✨│ai-help

[polar dove]

ok

how good is it on phones

[low shard]

No phones just use cloud

I was just using the wrong copy pasta, don't worry about it lmao

Use the links I sent above

[polar dove]

ok thanks

[low shard]

yw and lmk

[cobalt vine]

can someone tell me why like this ?

its my first time use kaggle

andi cant see the notebook settings there

[visual tendon]

Help me pls I want to make voice model...😭

[low shard]

what's ur pc gpu

[low shard]

elaborate:

• what's ur pc gpu (check via task manager)
• what tutorial link are u using
• what u want to do
• what links are u using

[cobalt vine]

i click the one wrote open the colab

i want to u se the voice changer but sorry the tutorial its confusing me

[low shard]

elaborate the things I asked you

[low shard]

@cobalt vine

[eternal aurora]

@low shard wsg did you know how to fix my issue?

[low shard]

elaborate

[brittle wing]

Hi guys, I wanna have a suggestion born from your experience with AI training.
I have prepared in these days a good Italian dataset with 16 female speakers and 26 male speakers, good amount of hours (haven't calculated totally yet, but it could be > 20 hours).
I already know how to train a model with Applio, and how to prepare the dataset ecc. just wanna know if there are some solutions I can use since my actual workstation is resource-limitated for that task. I mean, is there any cloud platform where I can just upload my dataset and do the fine-tuning on the original RVC model?

[low shard]

are you going to train a pretrain?

[brittle wing]

Yes

I just want to understand the best practices to do it since my dataset is ready

[low shard]

free or paid cloud providers? because i'm not sure you will be able to do it for free

[brittle wing]

Both of them, I want to invest on that

If there's no free alternative then go for paid

[low shard]

Train (make) RVC Models on cloud:

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

• Google Colabs (max 4 hours of daily gpu not granted for free, not much hours for training, but easy to use, there's a paid tier):
  • Applio (UI)
  • Mainline (U, broken right now)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus, only free):
  • Applio (UI)
  • Mainline (UI, broken right now)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly, but there's a paid tier):
  • 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.com/ which ofc uses RVC

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

You can use either:

• Weights.com: Easiest Possible Ever Automatic
• Ilaria RVC Zero: Fastest free on cloud
• Applio UI Colab: RVC Fork with some extra features like TTS
• RVC AI Cover Maker UI: Automatically Separates the Vocals and Instrumentals, converts the voice and mixes them back

for training a pretrain, it would be better you may pay for colab or use another service (like runpod) with your own notebook (should be similar to the google colab code), there isn't much support for paid ones because usually they don't use cloud to train pretrains as people don't really want to pay

the max you could get for free is kaggle applio, but i really doubt it will be enough

[brittle wing]

30 hours... mh it depends on which GPU they are going to assign

How many epoch should I use for training a pretrain? Not from scratch, it's a fine-tune, I want that the RVC model is going to understand better the Italian accent

[low shard]

either T4x2 or P100

[low shard]

yeah that's the thing, we can't know, you have to use the tensorboard https://docs.aihub.gg/rvc/resources/training/#epochs--overtraining

https://docs.aihub.gg/rvc/resources/training/#how-do-i-make-pretrain this can also help you

I'm italian too btw, goodluck

[brittle wing]

Thank you man, hope I will succeed on that. If I will get a good result, I'll share it for sure

[low shard]

there's #1235952130855010365 for that

you need to be model maker first of all though

[restive plinth]

it works so much better now! only now if i play the game and use it it will freak out again, is this just a graphics card issue atp?

[low shard]

what game? did u set the graphics to the minimum?

[restive plinth]

overwatch, and yes everything lowest possiblee

[low shard]

show a screenshot of ur wokada while running it in game, and close everything else in background

[low shard]

this is while running the game?

[restive plinth]

yeahs

[low shard]

it doesn't seem lagging, the perf value is green

go in game in a match and see how's the perf value

[restive plinth]

okay

it did spike a lot

and it times out when i am in game for too long

[brittle wing]

So I think I have to post a model on Weights.gg

Calculated the dataset length, currently is 13:46:59

For 17 female, 28 male speakers

[low shard]

set extra to 0.2, uncheck sup1, set chunk to 1400

[low shard]

#outdated-model-maker-role

[restive plinth]

kk should i send another screenshot when i do?

[low shard]

yes

[crude flame]

you dont, you can post it to huggingface or weights

[simple ore]

13.5h is not a particularly good size for a pretrain dataset

if you're doing it from scratch

[brittle wing]

I want to do a fine-tune of the original model, not from scratch, it's not my intention

[simple ore]

okay, just don't expect to get 45 unique voices out of that

[knotty moth]

you were unsatisfied on finetuning a single speaker normally?

[brittle wing]

I read on the docs that for doing a pre-trained model I must have at least 10 hours of data with many speakers

It's not a question of not getting satisfied with single speakers

[knotty moth]

for some inspiration, you can take a look on this pretrain https://discord.com/channels/1159260121998827560/1339155300720054316

it was trained on a lot hours of dataset

imo, if you want to make a pretrain dedicated for a specific language, it tends to be better to use much more amount of dataset

[brittle wing]

Thank you for the advice, I will consider that

[restive plinth]

atp it might be a gpu issue

[low shard]

yeah seems like your gpu is too weak

i mean

it's pretty old

u can't expect so much

[restive plinth]

it is lol well thank you so much for the help

[low shard]

you can try https://docs.aihub.gg/w-okada/local/w-okada/#reduce-more-delay , setting the f0 to fcpe, setting extra to 0.1

this would be fully focusing on speed rather than quality

or if you want, there's cloud (remote good pc), the only working one at the moment is wokada deiteris fork kaggle https://docs.aihub.gg/w-okada/cloud/w-okada-kaggle/

which gives 30 hours weekly of good pc, but you need to login + do a phone verification + it's kinda harder to setup

[restive plinth]

i see, ill just stick to what i got now and later save up for a more modern and better gpu ;p

[low shard]

alright

[elfin nebula]

Just a question. What are epochs?

I see it in the #1175430844685484042

[eternal aurora]

Let me show u

[analog obsidian]

number of times the model has seen its dataset

[visual tendon]

Yo bro, Do you know what's the open-source version of ElevenLabs? I noticed there's a lot of great TTS but it seems like none of it have the quality of ElevenLabs...

[trim sparrow]

does anyone know the solution to this problem in applio google collab plz ?

[somber canyon]

I wanna start making models but dont know understand anything about Datasets or how to do it, any guides out there?

[crude flame]

https://docs.aihub.gg/essentials/how-to-make-voice-models/

[trim sparrow]

yes I saw this message but I am used to using Applio and I am sure I didn't forget anything, I followed each step

I have the impression that the problem is at the extract features step. Because it almost skips the step by telling me "extracted successfully"

[simple ore]

show a screenshot of preprocess log and extractr features log

[low shard]

epochs are a unit of measuring the training cycles of the AI model

basically the amount of times the model went over its dataset and learned from it

they don't mean how good is the model, it's just an info provided on how they trained the model by the model maker
More ≠ better
Less ≠ better

There's no way to determinate how good the RVC model is until you try it out or listen to the audio samples if there are

[low shard]

?

