#✨│ai-help
patent trellisBOT
golden yew
Did I Do It Right?
ember bay
how to download w-okada? github don't have anything
odd shale
how to download w-okada? github don't have anything
I can suggest you to download Deiteris' W-Okada fork, it's easy.
What's your GPU?
ember bay
What's your GPU?
rtx 3060
ember bay
Read the guide above 😄
thx
ember bay
You're welcome buddy.
hey, can you help em again? i'm using it in discord but don't have any sound
foggy maple
Hello, I used to create covers with AI voices, but I can´t find the link or a google doc that works 😦 can someone help me? pls
azure marshBOT
Hello, I used to create covers with AI voices, but I can´t find the link or a go...
simple ore
labeling in Audacity works reliably for a specific audio
and whatever parameters you have for one, likely wont work for another
so just trim the silence and let the application split the big file
cursive salmon
gotcha, ty
golden yew
I Added A New Voice Changer From Weights.GG, And There Was An Extra File In The Folders I Downloaded, So I Messed With The Output, Or File Type, I Believed One Of Them Was Like MMV18 Or Something (I Cant See It Anymore Because Of The White) Maybe It Was The Template, But Ever Since I Did That, The Client Is Now White
After I Changed That One Specific Voice, I Could Still Use The Client, But After I Clicked On It, It Now Only Shows White, After Exiting And Resetting Computer, Its Still Like This, Please Help
blazing crane
Trying to train thru Mangio-RVC-Fork and this error keeps popping up while using Ov2Super pretrain
Been trying to fix it for the past like 2 days, re-preprocess & feature extraction the file multiple times on different settings still doesnt work
simple ore
Trying to train thru Mangio-RVC-Fork and this error keeps popping up while using...
mouse over your cursor over Mangio folder and select delete
then install Applio
blazing crane
then install Applio
provide link and i will 🙏
blazing crane
thank you
simple ore
unzip into C:\Applio or other drive
blazing crane
unzip into C:\Applio or other drive
its able to train right? never heard of it
blazing crane
its able to train right? never heard of it
@simple ore
simple ore
as long as you prep the dataset properly
i've never ever seen the loader failing at the spectrogram size
blazing crane
i've never ever seen the loader failing at the spectrogram size
what am i doing wrong?
blazing crane
trying to install applio
simple ore
unzip it to C:\Applio
blazing crane
i did
simple ore
I dunno what've you've done
blazing crane
me neither bruh 😭
i literally just installed the file, made a folder named applio in C: and ran the run-applio.bat
blazing crane
ill retry
simple ore
use 7-zip
blazing crane
this is what happens, i unzipped and tried again
@simple ore
waited for like 5 min last time ill wait again
simple ore
weird
blazing crane
weird
also is there a way to use pretrain models on applio?
or could you link me a applio guide to training rvc? all the ones i found on youtube are mangio
simple ore
hallow thistle
unique rock
can i run applio on termux android?
hallow thistle
It is possible to run Applio Linux on Termux terminal on an Android phone, but I'm not sure where its GUI should appear., and it won't be that fast like on desktop.
glacial pollen
It is possible to run Applio Linux on Termux terminal on an Android phone, but I...
it won't work because, well.. how do you wanna run pytorch and do cuda operations on android arm
👀
knotty moth
should be in cpu mode, though will be extremely slow
low shard
Hello, I used to create covers with AI voices, but I can´t find the link or a go...
what’s ur pc gpu
low shard
it won't work because, well.. how do you wanna run pytorch and do cuda operation...
Actually you can inference even on a 2019 samsung phone as tested lol
slow for sure but works
low shard
can i run applio on termux android?
ye but it’s not much suggested
glacial pollen
Actually you can inference even on a 2019 samsung phone as tested lol
pytorch and not onnx?
