gtx 1660 super

[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

[primal galleon]

whats the cloud and local

whats the difference

oh i see

sorry

what do u recommend

[low shard]

if ur going to only use models, do it locally with applio

if ur gonna train, use cloud like kaggle

if u need the easiest possible thing ever made to do both, use weights.gg

[primal galleon]

so if i want to train my own voice then i should use kaggle?

[low shard]

I mean u could also train locally but ur gpu could have issues, so it's better u use cloud, kaggele has the best gpu time for free

[primal galleon]

and how can i download rvc?

or what exactly is rvc

[low shard]

RVC = Retrieval-based-Voice-Conversion, speech to speech AI

[low shard]

I shared you hyperlink with guides (blue text that when clicked is a link)

[primal galleon]

where can i download it?

oh okay

what is rvc v2?

@low shard

[low shard]

it's the type of rvc version the model has been trained on

it's the latest, use it

[primal galleon]

idk which one

whats the name of it

applio?

cus i dont understand

where can i download rvc v2

the program

[low shard]

Applio is a fork (modified version) of RVC which has more features

[low shard]

u can follow the applio guide up there

[latent kettle]

Hey @low shard do you know any good TTS for English and hindi language?

[low shard]

hindi idk

for english, u could try gpt so vits or f5 tts or fish speech

u can also check our tts index https://docs.ai-hub.wtf/tts/tts-tools/

[latent kettle]

Can I use my rvc voices in this tool

??

[low shard]

@crude flame u also gotta add how to use the tts in 'realtime' btw

[low shard]

Rvc is speech to speech

[latent kettle]

Then how applio convert text to speech using rvc models

[low shard]

Technically not unless u generate a tts Audio and use it as input in rvc

[low shard]

They don't, they just generate tts with edge tts then use it as input in rvc

I think that's also said in applio

[latent kettle]

Oh I see. Is edge TTS in good for English ?

[low shard]

Edge tts is multilingual and only premade voices, using Microsoft edge tts API, not local

[latent kettle]

Do it have emotions in speaking like ElevenLabs.

[low shard]

Well 11 labs is still the best tts

Edge tts doesn't have good emotions

[latent kettle]

But it's paid

[latent kettle]

Then which tts is a good alternative for 11labs

[low shard]

The ones I said before: gpt so vits or f5 tts or fish speech

Those are good alternatives but 11labs is better

[viscid moss]

I think F5 TTS

It's good with emotions

[latent kettle]

Do it supports training models for other languages?

[low shard]

Did it get updated?

[latent kettle]

Em. Do I support?

@low shard

[primal galleon]

@low shard i uploaded the vocals and i want to replace it with mariah carey but it says error

can u help me

pls

[covert bough]

yeah now it's working a lot better

at the GPU slot tho it shows cpu and i cant change anything

i also cant edit the chunk

[viscid moss]

Hmm not sure, I tried the F5 TTS in Spanish and it already had that feature.
https://huggingface.co/spaces/jpgallegoar/Spanish-F5

[latent kettle]

@viscid moss do you know what is fine-tuning?

And how to do that ?

[viscid moss]

Basically improve an already made model

[viscid moss]

Nope

idk

[latent kettle]

Sounds crazy, how to do that

[latent kettle]

Oops, I see

Sorry

[viscid moss]

dw

[low shard]

Are u talking about Hindi specifically

[latent kettle]

Hindi and some other languages. Like Japanese or French.

[low shard]

Well I don't think any of the 3 I mentioned support Hindi as of right now, unless u fine-tune it, I never tried that but I know it takes a lotttt of time and a lotttt of power

[latent kettle]

Cam we fine tune a RVC model too ?

[crude flame]

wdym using tts in realtime? Congrats on senior mod btw

[covert bough]

what do i do to use my gpu for rvc

[low shard]

F5 supports English and Chinese,

gpt so vits supports English, Japanese, Korean, Cantonese and Chinese

Fish speech supports English, Japanese, Korean, Chinese, French, German, Arabic, and Spanish

[latent kettle]

Okay, cool.

[low shard]

Are u talking about Wokada as we did in the other channel? Show me a screenshot in #🔍│help-w-okada

[latent kettle]

Is it possible to fine tune a Rvc Model

Or it is only applicable to TTS models

[low shard]

That's what you do when you make any model using a Pretrain lol

[latent kettle]

Emm. Idk about that

What if I don't use a pre trained model

[low shard]

Pretrain: already trained model to use as a base

Fine-tune: train a model using a Pretrain, which is what everyone does when training their own RVC model

Train from scratch: train without a base, takes way more time and resources

If you ever trained an rvc model, you actually just finetuned based on the Pretrain

Unless you got like 50 hours and trained from scratch which I don't think you did

[latent kettle]

Okay. How can i train from scratch

[low shard]

That's only for when you want to make a Pretrain, it's highly not suggested to train a normal model without pretrain

What are you trying to do?

[latent kettle]

Nothing, just doing experiments. Want to learn something new

[low shard]

All you have to do is normal training but without selecting a Pretrain

[latent kettle]

I see many custome pre trains, can I create my own ??

[latent kettle]

And the model in output will be a pre trained ? Which I can use as a base for training my other models ?

[low shard]

The bottom of my guide, it explains how to use any TTS "like Wokada"

Also ty

[low shard]

Did you upload the model or just write the name of the person

[low shard]

Well technically yeah but it would take A BIG dataset and A LOT of time

I'm not sure if your GPU would be good enough

[latent kettle]

8GB VRAM ?