[low shard]

there's no open source better than 11labs

[somber canyon]

is training voice models locally better or worse?

[low shard]

but you can check https://docs.aihub.gg/tts/tts-tools/

[low shard]

what's your pc gpu?

[somber canyon]

i got a 3060 with 12GB Vram

[low shard]

it all depends on how powerful is ur pc

[somber canyon]

i heard Vram is what matters

[low shard]

great

[somber canyon]

also i got a APU but im not sure if that matters

[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

[trim sparrow]

@simple ore

I only have that in the console

[somber canyon]

is applio better and easier to use than the regular RVC? because its my first time making a model

[low shard]

applio has performance improvements + easier

[simple ore]

okay, so 3 min, 68 slices.. how big is the batch size?

[trim sparrow]

I set batch size at 15 on the ui

[simple ore]

for 3 min? make it 4

[somber canyon]

umm i went to Applio's huggingface site and i cant find the download button. I'm i blind?

[trim sparrow]

ok I will try

[somber canyon]

Do i download the one called "ApplioV3.2.9.zip"?

[simple ore]

yes

[trim sparrow]

It doesn't seem to work to change the batch size, but am I the only one who have this problem ? because in this case I will try with another dataset

[somber canyon]

i have a voice model but i wanna add a few things to it, is there a way with Applio?

[brittle wing]

-colab

[patent trellis]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[river gulch]

I'm not good in terms of PC things, but for what I read, the colab use part of your PC for run? I'm using RVC and I can only ran on 2 VRAM, because if I put 4 (my max VRAM) it will stop the process.

And what is the difference between 50 epochs to even bigger ones? I have only done some models with 50 Epochs, they do work well, but I don't know what it will be the difference with 100 or more (and not the laptop for train one of 100+)

[odd shale]

Actually no, colab doesn't use your PC's hardware to run because it's cloud based to work

In few words, it works using a remote PC from a server.

[somber canyon]

.

[odd shale]

And actually i would suggest you to train your model till it reaches it's lowest point on the tensorboard graph. and bigger epochs doesn't mean better results because overtraining exists.

[odd shale]

Add a few things like?

[somber canyon]

like maybe laughing and stuff

i heard someone who can laugh with rvc

[odd shale]

You only can by expanding the original dataset you used to train it on first place.

[somber canyon]

oh

alright

[odd shale]

And then retraining it of course.

But keep in mind RVC and realtime/w-okada struggle to reproduce laughing properly without breaking even if you add laughing to the dataset. (but it can at least reproduce some light laughing types but not all)

[river gulch]

Thanks god.
I did also read that you need keep attention while training for some errors? I read this I believe a year ago so I never have try out.

[odd shale]

Welp, you need to keep attention on tensorboard to avoid overtraining.

[analog obsidian]

it can do a couple of laughing types but not all

[odd shale]

factos.

[river gulch]

Tensor... what? the image generation page? Sorry, I got a little confused there.

[odd shale]

Nope, it's unrelated to image generation.

In few words it's a graph that allows you to see when to stop training a model.

It's included within Applio colab. (but you won't be able to see anything on TB until you actually start training a model)

There you have more info regarding training/tensorboard:

https://docs.aihub.gg/rvc/resources/training/

[river gulch]

I was just reading it. Still it doesn't clarify if stopping the model generation will cause it to finish it or fix the low bar

[odd shale]

Do you mean stopping model's training?

[odd shale]

But in case of any confusion, there you have a guide.

[river gulch]

Yes, because it says it is used for see when it reach the overtraining, so I kinda though it will give you a "stop it there" warning for one to see? Sorry, I never used this method

[odd shale]

Nope, it won't give you any sort of warning about when to stop training.

You must figure when to stop training (for example: whenever the g/total graph goes up again after a while)

[river gulch]

Wait, so it wouldn't give "it will take this xxxxx amount of time" or stop on itself when is done?

[odd shale]

Nope.

That's not how tensorboard/training works

[river gulch]

Good to know. So in resume I have to put to train, see the bar how gets low and when it reach once again the top tier I have to stop the training, no?

[simple ore]

have you read the doc? there's no bar

there are different charts for losses, the supposed to go down and converge/stabilize at some level

once that is done the model is ready, if you keep training some charts may go up (or may not) so the model may end up screwed up as it forget the pretrain information

[wary oxide]

weights its saying i have trains avaliable but its saying no files found

its still doing it

[analog finch]

hi where can i get the voice model

[eternal aurora]

Mb I was supposed
To show u but I was on my way back from the gym 😂😂

[low shard]

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @earnest musk
• https://weights.com/ (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.aihub.gg/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

[eternal aurora]

I’m sorry

[distant turtle]

-colab

[patent trellis]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[fading barn]

Any easy way to lower the audio quality output when using okada? It's almost too clear and I want it to sound a bit worse if that makes sense. I'm currently on linux

[eternal aurora]

@low shard so i got my audio all ready

it just doesnt let me select a voice model

[lunar bone]

I need help with Image generation

[echo grail]

How does Voyages compare to Midjourney

[trim apex]

Hello, what gpu is good recommend atm for voice changer ai? i have rtx 2060 but it feels rly bad for stream

[knotty moth]

it is still better than GTX gpus, but you might want a 4060 for streaming+vtubing

[trim apex]

ty
what can u say bout 5060-5070?

idk what should i choose and what settings are best for this

[knotty moth]

if 5060 Ti 16 GB is not more expensive than 4060 Ti or far above MSRP, sure

[analog obsidian]

if u want to stream with this buy a cpu with a gpu

[knotty moth]

note that we can't guarantee the voice changer for 50-series is stable enough

[trim apex]

for example?

[knotty moth]

though we've got another person getting it work #1367649960685862952 message

[analog obsidian]

i have a i7 13700k
so i can stream in high quality without stressing my main gpu

qsv = intel gpu

nvenc uses ur main gpu so it may stress the gpu too much if you are running the voice changer

[knotty moth]

the gpu should be capable to do AV1 streaming

[analog obsidian]

^

saves you resources and allows for smaller chunks

and saving you from cracking sounds

[trim apex]

sounds good, thx a lot

[knotty moth]

adding more cpu usage could bottleneck the performance

[analog obsidian]

its dumb but it can be a real issue in cpu heavy games lols

at least i can say my i7 is enough for this, never had such problem

[knotty moth]

the last gen i7 can be > older i9

given the same P-core count

[visual tendon]

So I was right all along... thx for confirming bro👍

[hallow thistle]

Minecraft

[elder coral]

How do i fill the epoch with my weights voice model

[noble wasp]

does this app can be used using AMD?

[knotty moth]

RVC? voice changer? stable diffusion?