Since when? last time I checked, think it was around 1 and a half year ago, nothing could be done
dang
low shard
pytorch and not onnx?
yea it runs on cpu
glacial pollen
That's interesting actually
glacial pollen
in that case, onnx mobile could make it better
Maybe I should resume thinking on the mobile port 🤔
low shard
Since when? last time I checked, think it was around 1 and a half year ago, noth...
did you have problems with installing it on termux?
with raw termux it has issues, i installed ubuntu with proot-distro to install it correctly
glacial pollen
when did you test it last time
glacial pollen
then that's an option now it seems, back then I couldn't see it happening
and was considering onnx
low shard
before it used ubuntu-in-termux which is 3rd party and doesn’t give the ubuntu 22.04
but proot-distro gives it with ubuntu-oldlts
low shard
then that's an option now it seems, back then I couldn't see it happening
If u want u can test it urself
glacial pollen
I could actually experiment once found some free time
🧐
hmmmmm actually yeah 🤔
perhaps that'd be a good usage for lightweight hiftnet vocoder or some other istft based approaches
low shard
I could actually experiment once found some free time
let me know when u do bc i’m kinda interested
glacial pollen
bet ✨
next geyser
D:\Downloads>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7897
The system cannot find the path specified.
How do I fix that?
simple ore
D:\Downloads>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7...
where did you unzip the file?
hallow thistle
Who on earth extracts files in RVC zip into "Downloads" folder? You could've put it in a path for an example like D:\RVC\your RVC files here. 
Putting RVC content in a folder that isn't named "RVC" and also containing other unrelated files can lead RVC to be unable to run. And I see that batch command looks like that batch file for the OG version of RVC GUI, which has been long outdated.
haughty pendant
Can anyone help me in coding? Because I'm trying to make communication between agents, if anyone can help please reply to me
hallow thistle
Can anyone help me in coding? Because I'm trying to make communication between a...
Uh? Is it about tweaking some RVC codes to talk with your agents? Because I've already said it at #🧬│ai-chat.
woven matrix
!help
dull ironBOT
tawdry gullBOT
!help
LunaBot 🌙 is the perfect music bot! Feature rich with high quality music! And Custom Playlist
You can start listening music by just joinning a voice channel and typing: /play [song name or link] (Remove brackets).
We support only Spotify, soundcloud, bandcamp and more!
To view more help on a specific command or category, run
/help <command> or /help <category>
Important Links:
Support
Premium
Invite
Command Categories:
🎶: Music
💰: Premium
⚙️: Utility
📕: Admin
Select A Page From Dropdown Menu Below
LunaBot
glacial pollen
D:\Downloads>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7...
You should download the precompiled zip of mainline RVC ( full package with all components needed ) or Precompiled Applio if you're into that
tho I'd personally recommend the latter
brittle wing
-guides
AI HUB Docs
golden yew
I Added A New Voice Changer From Weights.GG, And There Was An Extra File In The ...
Can Someone Help Me Please? After Puling Up The Dev Tool This Is Its Re-Direct, The Current Voice Model I Got Was Corrupted, And Because Its Corrupted, It Doesnt Load Anything, Is There A Way To Change My Model Slot At Lanch? Otherwise Its Still A White Screen
glacial pollen
Can Someone Help Me Please? After Puling Up The Dev Tool This Is Its Re-Direct, ...
what are you even using, which version, from where
which w-okada
golden yew
Thank You, Im Using MMVC- W-Okada Voice Changer
golden yew
v.1.5.3.13 Off Of Github
glacial pollen
your gpu?
@golden yew
If you don't know how to check it:
ctrl+shift+esc to open up task manager
then go to performance tab and check for your gpu
glacial pollen
heck
you should use:
https://rentry.co/ForkVoiceChangerGuide
And generally, check all the information in there
And for future, w-okada topics should be redirected to #🔍│help-w-okada
👀
golden yew
Again Thank You, Im Not That Teck Savey, Thank You For Your Time!
glacial pollen
No problem! Hopefully it goes well ✨
In case of issues, go to the channel I mentioned and ask there
Lots of helpful people in there
glacial pollen
There is a way to use uvr with amd gpu?
ye
solemn shell
https://github.com/Anjok07/ultimatevocalremovergui/releases/download/v5.6/UVR_12...
oh thanks bro
glacial pollen
tell me it ain't some rx 4/5XX
solemn shell
btw, ur vram?
rx 6600 8 gb vram
glacial pollen
a, then you mostly good I think
But ye, that is the one you wanna use. For more uvr related help I'd highly recommend visiting their discord
Audio Seperation discord
solemn shell
Ok
solemn shell
But ye, that is the one you wanna use. For more uvr related help I'd highly reco...