[low shard]

The output will be the same as when you train any other models that you trained

The only difference is when you use a Pretrain, u use the G & D files, while when u use a normal model u use the index and name.pth

[latent kettle]

Will it impact on model quality and accuracy if I Train my own pre trains in my native language

Training pre train will give me a benefit in output or not ?

[primal galleon]

i uploaded everything

is it because its rvc v2 model?

[latent kettle]

Can you please tell me which type of information is stored in a pre train.

[low shard]

send me the download model link u used

[low shard]

yeah it depends on how good u trained it, a good pretrain can help, a bad one can make it worse

[low shard]

it gets trained the exact same way as a model basically

[latent kettle]

Is it beneficial to train a pre trained of any model before training actual model

??

[low shard]

what

[latent kettle]

Emm.. lemme try to explain

[hazy grove]

I didn’t understand how to create my own model

[primal galleon]

@low shard i cant send u friend reqest

[low shard]

no need, just send the model download link here

[primal galleon]

kk

[low shard]

what's ur pc gpu

[latent kettle]

For example I'm training a model of a person named "nick" and I have collected a lot of voice data of that person, and I trained a custom pre train model of nick voice (fine tuning) and after that I'm creating the model normal model of nick.

[primal galleon]

@low shard https://www.weights.gg/models/cm32eshds00k4xn22owrk9k5w this

[low shard]

nah, it would be the same data, also u don't really need to make a pretrain for just a single voice

[low shard]

did you download it, extract the model, then upload it manually since its a weights.gg model as explained in the guide?

[latent kettle]

Okay. So I should stick with original pre trained model?

[hazy grove]

rtx 3070 ti

[low shard]

that depends on ur dataset lenght and language

[primal galleon]

ive just uploaded the index and the pth

[low shard]

could you retry and tell me what's the error that appears on top right?

[latent kettle]

Can I train a custom pre train on 8GB vram

[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

check also: https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/

[primal galleon]

it only says error

[latent kettle]

Error will be displayed here

[primal galleon]

its not

[hazy grove]

I have a file with voice models, but how do I use it?

[low shard]

I mean u can, but I'm not sure since it's considerated as D tier

[low shard]

retry the convert button, it will be there

[latent kettle]

D tier ? What does it mean

[low shard]

then that's not training (making) a model, that's inference (use models), you're looking for realtime voice changer for calls or inference on pre-recorded audios?

[low shard]

like not that good

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

[primal galleon]

it says succesfully acquired gpu but still a error

[hazy grove]

I want to make a model and then use it to record all sorts of audio recordings and so on for video

[latent kettle]

Okay. Btw we will talk about this topic later. I'm going to bed now. Thank you for assisting me. And congratulations for senior mod role. Byee

[low shard]

try screenrecording it

[primal galleon]

kk

[low shard]

then yeah u can use what I said above, download Applio, and u will be able to train and inference on pre-recorded audios

those are hyperlinks, blue text that when clicked are links

[low shard]

Yw and gn

[primal galleon]

@low shard

[latent kettle]

You are too late bro. If you have sent it me when I was going to buy a GPU ill definitely spend more money on my GPU 🤣

[hazy grove]

thanks bro

[low shard]

refresh models, and be sure u click a option not just put random things

[primal galleon]

kk

[low shard]

I mean ur gpu is still good for nornal models, just not THAT good for pretrains

[primal galleon]

wait im dumb xd i put istrumental not vocal lmaoooooo

[low shard]

😭

[primal galleon]

xddd

not working either

im done

ive put vocal

[unique blade]

ahhhhhhhhhhhhhhhhhhhhhhhhhhhhhh my eyes

[primal galleon]

bruh

[low shard]

oh u got no quota

well

just use https://weights.gg

[primal galleon]

Whats that

I need a tutorial😭

[low shard]

the zerogpu quota since its an huggingface zerogpu space

anyways

just use weights

its easier

[primal galleon]

I dont know how to

Im dumb

[low shard]

https://weights.gg > cover

[brittle wing]

How do I prepare a sample for converting, so it doesn't sound noisy?

[primal galleon]

Thank you @low shard

Not home yet but ill try

[unique rock]

Does anyone know why when I try to use KLM 4.3x4 in RVC disconnected, it downloads and everything but when I start it says there is an error in the pre-training, it doesn't train, and for other people it does work?

and the KLM4.3x3 is fine

[worn raven]

How do you guys replace someone voice in a vid with a character voice?

[brittle wing]

It works fine with me, it's probably a skill issue.

[unique rock]

I usually download it like this but I think it's wrong since only the pth should go and I think that could be my mistake:

https://huggingface.co/SeoulStreamingStation/KLM43/resolve/main/D_KLM43_x4_32k.pth?download=true

https://huggingface.co/SeoulStreamingStation/KLM43/resolve/main/G_KLM43_X4_32k.pth?download=true

and it should not go "?download =true" according to me

[low shard]

use UVR to clean it

[low shard]

elaborate more

and tell ur pc gpu

[brittle wing]

Before or after inferencing
Uvr denoise or Mel-roformer denoise?

[brittle wing]

Select custom paste links and execute the cell

[low shard]

before

mel denoise

[brittle wing]

I do that and there's still noise...but thx!