[noble wasp]

voice changer

[knotty moth]

get the AMD version here https://rentry.co/forkvoicechangerguide#download-for-amd-intel-and-cpu-on-windows

[noble wasp]

thanks!!

[muted coral]

so my friends have the dml files for the correct ai voice changer but for some reason the program isnt letting them run on their gpu it keeps it on the cpu even after all the changes

[knotty moth]

make sure the GPU is supported

if it's an Nvidia gpu, should use the cuda version instead

[muted coral]

they're using a amd and so i at first thought that they had a nividia like me but they didnt so i had them redownload the other package and its all set up but they are still not getting any slear feed back, shits all choppy as if they're still running on the cpu. they clicked on all of the gpu options 0-3 and none of them are the gpu. i also had them check in task manager which one their gpu was on and it says 0 but nothing works on 0 either so im at a loss

[knotty moth]

the fork version is strongly recommended for AMD gpus, which has GPU dropdown instead of the dml button crap

it should detect the GPU, as long it's RX 580/5000-series or newer

[muted coral]

They have the 6600

[muted coral]

Do you have the package name for the fork version? Im not to fluent with how everything is labeled yet

[viral ivy]

when monitoring myself on wokada im noticing my mic getting quieter?

is that normal

recording myself 3rd party seems to be fine but I can't figure out why it suddenly gets so quiet

through monitoring

[knotty moth]

you should find it there #✨│ai-help message

[worthy fox]

i mean that i record an audio with headset, the voice is clear, but after using the voice clone, it have some effects like an echo.

[knotty moth]

it is perhaps a post processing effect in the process, check the settings

[sweet cave]

Hi yall, I use a voice from this server and it works pretty good except for in vr chat it’s horrible. If anyone knows what I’m talking about or has any solutions in mind lmk

Thank you!

[knotty moth]

explain how horrible is it

[sweet cave]

Hard for me to since I haven’t heard it but my partner says as soon as step into vr chat it goes so wonky

I’ve tried adjusting my mic input setting multiple ways and it’s weird and tricky

[knotty moth]

could you record it and see if so?

[molten sentinel]

maybe you need to give it a bigger window to process audio?
idk what kinda application you're using

[knotty moth]

how about in discord or another game?

[sweet cave]

Discord is good and I didn’t get and I don’t use game chat so I haven’t tried

[sweet cave]

It was a free one

[hallow thistle]

VRChat uses your PC GPU. That's where it makes W-Okada to lag.

[knotty moth]

if the output voice cut out, it was lagging due to not enough gpu performance or that you need to adjust chunk & extra

[sweet cave]

Probably my gpu

I have a nvida 4070 tho

[knotty moth]

note that the VR mode may be demanding

[sweet cave]

Do u think the model is too demanding and if I find a smaller one it make work better

[knotty moth]

I don't think the voice changer performance vary on different models

unless perhaps 32k vs 48k models

slightly in the memory usage

[sweet cave]

I see

[spice thorn]

tqdm and gradio is installed but anyways i getting this shit

[low shard]

elaborate:

• your pc gpu
• what you want to make
• what tut link are u using

[spice thorn]

2060
for now start rvc
https://github.com/Eddycrack864/RVC-AI-Cover-Maker-UI u mean this?

[low shard]

2060

for now start rvc
btw, the project you mentioned is made specifically for automating the project of making ai covers, you're looking for that?

if so, tell what you did step by step

[spice thorn]

yep but i also would like to train my own

pip install -r requirements.txt

run bat and got error

installed other modules manually

again run and error again

[knotty moth]

speak English or use a translator

[low shard]

the project you said isn't meant for that, it's meant only for automating the ai covers making process

else you could use Applio, an RVC fork, modified version, that can do both inference (use models) and training, but it won't automate the process of making ai covers, you will have to manually separate the vocals and instrumentals

or do you need both tools?

@granite dew
this server is english only, be respectful to everyone here

[spice thorn]

applio is local?

[low shard]

if you don't be respectful, we will take an action

[low shard]

yes, it's open source

soo, do you rather use applio? or both projects?

[spice thorn]

i can separate vocals so i can use this too

i just want generate covers and make my models

https://github.com/Render-AI/Applio-RVC-Fork this?

[low shard]

no

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

[knotty moth]

that unofficial fork seems sus, if it ends up having some issues

[low shard]

it's 2 year old

[spice thorn]

alr, if i install this what needed to make model? just audio sample of clear voice?

[fading barn]

Anyone use okada on linux? I'm struggling to set up a virtual mic to output the changed audio

[indigo saffron]

Could someone help me with some img2vid? Trying to make something for fun but im not really getting what i want

[simple ore]

there are many img2vid, which one you're trying to use?

[indigo saffron]

Tried to use googles gemini, read a couple times its the best free one

[simple ore]

there are plenty of options for local as long as your PC has enough memory and a decent GPU

[indigo saffron]

I think it does, what are some options?

[simple ore]

newest one if FramePack, based on Hunyuan, can generate basically unlimited length videos

ideally you want 48GB RAM and 12GB VRAM

but with lower resolution it can be used with 32/8GB

[indigo saffron]

Tysm, appreciate you

[low shard]

https://docs.aihub.gg/essentials/how-to-make-voice-models/

[spice thorn]

thx\

[low shard]

https://docs.aihub.gg/w-okada/local/w-okada/#linux

[low shard]

yw and lmk

[fading barn]

That doesn't say anything, I have it installed already

[low shard]

can you elaborate exactly what's the issue then

[fading barn]

I'm not able to output from the program into other programs without a virtual audio cable, The deiteris guide doesn't elaborate on how to outside of google it, which I haven't been able to find anything on how to set it up

[spice thorn]

Please run 'run-install.bat' first to set up the environment.

[low shard]

where do you see that phrase?

[spice thorn]

in console when i tried to open run applio

[low shard]

delete the folder, re-download the Windows 3.2.9 precompiled (which is the latest version right now), then, be sure to use Winrar/Peazip/7-zip to extract the .zip file, and run the run-applio.bat

[spice thorn]

its exatly same as i installed

[low shard]

are you sure that you extracted the same version with another tool rather than window's default extraction method?

[spice thorn]

¯_(ツ)_/¯ i try again with winrar

[low shard]

also, where are you exactly extracting it?

[spice thorn]

?

[low shard]

what is the full absolute path you're extracting the .zip at?

[spice thorn]

E:\ApplioV3.2.9

i got the same thing

[low shard]

could you try extracting to the C drive?

[spice thorn]

same

oh

wait

it worked

[low shard]

so it works only on the C drive? or was it related to smt else?

[eternal aurora]

@low shard so i got my audio all ready

and its not letting me select a model on the interface page of applio

[spice thorn]

maybe it because of start as administrator

i have a problem on my pc that most of things wont work good if not launched as admin

Applio is auto or i need separate first?

[low shard]

weird? so it worked only after running it as admin?

[spice thorn]

no

[low shard]

@simple ore have u heard others running in the same issue?
Edit: nvm

[spice thorn]

as normal

[low shard]

oh lmao

[spice thorn]

as admin not working

[low shard]

yeah it's said to not run it as admin in the docs lol

[low shard]

show a screenshot

elaborate more

[spice thorn]

im lazy ass xd

[low shard]

it's not auto

[spice thorn]

so i need to put vocals only?

[low shard]

yes

[eternal aurora]

i uploaded the audio

and i cant select a model

[low shard]

refresh

click the voice model dropdown menu

[spice thorn]

@low shard lowest_value, lower = better?

[eternal aurora]

still nothing

would this be helpful at all?

[spice thorn]

Why don't just drop files?

[eternal aurora]

thats what im about to do

[dry jewel]

hello guys

i'm having issues with w-okada real-time voice changer

it has robotic glitches, is there a way to make it sound like real voice?

[eternal aurora]

@low shard what should these vlaues be

[sharp marlin]

Are there any tools like RVC available on Linux? I wanna make AI covers without shelling out money or using Michaelshaft Winblows.

[mystic orbit]

just type 50k lines of code and you might get it working

[sharp marlin]

Hardy har har.

Seriously. Is there anything out there?

[mystic orbit]

Don't know. I'm on Win10

[sharp marlin]

Then why answer my question?

:/

[low shard]

elaborate:

• what's ur pc gpu
• what tutorial link did u use
• what u want to do
• a screenshot of the program

[low shard]

https://docs.aihub.gg/rvc/resources/training/#batch-size, https://docs.aihub.gg/rvc/resources/training/#epochs--overtraining

[low shard]

what's your pc gpu?

damn never heard of this parody

[eternal aurora]

ty

@low shard wait when do i need to use tensor?

[low shard]

tensorboard? you have to always use it when training

[eternal aurora]

oh okay, but i havent started the training bit yet

[low shard]

you can just set the training to an epoch like 500, then monitor the training process in the tensorboard to see if u need to stop earlier

[eternal aurora]

yes i just done that thx

how do i set up tensorboard?

[low shard]

you're on cloud right?

[eternal aurora]

yh

[low shard]

wait for it to train a bit, open the cloud url link, select scalars, then click the gear icon and select Reload data

[reef flax]

https://huggingface.co/Shadicti/deiteris-Fork/blob/main/voice-changer-windows-nvidia-b2332.zip I had this version of W-okada and it ran well until it suddenly just delayed a lot, i got like 2 normal versions of W-okada and it sounds a lot worse than the voice used to be. and its glitching, what can i do?

[eternal aurora]

okay ill let u know how it goes

[low shard]

elaborate:

• what's ur pc gpu
• what program are u using wokada with
• show a screenshot of ur wokada

[reef flax]

4060 ti

[low shard]

nice, reply to the rest of the questions too

[reef flax]

program ? there was no program running other than discord.

and i only have on escreenshot of the w-okada

the link is the w-okada iw as using

i already deleted it

because it suddenyld idtn work anymore

i just need reccomendations

or a good program to get for w-okada

because i feel like im doing something wrong.

[eternal aurora]

tf it was done in like a minute....

[low shard]

yes, like discord, I thought you were running a game with wokada

[low shard]

that's the best wokada version as of right now, there was 100% something wrong with the settings, please reinstall the latest version of wokada deiteris fork (b2332, the one you sent), and show me the settinga along with the discord one

be sure you're using https://docs.aihub.gg/w-okada/local/w-okada/

and aren't using vb audio cable

[reef flax]

im using vb audio cable

definitely.

[low shard]

lemme guess, you firstly used a youtube tutorial that suggested you that, then went here, directly got the new wokada download link?

[reef flax]

im downloading it again right now, it's weird how it started delaying like crazy

[reef flax]

no, i bought a model,.

the person who i bought it from reccomended it to me

[low shard]

vb audio cable has been reported by very many users here to cause issues of it not working for windows

[low shard]

who told you that?

[reef flax]

oh i was talking about the w-okada

not the vb audio cable

[reef flax]

vb audio cable i was using that yeah

is it ass?

😭

[low shard]

vb audio cable is a common VAC (Virtual Audio Cable) known for having issues on windows, so yeah as many others said it will give you issues out of no where lol

[reef flax]

cause now im using the virtual audio cable

that is in the site.

im downloading it rn

the link you sent.

[reef flax]

oh yeah okay makes sense man.

i didnt know that.

[low shard]

this right?

[reef flax]

yeah

and im sure a lot of people didnt know that?

[low shard]

ohh, I thought the dude you bought the model off told you to use vb audio cable lol

[reef flax]

nahnah

poor guy didnt know

haha

im doing this right now and ill let you know

[low shard]

at first when original wokada and RVC AI Covers were trendy alot of people used it, it's been months that it's not suggested anymore (prob like half a year)

[low shard]

alright

[reef flax]

i see

its weird how that ver got laggy somehow

outof nowhere.

[low shard]

which is the reason I thought you used a youtube tutorial, they suggest an old version of wokada with vb audio cable because that was trendy like 2 years ago

[reef flax]

makes sense yeah and ive seen those tutorials are like 2 years old etc.

[low shard]

dw, I will also help you with the settings, prob some wrong settings + vb audio cable fucked it up

[reef flax]

carrying me man

😭🙏

[low shard]

save every epoch should have been a bit lower for accuracy like 10

also, check the output in the colab/kaggle

[eternal aurora]

ohhh shi let me just fix thst

[eternal aurora]

apparently not enough data was present

[delicate whale]

Whats the best settings to put in voice changer native client?

[low shard]

elaborate

[reef flax]

right so i downloaded everything

[low shard]

• what's your pc gpu
• what tutorial link did u use
• what are you going to use the program with
• show a screenshot of the program settings

[low shard]

you forgot to slice audio files in the preprocess part

[low shard]

alright, want me to check your settings ?

[reef flax]

wait let me do the uh new audio thing.

[eternal aurora]

oh the audio cutting part? i wasnt too sure if it should be on simple or auto

[reef flax]

i might sound dumb, is this wrong or right?

[low shard]

it's right, you can also optionally set monitor to ur headphones to hear urself

[reef flax]

i wanna hear like the voicechanger.

do i have to change it to my headphones then?

[simple walrus]

I know how to create vocal models. I just wanted to know if anyone has a finished vocal model from Turkish star Derin Talu.

[reef flax]

i just set it to my headphones and it worked.

[low shard]

yup

[low shard]

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @earnest musk
• https://weights.com/ (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.aihub.gg/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

[low shard]

yes

[reef flax]

you fixed the delay issue for me, thank you! , as of right now these are my settings, and its well. im not complaining, but do you see any issue further just to be sure?

[delicate whale]

So basically when I talk It doesnt sound crispy its pretty choppy sound

[low shard]

elaborate the questions I asked you: #✨│ai-help message

[eternal aurora]

should it be simple or auto?