[low shard]

well u need to get a better model and be sure to have a cleaner input

[worn raven]

Basically i wanted to replace someone voice with the one of a Half life scientist, just to make a meme, but i couldn't find any website that could do that

[brittle wing]

Mhm I always use the best models

[low shard]

what's ur pc gpu

[worn raven]

Idk

I have a lenovo laptop

I think it razen or smth like that

[low shard]

amd ryzen is a cpu type

[low shard]

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

[proven hill]

can you check dms please? ☺️

[worn raven]

Alr i'll do that when i can, and after i get my cpu info, what's next?

[brittle wing]

Not to be salty but who removed my BTS group chorus model from #1175430844685484042 ?
I'm actually glad tbh it was really a bad model

-colab

[azure marsh]

📒 Google Colab Notebooks

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

ℹ️ Note

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

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

[primal galleon]

@low shard how can i train my own voice?

[hollow dome]

is there a way to do batch inference non-locally?

[finite galleon]

The X4 in the G link should be x4

[unique rock]

So there are no more alternate links?

[finite galleon]

No i meant the link is wrong. Its supposed to be https://huggingface.co/SeoulStreamingStation/KLM43/resolve/main/G_KLM43_x4_32k.pth?download=true not https://huggingface.co/SeoulStreamingStation/KLM43/resolve/main/G_KLM43_X4_32k.pth?download=true

[unique rock]

thanks for sharing the one above!

you are an angel

[finite galleon]

Np

[knotty moth]

remove the "?download=true"

[unique rock]

Yes I do, but they told me that the "G" is wrongly named, that it should be "x4" and it is "X4" and the D is "x4"

[simple ore]

just go to https://huggingface.co/SeoulStreamingStation/KLM43/

and download what you need

[knotty moth]

no it's both "x4"

[unique rock]

so it's correct now?

https://huggingface.co/SeoulStreamingStation/KLM43/resolve/main/G_KLM43_x4_32k.pth?download=true

https://huggingface.co/SeoulStreamingStation/KLM43/resolve/main/D_KLM43_x4_32k.pth?download=true

[simple ore]

seems so

[brittle wing]

With or without download=true?

[brittle wing]

Should I?

[simple ore]

what are you trying to do?

just click the link to download

[merry sand]

Where else is a good place to train a model, if I can't use RVC Disconnect for hours, like a 300 epoch thing from a 40 something minute audio of a character?

[knotty moth]

if you have at least 6 GB gpu, recommended to train locally without worrying of such limits

another cloud option is:

-kaggle

[azure marsh]

📘 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

[merry sand]

Is it possible to resume with another account, after it's saved?

[knotty moth]

don't lose preprocessed files and G & D files, or save them before and then load

[merry sand]

Does the rvc logs part count?

Edit: Nevermind. "File unreadable".

[hollow dome]

this is stupid but i tried to batch inference myself and my computer bluescreened lmao
could i pretty please get someone to do 2 folders of short clips for me :3? i can train a model for you in return

[tiny totem]

I was already making a CVC UTAU Vb out of this anyway so maybe if I try messing around with something like that too I mught could get something

[tiny totem]

which notebook is this that has the hybrid??

Actually is there somewhere that has an Explanation of each of the methods? Ik I can just look it up but in context of TVC I'm sure there can be a large difference

[formal wind]

Ok, I would like to go into more detail than I did yesterday. I'm having an issue on applio rvc where It takes a really long time to train. Basically what happens is, when I go to train a model it compiles for like 15 mins. Then it goes to the first epoch and takes like 30 mins to finish. Then it compiles for like 5 mins. Then Trains the next epoch for 30 mins, and so on and so on. The feature extraction and all other steps are fine its just training. I've asked for help last night but I couldn't fix the issue. My graphics card is an "AMD Radeon RX 6600" I think I set it up right for amd. If I didn't then someone would have to message me a video or somethin cause I tend to fail when im reading words as a tutorial. (Im an idiotic visual learner.)

[jaunty nova]

hey anyone know how to make voice ai less choppy when recording?

[knotty moth]

u don't mean that paywalled garbage?

there's no way a recording could be choppy unless the audio driver or device may be faulty

[jaunty nova]

hmm maybe it was on my end. would increasing the gain help?

[knotty moth]

it would do nothing at all

[rustic hare]

my friend with a geforce rtx 1650 is asking what should his settings should be

[knotty moth]

rvc inference can run well and not so slow on 1650

[rustic hare]

yeah but like what chunk and stuff

[knotty moth]

rvc doesn't have such that option

[rustic hare]

i forgot which one did

[knotty moth]

I answered so because you have asked in this channel instead of #🔍│help-w-okada if it means the voice changer context

[rustic hare]

yes

[swift crag]

what does the index slider do

[formal wind]

increases the accent, or effects depending on the dataset

[formal wind]

So anyone got a fix for this?