[low shard]

f0: rmvpe without onnx
extra: 2.7
chunk: 192ms

you can optionally set force fp32 mode on for better quality and stability, but more delay

you can also optionally use https://docs.aihub.gg/w-okada/local/w-okada/#reduce-more-delay

[low shard]

did you already manually slice your audios? if not, set it to simple if u deleted atleast the silence, or if it still has silence set it to automatic

@delicate whale i saw u deleted the message, if you do shift+enter you can make a new line on discord btw

like
this

telling ya since i thought that's what you wanted to do

[eternal aurora]

naa i havent sliced it i put the audio into audacity thats it

[low shard]

did u remove the silence too?

[eternal aurora]

truncate silence?

yes

[low shard]

then set audio cutting to simple

[eternal aurora]

okay

[acoustic thunder]

What is the best tool to make high quality voice conversion (not in relatime). Like upload, wait for processing and have quality output. Would be there any difference between real time conversion in w-okada and not real time

[eternal aurora]

it seems like its working

[low shard]

RVC, more specifically Applio since it's a fork that gets maintained and has performance improvements

what's your pc gpu?

[analog obsidian]

local rvc

[low shard]

I mean, they can also use cloud if they don't have a good pc gpu

[delicate whale]

1. PC GPU= Geforce RTX 2060 2. Tutorial https://youtu.be/dZ_2HELnWJU?si=4xXWwmYt_gqgS-Zj 3. Im going to use it for discord talking 4. I Will give the screen shot right now

[analog obsidian]

[low shard]

the program (and code) is the same

[acoustic thunder]

rtx 4060

[low shard]

youtube tutorials are outdated, that's the issue

@delicate whale i just checked the tutorial shortly, and it uses an old version of original wokada

please, just forget everything you got off that tutorial, 1 year old is prehistorical in ai field

uninstall everything out of it

-realtime

[patent trellis]

💻 Local Realtime RVC

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

• Wokada Deiteris Fork

Most suggested. GUIDE

• Original Wokada

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

• **RVC GUI Mainline Realtime**

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

[low shard]

use the 1st link

[analog obsidian]

rvc = retrieval based voice conversion

[low shard]

wokada deiteris fork is the best rn

[delicate whale]

So you would like me to delete the voice changer app itself right now?

[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

[acoustic thunder]

thank you!

[low shard]

yes, all video tutorials are outdated, that video is using a year old version, AI progresses at sonic speed

[acoustic thunder]

what is there most quality TTS nowadays? With emotion control and etc. All i hear is F5 TTS and FISH

[low shard]

https://docs.aihub.gg/tts/tts-tools might help

[acoustic thunder]

thank you!

[delicate whale]

I just deleted it what do you recommend me to do now?

[eternal aurora]

@low shard apparently its done?

[acoustic thunder]

i wonder how are you not lazy to answer all the reapetitive questions from stupid(like me) users

[low shard]

to read the 1st written guide link the bot just sent after I wrote "-realtime", the wokada deiteris fork

there's no updated video tutorials

[analog obsidian]

can we have a pinned message that explains rvc doesn't mean realtime voice changer

[dry jewel]

gtx 1660ti (laptop)
https://www.youtube.com/watch?v=_JXbvSTGPoo (don't mind title)
i want it to sound like real audio with good enough quality (it's scratchy even in the video introduction in the link)

edit:
it seems like some voices work better than others

[delicate whale]

Could you send me the link I dont see it

[cloud finch]

is this W okada?

[crude flame]

I second this

[dry jewel]

yes

[reef flax]

damn u carried, thanks man.

[tulip cloak]

Technically I did

[cloud finch]

https://huggingface.co/Shadicti/deiteris-Fork/blob/main/voice-changer-windows-nvidia-b2332.zip

is this the right W okada link

[delicate whale]

How long does it take once the MMVServerSIO finishes downloading?

[cloud finch]

Every time I extract it I get a checksum error saying "corrupted archive"

[analog obsidian]

depends in how fast is your internet
for me it takes like 50 seconds

[analog obsidian]

uh try reinstalling 7-zip/winrar

[cloud finch]

Are you sure? cause its with one specific file (torch_cuda)

and it doesnt happen when extracting anything else

[analog obsidian]

idk

u could download the thing from github

[cloud finch]

I have a nvidia

I only see AMD options on there

unless its fine to get one of those???

[analog obsidian]

amd64 = 64 bit cpu

😭

[cloud finch]

I can get that even though I have a nvidia?

?

[analog obsidian]

read the file name

cuda = nvidia

[eternal aurora]

@low shard how comes my graph isnt showing

[cloud finch]

Oh

[analog obsidian]

amd64 = x64 bit cpu

[cloud finch]

this is not common knowlege nerd boy /j

[analog obsidian]

🤓 ☝️

[cloud finch]

wait so

do i get 001 and 002

or just 001

[analog obsidian]

read the github readme

https://github.com/deiteris/voice-changer?tab=readme-ov-file#for-amdintelcpu-users

[cloud finch]

thanks

[analog obsidian]

🔥

[cloud finch]

weird the docs give a bad link

[analog obsidian]

i havent never tried that link idk

[low shard]

https://docs.aihub.gg/w-okada/local/w-okada/

[analog obsidian]

its the same program but they merged both zip files

[low shard]

this was the link given by the bot

[cloud finch]

interesting

[low shard]

do not follow video tutorials for AI lol

you got the latest version of original wokada, but it has worse performance than the latest wokada deiteris fork, also you prob got vb audio cable that gives issues

[eternal aurora]

now why is there no graph

[low shard]

@dry jewel uninsstall everything you got off youtube

get wokada deiteris fork https://docs.aihub.gg/w-okada/local/w-okada/

[eternal aurora]

why tf is there 4 graphs

[low shard]

you're welcome

[eternal aurora]

forsome reason i generated index before training my model tf

[delicate whale]

Is it okay if i follow this tutorial? https://www.youtube.com/watch?v=VAtnHB9Y9UA&ab_channel=Mourn

It says to download the 3.24GB one and I think it might be new i dont know

[low shard]

no, i just told you, video tutorials are outdated

that video tutorial uses original wokada too

read the written guide

[analog obsidian]

there's not even a single thing correct in that tutorial

[delicate whale]

Once I click the link which link is for windowas?

[low shard]

read the tutorial

it explains you

the only way to get the correct program is by reading the guide

[eternal aurora]

@low shard does this look correct?

[delicate whale]

There isnt a Guide when I click the link

[low shard]

https://docs.aihub.gg/w-okada/local/w-okada/ is a guide

you have to read it

[low shard]

the training just started, also you have to look at g/total

[cloud finch]

BRO

WHYY

😭

[analog obsidian]

reinstall 7zip or winrar bro

lol

something crazy going on in your system

[delicate whale]

Could you please send the download link of it here, I dont see anywhere where it says Windows download I only see Mac Download

[cloud finch]

Do you think its cause im extracting it in the E drive?

[analog obsidian]

try drive c

not a onedrive folder

[low shard]

https://docs.aihub.gg/w-okada/local/w-okada/ it shows a Table Of Contents at the Right Part

[cloud finch]

aight

[low shard]

you have to get the Vac Lite at the Virtual Audio Cable step, then downlaod the nvidia version for windows

[analog obsidian]

users > name here > voice-changer

[low shard]

all of your questions are probably already answered in the written guide

[eternal aurora]

oh okay why do i need to look there for>?

[cloud finch]

oh just straight in the root eh?

alright

[analog obsidian]

can be anywhere tbh

as long it doesn't say onedrive

[cloud finch]

I think i removed one drive

hate that shit

[low shard]

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

[delicate whale]

Do I download this one

Download AMD, INTEL and CPU on Windows
The lastest version as of December 7th 2024 is: dml-b2332 (click here to download)
Intel UHD Graphics do NOT work at this point in time. Use Online Alternative.

[analog obsidian]

i forgot to remove onedrive when i installed ts 😭 😭 😭

[low shard]

I just told you get the nvidia windows version

you got an RTX 2060, which is made by Nvidia

it's an nvidia gpu

[rain axle]

For anyone on Win11 who's using Edge as their browser to have the voice changer interface open:
Microsoft did some kind of update and it completely fucked the voice changer for some reason. I don't know why. It introduced a huge amount of delay and I was going crazy trying to figure it out.

Using Chrome works just fine. I have no idea what they did and why it would do that.

The update happened some time today.

[cloud finch]

GRAAAAH

im reinstalling midrar

[analog obsidian]

hold on

don't make a folder with spaces

just name it voicechanger

lol

[cloud finch]

alright

[analog obsidian]

if that doesn't work yw what to do next

[cloud finch]

[delicate whale]

There doesnt say anything about Nvidia Windows Version, It only says

Download NVIDIA RTX 5000-series on Windows
NVIDIA RTX-5000 series, the newest release of GPU's, require a separate download. You do not need it if you have an older GPU, follow the normal Nvidia link in that case. nvidia-5000-Series (click here to download)

[low shard]

that's for the rtx 50 serie, you got an rtx 2060.

https://docs.aihub.gg/w-okada/local/w-okada/#download-nvidia-on-windows it's this one for you

[eternal aurora]

ty

oms

omd

my shit just crashed

[low shard]

@delicate whale don't skip steps on the guide, be sure you download vac lite too

[eternal aurora]

wtf why is my tensor still running

[low shard]

please have a careful reading of the whole guide, skipping steps (except in cases of the step being related to another OS or GPU) can cause issues

[low shard]

wdym

[delicate whale]

Is there anyway I could get you on the VC call very fast

[low shard]

I can't VC

what's the matter?

[delicate whale]

You dont have to talk Im going to share my screen

[low shard]

join support one

[eternal aurora]

my browser crashed rn and my applio just shut down however my tensorboard is still running

[low shard]

@delicate whale you have to download only the normal NVIDIA on windows version

[delicate whale]

so the first one

[low shard]

how did ur browser crash 😭

[cloud finch]

im gonna reinstall midrar.

[low shard]

yes, the 2nd one is only for RTX 50 serie users

so yes the 1st for you

[analog obsidian]

try 7zip

[low shard]

winrar goated

[delicate whale]

now it shows that

[low shard]

download

[delicate whale]

do I click download

[low shard]

yes

[eternal aurora]

i have like 70+ tabs open xddd what should i do now lmao, should i restart?

[low shard]

the training isn't affected because it runs on a remote good pc

[delicate whale]

Once Im finished downloading whats gonna happen and then?

[low shard]

also don't torture your browser with over 70+ tabs pls

[eternal aurora]

ohhh okay

[eternal aurora]

let me show u

[low shard]

then you have to extract it with winrar or 7-zip or pea-zip

[cloud finch]

i've been a d1 7zip hater since day one. If it works im finna be mad

[low shard]

whatever tool you use

[analog obsidian]

lmaoo

[delicate whale]

I have Winrar

[cloud finch]

pea zip 😭

[eternal aurora]

what should i do abt the applio part?

[low shard]

oh god. and I thought i had many tabs

[cloud finch]

BRO

[eternal aurora]

im a student lmao half of them are for revision n shit xddd

[low shard]

you should be able to just reopen the colab/kaggle and continue from where you where, firstly close some tabs

[analog obsidian]

okay we need a better nerd than me to know how to fix this

[cloud finch]

AUAAUAUAUU

[low shard]

then use that to extract

[delicate whale]

and then thats it?

[eternal aurora]

yeah ngl some of these tabs have been here since deceber

[tough quiver]

how do i chat to the ai

[low shard]

why are u using the github one?

[cloud finch]

because the doc one gave the same error

[analog obsidian]

because the huggingface one havent worked

lol

[low shard]

[cloud finch]

can i install like

[low shard]

did you extract it? i didn't see you do that

[cloud finch]

the dependency seperately

since its only one

thats throwing an error?

[delicate whale]

I got 1 minute till its done downloading

[low shard]

even with 7-zip?

[cloud finch]

yup

cusolver64_11.dll

[low shard]

can you try makin your own zip for a sec and extract it

[cloud finch]

for the voice software or

any

[low shard]

just any to test if ur extracter is actually extracting

[delicate whale]

im extracting

[low shard]

alright, meanwhile, did you also download vac lite?

[cloud finch]

it worked

waity idk why i sent that lol

[low shard]

could you also extract it yourself?

[eternal aurora]

let me go onto collab

[low shard]

also yeah why did u send it lol

[delicate whale]

no

could you send me the download link

[low shard]

A VAC (Virtual Audio Cable) makes a fake audio device, used to re-route the audio of different programs

In Wokada context, it's used to get the output of wokada as the input in other programs

Vac Lite is the suggested one for windows: https://docs.aihub.gg/w-okada/local/w-okada/#virtual-audio-cable

[analog obsidian]

ive read that if ur getting those weird checksum errors in both 7zip and winrar it could mean your ram is dying

[eternal aurora]

@low shard uhm i see nothing?

[cloud finch]

yes

[low shard]

could you try maybe some big files?

[cloud finch]

Its weird though since this is the only thing I've ever gotte nthat on

[cloud finch]

sure

how big

like 1g?

[low shard]

whatever biggest file you got

[delicate whale]

I jsut downloaded it

[cloud finch]

ill pack up a 7gb one

its only 2 feils though

files*

[low shard]

extract that too

[delicate whale]

ok

[low shard]

now just open the extracted vac lite, run setup64.exe, and select yes when asked

[cloud finch]

Its still packing but, have you ever extracted W Okada before and it worked?