[formal wind]

Nvm its doin ok for now

[formal wind]

Ok well the speed is fine now but it keeps having an error every like 9 epochs so I have to keep training it from where it left off everytime. Which is kinda annoying ngl

When the error happens again ill let yall know what it says

[twin kestrel]

can u all help me

gimme training link pls

[formal wind]

It says:

Process Process-1:
Traceback (most recent call last):
File "C:\Applio\ApplioV3.2.8-bugfix\env\lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Applio\ApplioV3.2.8-bugfix\env\lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "C:\Applio\ApplioV3.2.8-bugfix\rvc\train\train.py", line 484, in run
train_and_evaluate(
File "C:\Applio\ApplioV3.2.8-bugfix\rvc\train\train.py", line 739, in train_and_evaluate
save_checkpoint(
File "C:\Applio\ApplioV3.2.8-bugfix\rvc\train\utils.py", line 118, in save_checkpoint
os.replace(old_version_path, checkpoint_path)
PermissionError: [WinError 5] Access is denied: 'C:\Applio\ApplioV3.2.8-bugfix\logs\Zoom\G_2333333_old_version.pth' -> 'C:\Applio\ApplioV3.2.8-bugfix\logs\Zoom\G_2333333.pth'

and i figured out that it ends when this ends

[twin kestrel]

.

[tame mica]

-cloud

uh

[tame mica]

-docs

[azure marsh]

📚 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

[tame mica]

click on how to make voice models

[low shard]

since you got a gtx 1660 super, i would suggest cloud

Train (make) RVC Models on cloud:

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

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

3. Be sure to know about the tensorboard

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

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

You can use either:

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

[low shard]

what's ur pc gpu

[low shard]

did u get it?

[twin kestrel]

im using handphone tbh

i dont have pc

[hallow thistle]

Handphone? Headphones or smartphone?

[low shard]

ig u meant smartphones

[hallow thistle]

[low shard]

well, that's not really the best, you can do it on Cloud (remote good pc)

Train (make) RVC Models on cloud:

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

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

3. Be sure to know about the tensorboard

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

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

You can use either:

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

However, the User Interface won't be much mobile friendly

[knotty moth]

local training is still viable for relatively small dataset and batch size 4 though less optimized than RTX cards

[low shard]

that's true that he can still train, which I also said before to him, but isn't it better letting him use Kaggle, having higher speed and not being limited to certain batch size and dataset lenght?

[knotty moth]

true, a single T4 has less raw performance and clock speed than 3060, but has more vram

[simple ore]

seems like you somehow have a problem with permissions

[low shard]

a T4 is surely better than a gtx 1660 super, and with kaggle he can get a P100 or T4x2 which is surely even better in raw performance

[simple ore]

@formal wind make sure your account owns C:\Applio and every child folder as well, dont run Applio as admin

[knotty moth]

tbh T4 is better, because it is turing vs pascal in terms of optimization in AI performance

[formal wind]

How do I do that 😭

[hallow thistle]

"Every child folder" means every subfolder and its content inside Applio folder.

[brittle wing]

Does anyone know where I can redownload the program to use the voices? I can't find the link on the server?

[formal wind]

yeah I mean like how do I make an account own a folder

[simple ore]

right click on Applio, properties, Advanced

[hallow thistle]

Redownload the program of which?

[brittle wing]

The program that uses the voice models? The folder download. It's been a while since I used it

I'm not sure if they updated it

[hallow thistle]

W-Okada or RVC? Both uses RVC voice model.

[brittle wing]

I think RVC

[hallow thistle]

-rvc

[azure marsh]

📚 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]

RVC is audio conversion program. But I'd suggest for Applio for easier access.

[formal wind]

Alr I think I did it'

[brittle wing]

I don't know anything about that, what are the differences?

[formal wind]

Ill try training later on

[low shard]

RVC = inference (use mmodels) on pre-recorded audiso and training (make) models

Wokada= realtime voice changer for calls

what's ur pc pgu

[hallow thistle]

W-Okada is "realtime" audio conversion program that uses RVC voice model to convert voice in realtime.

[low shard]

and what are u looking for

[low shard]

mm I'm not sure about that

[brittle wing]

Ah okay, that makes sense, I'll try to figure out how to download that

[hallow thistle]

NVIDIA T4 is older than H100, but it's also used in Google Colab.

[low shard]

so you need RVC or Wokada?

because it's different links to download it

[hallow thistle]

Which?

W-Okada and RVC are two different programs. Pick one for your needs.

[brittle wing]

W-Okada

[low shard]

tell me ur pc gpu

to check if ur pc is good enough

so i can send u the links to use it

[hallow thistle]

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

[low shard]

^

[simple ore]

and dont use overtraining detector

[brittle wing]

I have an AMD Ryzen 7 5800H

[low shard]

that's a cpu not gpu

check what I said in #🔍│help-w-okada

[hallow thistle]

AMD Ryzen is CPU, while AMD Radeon is GPU.

[formal wind]

Sure but whats wrong with it

[brittle wing]

Oml I read it wrong lmao, hold on lemme check

[low shard]

lol dw, just please let's go in #🔍│help-w-okada since this is the wrong channel

[hallow thistle]

If your AMD Radeon has RX in its name, it's dedicated GPU. If not, it's an integrated GPU.

[formal wind]

so the thing is. the bigger datasets are the ones that take literally forever. (like 20 or more mins) and the smaller datasets finish way faster. How do I speed it up so im not waiting for years.

So basically the original issue apart from the fact that it doesnt say compiling

[simple ore]

how much vram is being used with your big dataset training?

what's batch size?

[formal wind]

8

[knotty moth]

it's just ur usual toyota car vs porsche

[tidal cairn]

https://huggingface.co/Yohimmu/max/resolve/main/D_2333333 (1).pth @low shard

[knotty moth]

not just that single file, but the entire preprocessed files in logs/yourmodel and G & D files

[tidal cairn]

Where do I find this?

[low shard]

the specific file you linked isn't related to using the model, are you talking about using a model or resuming training?

[hallow thistle]

You must have put pth and index files into a single zip file.

[tidal cairn]

I wanna use the model I created

So confused sorry for the lack of clarity. I just wanna use the model I made

[low shard]

the zip has only the D and index

[tidal cairn]

What file am I missing?

[hallow thistle]

[low shard]

the namemodel.pth file

you can't use it unfortunately

the index is basically the accent, the namemodel.pth is the actual voice

[tidal cairn]

So would I need to retrain?

[hallow thistle]

Yes.

[tidal cairn]

If someone can create the model for me and upload it properly, i’m willing to send money for it. I know it doesn’t take long but I obviously don’t have a clue of what to put 😂

[knotty moth]

start over from preprocessing, since it seems you have lost all those files

[low shard]

rip

[low shard]

If u want there's #1159289738314919936 or #1191429836321849435

[tidal cairn]

Okay

[low shard]

or u could try again doing it urself

your choice

[knotty moth]

I'd recommend alternative one like the kaggle notebook, though it may require phone verification to enable some needed features

-kaggle

[azure marsh]

📘 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

[low shard]

yup that's true

Kaggle gives better GPUs and more gpu time

However it's harder

[tidal cairn]

Trying to do it again, now it’s not displaying my path 💀

But gave me a index

?

[tidal cairn]

Got sent here just to get ghosted, tough. Thanks for the help anyways. I’ll continue to post updates hopefully I can get some assistance. Been in the server for over two years, this the first time I been left stranded

[raw hedge]

im training an rvc voice on applio, i have turned on the overtraining detector, but for some reason i cant understand it?

[low shard]

You are sure the colab runtime didn't disconnect?

[tidal cairn]

Yeah it’s giving me no path only index.

I’ve tried restarting it 5 times now since we last spoke.

[low shard]

What if you click on "select pth file", do you see it?

[tidal cairn]

it finally just now gave me a path

[low shard]

Btw it's called pth not path

[tidal cairn]

so how would I save this so I can reuse it

rather than have to consantly retrain

[low shard]

That's the G pth, still not the actual one to use the model

Show a SS of the options in the menu

[tidal cairn]

@low shard

[low shard]

I mean a screenshot of the options in "select a pth"

[low shard]

Are there no other options?

It saved the model but only the epoch 10 and 20 ones

[tidal cairn]

yeah that's all I see

do I need to change any of these?

[low shard]

the training stopped I guess

[low shard]

how long is ur dataset

[raw hedge]

.

[tidal cairn]

Okay

[low shard]

put batch size to 8

[tidal cairn]

Gonna do it rn

[low shard]

Oh I just noticed u changed ur message

btw u should use kaggle instead of colab

[tidal cairn]

I didn’t mean to change it

But you said kaggle is harder right?

[low shard]

yes but you don't have the risk of it getting disconnected randomly (since google colab free isn't granted)

so u don't accidentally waste hours for it to just disconnect out of nowhere

[raw hedge]

@low shard

[low shard]

I haven't used Applio's Overtraining Detector

[tidal cairn]

Okay for sure

[knotty moth]

don't use it if not sure

[raw hedge]

i didnt thoguh for sure that you needed the IQ of a scientist to understand this shit either

[low shard]

this is the applio version on kaggle https://docs.applio.org/applio/getting-started/other-alternatives#kaggle

@crude flame btw u gotta add also the kaggle applio on the docs, and thanks for updating the docs too

[tidal cairn]

how to start?

[knotty moth]

Copy & Edit

again read the guide

[low shard]

@simple ore what's wrong with Applio's site? some links are broken like https://applio.org/learn/58

some are saved on the wayback machine like https://web.archive.org/web/20241222222208/https://applio.org/learn/58

[simple ore]

i think thing were moved

https://docs.applio.org/applio

[low shard]

I think some were removed instead of moved

like for example https://docs.applio.org/applio/guides/other-guides redirects to https://applio.org/learn which doesn't exist

or at https://docs.applio.org/applio/getting-started/other-alternatives#how-to-upload-your-dataset-to-imjoy-elfinder redirects to https://applio.org/learn/58

which doesn't exist either

[simple ore]

lemme ask

[low shard]

alr

[snow rose]

Why im getting error when im uploading any .index from weight.gg other sites work normally. Only index from weight.gg does not work

[simple ore]

the naming is wrong

but it may get fixed in the next release, regardless you just save the files under /logs/modelname/

without using UI

[paper fossil]

python train_nsf_sim_cache_sid_load_pretrain.py -e "Zul_Experiment" -sr 40k -f0 1 -bs 8 -g 0 -te 500 -se 50 -pg pretrained_v2/f0G40k.pth -pd pretrained_v2/f0D40k.pth -l 1 -c 0 -sw 1 -v v2 -li 16

FileNotFoundError Traceback (most recent call last)
<ipython-input-39-1e60057091e2> in <cell line: 86>()
86 set([name.split(".")[0] for name in os.listdir(gt_wavs_dir)])
87 & set([name.split(".")[0] for name in os.listdir(feature_dir)])
---> 88 & set([name.split(".")[0] for name in os.listdir(f0_dir)])
89 & set([name.split(".")[0] for name in os.listdir(f0nsf_dir)])
90 )

FileNotFoundError: [Errno 2] No such file or directory: '/content/Mangio-RVC-Fork/logs/Zul_Experiment/2a_f0'

I got this error, can somebody help?