[vast elm]

yo@low shard, wassup. just wanted to remind you that your amazing <33

[low shard]

@delicate whale you should have selected extract in it's own folder, don't extract everything into a big folder

[low shard]

thank you

[low shard]

yeah lol, a user literally extracted wokada infront of me in vc rn

[cloud finch]

WHAT THE FU

oh my lord

[analog obsidian]

thank god i read that twice

lol

[low shard]

i have a feeling there's something wrong extracting any big files for you

[low shard]

wdym

[analog obsidian]

first i read "a user extracted infront of me in vc"

😭

anyway nick check this

fumiama realtime gpu usage

[delicate whale]

I only see setup64 check my screen and im on the VAC folder

[low shard]

smh the unstable realtime program

[analog obsidian]

chatgpt goated for fixing every dependency for me

[low shard]

yes that's what i mean, .exe is just the file extension, every file has an extension like .mp4 or .mp3

you just didn't enable to show extensions in file explorer, anyways you don't have to, just click setup64

[eternal aurora]

@I used it to help me find my data folder

[low shard]

for you not even changing system locale could have helped me

[eternal aurora]

@low shard still nothing in applio 😦

[low shard]

are u using same account? you should check the colab output and tensorboard

not the applio ui

@delicate whale click yes

@delicate whale now, just select your usual devices for playback and recording as default ones

find playback usual device

set default

and do the same for recording tab

did you do the above that i said?

[delicate whale]

which one

[low shard]

#✨│ai-help message

did u do this part

[delicate whale]

I just selected ok and thats it

[eternal aurora]

ywah i should be however its been stuck on 56% for a while

[low shard]

oh, that's bad

click windows+r, type "mmsys.cpl", Enter

then select your usual devices as default one

it's a must for it to work

[low shard]

it says draft session stopping.. seems like you stayed without checking it for too long and it went afk and stopped ig, unless you stopped it manually

well shortly, rip your training

@delicate whale i can't know which speaker is your default one, you need to check the name, and if it's your usual device, click set default

[cloud finch]

i extracted a 5.71 GB file with no problems

[delicate whale]

I use a microphone and headset

[eternal aurora]

ah great time to resart it seems, how do i like check it? when im doibg iy

[delicate whale]

by speaker you mean the output audio right?

[low shard]

wtf? i'm gonna try to the extact exactly the windows amd one just to check again

[low shard]

yes
playback is output
recording is input

[low shard]

check what? the training stopped

you have to restart

[analog obsidian]

you've never had problems unzipping files before?

[delicate whale]

I have selected my output audio

[cloud finch]

https://github.com/deiteris/voice-changer?tab=readme-ov-file#for-amdintelcpu-users where i got it

though you probably already know

[low shard]

I just extracted the huggingface one, works all fine

[cloud finch]

bro

HOW

PMO

[analog obsidian]

do a ram test

lol

[cloud finch]

I mean I guess but like its literally only this weird ass file

[eternal aurora]

Okay, how do i stop tensorboard?

[cloud finch]

other big files extract perfectly fine

[low shard]

everything stopped, the whole session stopped

[delicate whale]

I picked the output

[eternal aurora]

oh right

[low shard]

did u do also the same for the recording tab?

u had to pick your input here too

[delicate whale]

ok

yes its selected

[low shard]

the default one?

[delicate whale]

yeah

[low shard]

like your usual device?

alr

now just go to the MMVCServerSIO folder

then open the MMVCServerSIO file

[delicate whale]

just did it how long would this take

[low shard]

depends on ur internet

it's downloading internal files

your internet is downloading in kilobytes

[delicate whale]

thats slow?

[eternal aurora]

pls tell me its near 1k kb not 100 kb

[low shard]

by alot

it's taking over 3 mins for some megabytes files 😭

[delicate whale]

holy shii, its cause im upstairs and thats why

[low shard]

300 kBs

[delicate whale]

let me open the door so its faster

[eternal aurora]

holy shit

i thought mine was bad

[low shard]

rip

[delicate whale]

LMAOO

it says enter to continue what does that mean?

[low shard]

it means that your internet was so extremely slow that it was slower than the program timeout error time for downloading 😭

basically it took so much time that it went pass the timeout time

[delicate whale]

do I redo it?

cause it went away

[low shard]

either redo it, or just follow https://docs.aihub.gg/w-okada/local/w-okada/#help

you will basically download the internal files from google drive, so it won't have a timeout

[delicate whale]

I opened the door it should be faster

[low shard]

@delicate whale btw when i said redo it, i meant redo it hoping that it won't be so slow that it will timeout again

[low shard]

slightly but yes

[delicate whale]

thank god

[low shard]

maybe it's just better to follow that guide part if you still get time out lol

[cloud finch]

ending it all

its over

jumping into the sun

[low shard]

what if i try to share u my extracted wokada amd

[delicate whale]

okay if I get it one more time Im doing it

[cloud finch]

that would be appreciated

I wonder if that would work

[low shard]

gimmie 2 mins

[analog obsidian]

but he has an nvidia gpu

[low shard]

@cloud finch what's ur pc gpu

[cloud finch]

RTX 3060

NVIDIA

[low shard]

😭

[analog obsidian]

[cloud finch]

💀

[low shard]

lemme redo the job for the nvidia one

[cloud finch]

I love you

[low shard]

@delicate whale did u place the downloaded internal files from google drive to the internal folder in the program?

no need to retry

your wifi is way too slow, you will always get a timeout

[delicate whale]

Do i extract the file there and then

[low shard]

nowdays having kBs is too slow

[low shard]

copy paste all files in the pretrain folder in the program

[delicate whale]

I think I have to download those files first because it wont paste it their

[low shard]

i meant the pretrain folder

[delicate whale]

I deleted all the files that were orignally in the pretrain folder

[eternal aurora]

@low shard why do i keep on getting this error

oh wait

i can make one

[low shard]

download the files, extract them into the pretrain folder

[delicate whale]

okie

[delicate whale]

I click extract here and then on the pretrain folder right

[eternal aurora]

yh im dumb

[cloud finch]

You still uploading the files?

[low shard]

you gotta extract them into the pretrain folder

[delicate whale]

ok

[low shard]

the nvidia version is bigger than amd's one

[cloud finch]

Oh really?

I see

tysm

[low shard]

so it's still downloading

[cloud finch]

Ohhh

I gotcha

[low shard]

i really hope it will work

[cloud finch]

If not then

its definitely my ram

[cloud finch]

Sorta unrelated but I tried download the AMD version out of curiosity and it extracted

why does nvidia hate me

[low shard]

because you didn't buy 99 pairs of the rtx 9090 ti super 💔

jokes aside

btw is ur windows updated and drivers too?

[cloud finch]

I updated my drivers about 2 weeks ago and windows is updated

[low shard]

I updated my drivers about 2 weeks ago
then they need to be updated

latest driver was 30th april

[cloud finch]

alright

I guess ill update them

[low shard]

nvidia releasing drivers weekly bc mostly are rtx 50 serie fixes, but there are also other fixes that might help for all other series

@delicate whale so?

[eternal aurora]

@low shard ah crap how do i fix this

[delicate whale]

could you tell me what you mean by that

[low shard]

it wasn't referred to u

im asking u how things are going

[low shard]

what's ur dataset lenght and batch size

[delicate whale]

its going good files are almost downloaded

[eternal aurora]

24 mins and 547 mb the audio size

it wasnt do this before but now its doing this

[delicate whale]

im done extracting the files

[low shard]

show

[low shard]

what's the batch size

[eternal aurora]

oh 15

[low shard]

@delicate whale now click the mmvcserversio file

[low shard]

that's why

15 is too much

[eternal aurora]

too high?

[low shard]

and also makes it use more vram which they don't have access to

[low shard]

keep it at 8

[eternal aurora]

ohhh

[delicate whale]

where

[eternal aurora]

okay ty

[low shard]

in the MMVCServerSIO folder

not the pretrain folder

[eternal aurora]

should i redo the index file or na>

[low shard]

it's better you retrain

[eternal aurora]

yeah thats what im gonna do

would u know roughly how long it will take to train?

it was reaching its 30 min mark last time

[delicate whale]

I started it and it opened it on a browser

[low shard]

@delicate whale set gpu to ur rtx 2060
set extra to 2.7
set chunk to 192
set f0 to rmvpe without onnx
set input to microphone
set output to line 1
set headphones optionally to hear urself

[low shard]

yeah that's completely normal

it runs on ur hardware

it just uses a web user interface locally hosted

[delicate whale]

So when ever in the future I wanna use it I open MMVCServerSIO folder

[low shard]

yes

use the settings i told u

[delicate whale]

ok

[low shard]

@delicate whale u gotta upload a model first

[cloud finch]

updated drivers, still doesn't work

[delicate whale]

I have a model downloaded where do I put the model file in my files at?

[cloud finch]

Did the guy who extracted it in vc do the AMD or NVIDIA one?

[low shard]

extract the model anywhere u want, click edit in the program, and upload it at any slot

[low shard]

nvidia

[cloud finch]

im cooked

[low shard]

i also got an nvidia gpu and extracted it on nvidia

i also was able to extract it on my integrated graphics laptop lmao

[cloud finch]

oh my lord

[low shard]

winrar so goated

[delicate whale]

I got a model which file do I put for which

[low shard]

.pth for the model part
.index for index

it depends on the file extension

[cloud finch]

I guess its my ram then??

[low shard]

@delicate whale if u can't see that, then just check https://www.youtube.com/watch?v=CvJn_8ZUdpk

[low shard]

i'm not sure, i'm still reuploading the file

i compressed it in a .rar

[cloud finch]

thank you

[simple ore]

I'd recommend 7-zip instead, but winrar is fine too

[cloud finch]

CAN IT

[delicate whale]

where do you set headphone to hear yourself?

[low shard]

even 7-zip doesn't work for Shy

[low shard]

in monitor option

[cloud finch]

https://tenor.com/view/midoriya-deku-boku-no-hero-gif-13174542

7 MID

[simple ore]

skill issue?

[cloud finch]

7ZIP has no soul

[low shard]

btw the speed and shi should be pretty similar between the 2, just 7 zip is open source while winrar is 'free'

[low shard]

why do u hate 7 zip btw

[cloud finch]

no soul

[analog obsidian]

this could be a hardware issue but we don't know

[cloud finch]

users think they better

Hate 7 zip users

hate 7zip

they all smell

[analog obsidian]

[low shard]

@delicate whale

i can't vc anymore i think

[delicate whale]

So IF i were to click start and record a video for example my voice would attomatically be in the video?

[low shard]

did u argue on reddit with someone about 7zip 😭

[cloud finch]

no?

[low shard]

yes

[delicate whale]

okay thank you very much bro

[low shard]

i wonder what type of argument u had with 7zip users bc i never had one

[cloud finch]

not much of an arguement just me shitting on it for no reason since i use winrar

[low shard]

btw are u going to use it only for pre-recorded audios?

[cloud finch]

also winrars red

so + points

red AND green

[low shard]

nvidia is green

[cloud finch]

the sky is blue too

like your FACE after I punch it...

[low shard]

we are on discord, we do NOT know what is the sky and grass 💔

[cloud finch]

fair point

[cloud finch]

-2 download speed 💔

[analog obsidian]

[low shard]

better than kBs tho

[cloud finch]

wasnt there a guy earlier with that 😭

[low shard]

yes, 300kBs

[eternal aurora]

@low shard how do i know that im done training?

[delicate whale]

probably yeah

[low shard]

wait then you got the wrong program

are you going to use it ONLY for pre-recorded videos?

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

Wokada = uses RVC for realtime inference

Wokada is meant for realtime like discord vc

or like youtube lives or whatever is meant for realtime, since it prioritizes speed over quality compared to rvc

[delicate whale]

So what should I do lol

what if i use it for Twitch Lives?

[low shard]

then yes wokada

[delicate whale]

And for PreRecording what would that be

[low shard]

i mean, if you're going to record youtube videos, you could set obs input to line 1 for example but this isn't really suggested, the best is you get applio (rvc fork for better performance) https://docs.aihub.gg/rvc/local/applio/

[delicate whale]

Can I have both of them installed or I have to delete one?

[low shard]

u can have both of them installed

read the guide and you will be fine

[eternal aurora]

how do ik when its done

[delicate whale]

This bout to take a long time LMAO

[simple ore]

not the charts you need to look at

[eternal aurora]

these ones?

[simple ore]

no

[eternal aurora]

oh

[simple ore]

also click on the square button under each chart to re-fit

[eternal aurora]

okay

how do i get to that?

[simple ore]

collapse the 'grad' line, expand the loss_avg_50 line

[eternal aurora]

aite

[simple ore]

dont type 'loss' in search

[eternal aurora]

okay

[simple ore]

just click the rows

[eternal aurora]

ywahh i figured

[delicate whale]

I clicked the link and then the part where it says download here and it sqays Linux, Windows and a File that says Gitignore

[eternal aurora]

@simple ore how do ik when im done?

[low shard]

you need to click the windows one

[simple ore]

change smoothing to like 0.5 or less

[delicate whale]

I clicked it and it now says Gitignore, AppolioV3.2.8-bugfix.zip. and another AppolioV3.2.9.zip

[simple ore]

AppolioV3.2.9.zip

[eternal aurora]

Done

[simple ore]

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

[delicate whale]

So what does the bug fix mean?

[simple ore]

previous version, dont bother with it

[eternal aurora]

How would the graph look when it’s done?

[delicate whale]

Ohh okayyy

[simple ore]

normally, fm loss should be going down or at least not up, kl usually goes down and flattens quickly, nek abd total loss take some time to converge to some value and flatten

[eternal aurora]

ahh okay its at its 30 mins mark

@simple ore like this?

[simple ore]

click this under each chart

you're at 750 steps = barely even started

[eternal aurora]

omg bruh

[delicate whale]

how do you execute a folder?

[delicate whale]

bro I downloaded it

what do I do now I ran the file but I guess my internet was slow again for Apollio

[low shard]

u extract it and run the run-applio.bat

[delicate whale]

yes

and it failed