[simple ore]

forgot to run extract features?

[paper fossil]

Already did

[simple ore]

well, that should've created 2a_f0 folder

[snow rose]

I know but still its weird that only one site does not work

[quasi steppe]

i do NOT know how to train an ai voice and i really need to 💔

[simple ore]

because weight.gg models are names model.pth model.index

[low shard]

I mean he's the creator, I was just saying since there are some guides like Applio Kaggle one that could help newbies

idk about other ones that were there

[paper fossil]

You are using torch.load with weights_only=False (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for weights_only will be flipped to True. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via torch.serialization.add_safe_globals. We recommend you start setting weights_only=True for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.

[paper fossil]

I got this message when I run that. Is this and issue??

[simple ore]

@paper fossil not an error, just a warning for torch 2.4+

[paper fossil]

I see

[simple ore]

shoud not be using 2.4.+ anyway

[paper fossil]

So what should I do

[low shard]

what's ur pc gpu

[simple ore]

perhaps colab by default has torch 2.5

[low shard]

be sure that feature extraction didn't say 'no-feature-todo'

[simple ore]

you need 2.3.1 max

[quasi steppe]

rx 6700 xt 🪦

[simple ore]

can run applio locally with Zluda

[low shard]

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

[low shard]

but let me know if it comes back

[paper fossil]

Where can I see that?

[low shard]

in the output of the cell

I'm guessing you're using RVC-Disconnected

[paper fossil]

Yes

[low shard]

yup check the output

[simple ore]

I dont that's the issue... f0 is extracted using rmvpe, it loads saved weights.. if there's a new torch it wont load weights, so f0 extractionf fails

[paper fossil]

And where can I check the output?

I'm sorry for troubling you guys, I'm still new to this

Just started

[low shard]

send a screenshot of the colab

[low shard]

worth to check

lemme see colab's patch notes

[paper fossil]

That's the thing, I can't send a screenshot here, that's why I'm copying and pasting😅

[low shard]

nothing changed since a like 3 months soo

!give-media-perms 30m @paper fossil

here u can now

[paper fossil]

Thanks

Do I have to install this?

[low shard]

u shouldn't have to

show a ss of the full feature extraction cell

[paper fossil]

Okay then how do I check the output?

[simple ore]

as I said

[low shard]

wtf

[low shard]

why don't they say that in the release notes

[simple ore]

so whatever notebook that is, it does not have a pinned torch version

[low shard]

nvm it seems updated since some months

[simple ore]

so the default is useed

[low shard]

it's rvc disconnected, however it's weird since I never seen this happen before

[paper fossil]

Maybe I'm using a link that is from a video upload a year ago. Is that the case?

[low shard]

youtube tuts are outdated 😭

don't follow them

[paper fossil]

I see

[low shard]

what's ur pc gpu?

[paper fossil]

3080

[simple ore]

so why are you using colab with oudated mangio lol

[paper fossil]

I didn't know XD

Where is the updated ones?

[simple ore]

https://huggingface.co/IAHispano/Applio/blob/main/Compiled/Windows/ApplioV3.2.8-bugfix.zip

download, unzip to C:\Applio

[paper fossil]

Okay

Thanks

[simple ore]

make sure you have cuda tools 12.x installed

https://developer.nvidia.com/cuda-12-1-0-download-archive

you only need 'runtime', not everything, so unselect the rest in the custom setup

[hallow thistle]

CUDA for NVIDIA GPUs.

[knotty moth]

bruh you should train locally, let ppl without decent gpu like yours use the colab

[paper fossil]

I see

[paper fossil]

Which one?

2022?

[simple ore]

10 or 11, what your windows version is

[paper fossil]

Ah, I see

Okay 11 it is

[hallow thistle]

If you use Windows, select Windows. If your Windows number is 11, select 11.

[knotty moth]

u dont know your own OS lmao?

[hallow thistle]

The architecture will always be x86-64 or simply x64.

[paper fossil]

Ofc I know, thought those were versions

[hallow thistle]

You can type winver in Run (Windows key + R), and let the program tells you which Windows you're using.

[paper fossil]

[hallow thistle]

However, if you're using "Windows Server", there are two most recent versions available for download.

[low shard]

ur pc is good enough to do it locally lmfao

google colab is a cloud computing service only for bad pc people

[low shard]

yeah

[low shard]

nuh uh

[low shard]

he's prob NOT using that

[hallow thistle]

[paper fossil]

My pc just died after I installed the Cuda

Now i can't turn it back on

No lights, no anything

[crude flame]

Okie dokie

[paper fossil]

Tried turning on and off the psu

[low shard]

What the actual fuck?

Are you 100% sure it doesn't turn on at all?

[paper fossil]

Nvm

I turned off the plug with my fat feet

[low shard]

lol

[brittle wing]

What happened with RVC V3?

[simple ore]

it never existed

[brittle wing]

There was a rumor that the developers are working on it...

[simple ore]

originall developers did all but abandoned it to play with their new pet project

[brittle wing]

Then what's all that in #🔊│ai-development ?

What's Codename doing?

[simple ore]

we are doing some enhancements, testing new vocoder

[brittle wing]

Enhancements of what

[simple ore]

a lot of under the hood improvements, faster training, less memory use

it is for Applio and Codename's extra features on top of that

[brittle wing]

Uh so no one will ever make rvc3

[simple ore]

there will be Applio v3.3

[brittle wing]

Will the Applio colab gradio be updated along w that

[brittle wing]

Colab?

Or fork only

Will the models of now work w it?

[simple ore]

yes and yes

[ancient ocean]

Why I dont have access to archived info channel links? I want to find ways in order to create AI Covers

[glacial pollen]

Waiting.

Until these " experiments " on figuring out what works and what doesn't end, I am at hiatus

No point wasting my gpu on testing potentially broken things

• electricity in my town, esp during winter has quite a high failure rate and that wrecks long trainings ( those that last 3-6 days ) so esp given that, I have no choice but to wait for now

[brittle wing]

Oh thanks for answering, so no RVC V3?

[brittle wing]

Nice

[glacial pollen]

Well no, that I didn't say

Once we figure out the most optimal way of doing stuff, " v3 " will be a thing

[brittle wing]

It was just a rumor then

[glacial pollen]

No, I mean I didn't say there won't be v3

If I was to be honest with you, changes we've incorporated so far could be even deemed as v4

It's far beyond what stock rvc was at this point in many aspects aside of the core

In any case, it'll take a bit so, patience would be appreciate from everyone ~ ✨

For more, anyone should be actively checking #🔊│ai-development
Tho please, without annoying asking about million things or requesting codes / models without prior knowledge on how to use it

[brittle wing]

I was just asking.

[glacial pollen]

and i was just saying.

never hurts to let others know in advance

[simple ore]

in short, there will be no "RVC V3"

[green chasm]

WARNING: The following packages were previously imported in this runtime:
[matplotlib,mpl_toolkits]
You must restart the runtime in order to use newly installed versions.

getting this when i run the Dependencies cell in the rvc v2 disconnected colab

i've restarted the runtime like 4 times

and i'm getting this when i try to train
ValueError: 48000 SR doesn't match target 40000 SR

[red bramble]

Running with the runtime Python, Please wait.
No supported Nvidia cards found, using CPU for inference
cpu
.pth file directory: ./models\Müslüm Gürses - Weights.gg Model\model.pth
.index file directory: ./models\Müslüm Gürses - Weights.gg Model\model.index
loading ./models\Müslüm Gürses - Weights.gg Model\model.pth
gin_channels: 256 self.spk_embed_dim: 109
<All keys matched successfully>
sid: 0 input_audio: C:\Users\Poyraz\Downloads\2024 mustafa.mp3 f0_pitch: 0 f0_file: None f0_method: crepe file_index: models\Müslüm Gürses - Weights.gg Model\model.index file_big_npy: index_rate: 0.4 output_file: C:\Users\Poyraz\Downloads\2024 mustafa_RVC_1.wav
Traceback (most recent call last):
File "C:\Users\Poyraz\Desktop\RVC-GUI-pkg\vc_infer_pipeline.py", line 285, in pipeline
index = faiss.read_index(file_index)
File "C:\Users\Poyraz\Desktop\RVC-GUI-pkg\runtime\lib\site-packages\faiss\swigfaiss.py", line 8261, in read_index
return _swigfaiss.read_index(*args)
RuntimeError: Error in __cdecl faiss::FileIOReader::FileIOReader(const char *) at D:\a\faiss-wheels\faiss-wheels\faiss\faiss\impl\io.cpp:68: Error: 'f' failed: could not open models\Müslüm Gürses - Weights.gg Model\model.index for reading: No such file or directory
Initiating prediction with a crepe_hop_length of: 128

[brittle wing]

I didn't want to annoy you

[brittle wing]

Nice

[brittle wing]

V4 then

So, it was just rumors

[glacial pollen]

Just an upgrade

" vN " naming convention is no more

V1 V2 etc
no point

the only reason it was v1 and v2 was due to the use of v1 and v2 hifigan's configuration and few other aspects

Anyway, I go dnd, got some stuff to do so, take care

[simple ore]

Again, it is very unlikely we'll see an official RVC Vanything

[brittle wing]

Have a nice day/night!

[brittle wing]

I understand

[sudden tree]

anyone know if an official applio update is coming soon for prepacked?

[steady kindle]

some one can help me with that (Modulator?)

[low shard]

With what

[steady kindle]

eee how to say itt

[steady kindle]

with say like as a girl but u re boy

you know

idk how to say it

you know

[steady kindle]

hello?

[low shard]

I'm guessing real-time voice changer by what you say

Wrong channel

[steady kindle]

oh

[low shard]

Rvc is not for that

Wokada is tho

[steady kindle]

W okoada?

okay

thx

[brittle wing]

This isn't overtraining,right?

[sudden tree]

looks perfect

[brittle wing]

can someone help me with enabling RefineGAN? I have the latest version of Applio, but when I have Refinegan installed, i don't see an option on training to set it to "Applio" like the forum post in #1235952130855010365 says to do

[azure marsh]

Hey, Echoes/DragonWinter01! Please use the command !howtoask to increase your chance of getting help by structuring your question in a way others can understand better. Also make sure you're asking in the right help channel:

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

[brittle wing]

nor do I have an option to set the sample rate to 44k, which is the version of the pretrained model I chose

[radiant quartz]

what the frick

🥴

[radiant quartz]

current settings:

dataset: 500 voice lines
dataset duration: 12 mins for 1st file, 15 mins for 2nd file, then
combined is 28 minutes

save frequency: 25
total epochs: 170
batch size = 8

was that me undertraining or over

[simple ore]

the standard chart like this is amost unusable for small datasets as it only logs the value at the end of the epoch

[violet swan]

im assuming loss/g/total just logs value of 1 sample? @simple ore

[analog obsidian]

Only logs the last value at the end of the epoch

so like if each epoch has 126 steps, the old graph only logs the last value

of those 126 steps

[simple ore]

yeah, it could be a good value or bad value, since the samples are shuffled randomly it is hard to tell

[violet swan]

and avg 50 is just avg over 50? only logging random 1 sounds stupid

[simple ore]

average 50 in Applio unreleased version does an average over last 50 steps

[paper fossil]

Does it have to be in C drive? Cuz it's almost full

[simple ore]

Any drive is fine

[formal wind]

Bro im boutta start TWEAKING if I can't sort this issue out. Noobies you gave up on me 😭

[simple ore]

mainly not doing it on Desktop or c:\windows\system32\

[simple ore]

dunno dude, plenty of people followed the guide and it works fine

as long as you actually read and follow steps

[formal wind]

Ill retry again ig 😭

[simple ore]

see how much VRAM is used when you're training

[formal wind]

how do I do that

[simple ore]

in task manager/performance, just like the last time

[knotty moth]

btw the grad metric also logs last step of the epoch, so it might be better to include avg grad

[formal wind]

Hey is there a applio vid tutorial for amd. I've checked on youtube and found nothing

[simple ore]

dont think so

[formal wind]

dangg

I just reinstalled it so ill see if its workin

[hallow thistle]

He was using the original version of W-Okada though, which is now outdated.

[brittle wing]

can someone help me

[azure marsh]

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

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

[hallow thistle]

Please be specific on which problem you're encountering.

!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.

[brittle wing]

i download the cuda vrc and it dindt help idk why i got a rtx 4060

[hallow thistle]

W-Okada or RVC?

[brittle wing]

rvc

[low shard]

rvc = inference (use models) on pre-recorded audios, and train (make models)
Wokada = RVC but for realtime, realtime voice changer for calls

[brittle wing]

OKADA my bad

[hallow thistle]

CUDA toolkit with RVC? What?

[low shard]

tell me the tutorial link u followed in #🔍│help-w-okada

[jovial dock]

how to make train

[hallow thistle]

I was about to say that I didn't know how to get CUDA to work with RVC, but damn you got me again. People be confusing W-Okada for RVC, perhaps.

The only thing W-Okada is related to RVC is that W-Okada uses RVC voice model to inference. Its codebase and GUI themselves aren't related.

[formal wind]

Yo is someone able to help me set up applio for amd (Or even nvidia and I'll do the amd options along side it) with like a video or sumin in priv messages. I'm too mentally stupid to do it reading words 😭

[hallow thistle]

Why did you send this screenshot to #✨│ai-help? This is the original W-Okada, the long outdated W-Okada, which uses an actual window to display its GUI.

[glacial pollen]

What's the issue?

Any particular / exact thing or an element that won't work for you?

[simple ore]

you still have not posted a screenshot of the task manager showing the VRAM used during training

like how many times do I need to ask for that?

[glacial pollen]

what's the issue they deal with btw?

ah, the speed if I get the context right

welp, ye, as noobies said, first off post the vram usage

1. ctrl+shift+esc to open up task manager
2. performance tab
3. screenshot the gpu section during training or whatever like so:

And if you don't comply with our requests, we can't really help you much

[hallow thistle]

My laptop doesn't have a GPU.

[simple ore]

there's always an option for external gpu

[formal wind]

i didnt see a vram option in ma task manager, and im just gonna keep trying in different ways until it works. 😭

[formal wind]

im not at my pc though atm so ill send it when i can'

[west obsidian]

how to make ai cover?

help me

[low shard]

What's ur PC GPU

[brittle wing]

ok guys so when i just rvc i get a good voice model however it doesn't sound like the character because i dont have their accent

[low shard]

RVC = inference (use models) on pre-recorded audios, and train (make models)

Wokada = RVC but for realtime, realtime voice changer for calls

Which are you talking about?

[brittle wing]

i use rvc inference

[low shard]

since you said "I don't have their accent", are you sure that you aren't using wokada for realtime voice changing for calls?

Or are you recording your audios and using them as an input in RVC?

[brittle wing]

i use my voice for audio recording then i convert them with rvc inference i do not use realtime

[low shard]

oh alright then, you could try playing with the index ratio, it controls how much the index is used, which is the accent of the model

you could set it higher

[brittle wing]

is this for apollo

[low shard]

You mean Applio?

Yes it is

[brittle wing]

that was a typo yes appilo

[low shard]

Applio is a fork (modified version) of RVC

[low shard]

https://docs.aihub.gg/rvc/local/applio/#4-modify-settings-optional

check in the advanced settings

in some RVC it's called index ratio, in applio it's called search feature index

[brittle wing]

welp i have deleted rvc because it wasn't working

i know that sounds dumb

so can i just send you the voice model

and you test it out for me

[low shard]

Could u tell me ur pc GPU?

U can download it back

[brittle wing]

ok i will download it back

but it will take a long time

[low shard]

wait could you tell me your PC GPU? You can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab -> GPU

[brittle wing]

its a gtx 970

[hallow thistle]

Applio won't work if you run Applio with Run as Admin mode on.

[low shard]

oh that's old

it works but will be kinda slow

[brittle wing]

some people told me thats the reason y it sounds bad