Do you have the discord invete of audio seperation discord?
glacial pollen
Google search
There's just one server named that way
Also, sending invites here triggers a bot so
discord . gi gi / P7FhQFH
gi -> g
@solemn shell
gluck ~
solemn shell
Ok, tks
brittle wing
Do I have to redownload the pretrain files anytime I resume?
glacial pollen
Do I have to redownload the pretrain files anytime I resume?
the G, D? Nah
they stay where you downloaded them. Those act like a base and aren't directly taken / modified in-place
brittle wing
they stay where you downloaded them. Those act like a base and aren't directly t...
Wdym
glacial pollen
?
You asked if you have to redownload them
I said nah
if you don't understand the rest, ignore that part
brittle wing
I don't have to paste the links when resuming, just inputting the G/D number as 2333333?
glacial pollen
What are you talking about?
When you do finetuning ( aka, train a model on your dataset ) the G/D files ( pretrains ) you download act like a base
azure marshBOT
What are you talking about?
When you do finetuning ( aka, train a model on your ...
glacial pollen
23333...3 ones are useless after training, as long you have your small model
brittle wing
I mean downloading a pretrain that isn't in the RVC Disconnected Colab's drop-down menu
glacial pollen
oh, then just say so you use colab
Well, if you restart the instance / colab itself, the files are gone so
you have to do so
if you didn't restart it / still have it running, there's no need
you can always confirm it by checking rvc's foler / assets / pretrained_v2 ( if .pth files are present, you don't have to re-download )
brittle wing
Well, if you restart the instance / colab itself, the files are gone so
So I have to reinput the G & D links after switching accounts?
And resuming the training
glacial pollen
ah, the resume
brittle wing
ah, the resume
Yeah cause I know where to place the links when beginning training
glacial pollen
nah, essentially you input the same config as you initially did ( first training )
requirements:
G/D with numbers files ( the 23333 ) in a corresponding model's folder + the most recent epoch / small model that was saved
brittle wing
???
glacial pollen
aka, you have to retransfer your model's folder ( from logs )
Mentioning the part you don't understand would be more helpful than writing goofy " ??? "
🙂
see, this is the " config " i talk about
You do the same stuff as you did initially
brittle wing
see, this is the " config " i talk about
...I use RVC Disconnected...
glacial pollen
- the requirement for another colab instance is to have ur model's folder in log folder
well, still aplies, whether it's cli ( command-line, or jupyter cell / colab cell for code ) or ui
model's dir / exp name, bs, and so on
that's the configuration
Why won't you just go for something simpler then? weights.gg or other stuff
brittle wing
Why won't you just go for something simpler then? weights.gg or other stuff
Me simpler?
glacial pollen
well yeah? if you can't grasp basic terminology I gave you then I suspect you're a beginner
in that case, using simpler solutions might be better for you than tinkering with resuming and synchronization and so on
brittle wing
What I'm asking is for example KLM isn't present in RVC Disconnected's drop-down menu, do I have to input the G&D links to these.What I'm asking is do I have to input them in the fields everytime I resume training or they get saved in my drive
brittle wing
in that case, using simpler solutions might be better for you than tinkering wit...
I can train models, I can resume
I'm not that much of a beginner
glacial pollen
Well then I don't get what you're talking about
You mention re-downloading pretrains but then mentione G/D233...3 files which are not pretrains
and now you're showing me custom pretrained models dropbox
brittle wing
Ignore what I wrote previously
glacial pollen
What I'm asking is for example KLM isn't present in RVC Disconnected's drop-down...
if it's not, it means it's custom
brittle wing
The G&D number is for the steps
glacial pollen
The G&D number is for the steps
Obviously my man, but that's not a pretrain, it is a checkpoint of your model's generator and discriminator
brittle wing
if it's not, it means it's custom
What is Italla pretrain?
glacial pollen
Never used it so can't give you any feedback
brittle wing
Obviously my man, but that's not a pretrain, it is a checkpoint of your model's ...
Ikr
crude flame
What is Italla pretrain?
glacial pollen
In any case, make yourself clear and use normal terminology
brittle wing
So I have to redownload the pretrain everytime I resume?
glacial pollen
if it's not available there, naturally you do
glacial pollen
but not in literal sense
G/D are enough
your model's files + most recent epoch
and call it a day
glacial pollen
If you made yourself clear in the beginning, there wouldn't be any need for such hassle really
brittle wing
https://discord.com/channels/1159260121998827560/1236418052959244341 take a look
I knew it was for Italian language
white bough
Am I understanding well? Is KLM 4.3 for speech and KLM 4.2 for singing?
brittle wing
If you made yourself clear in the beginning, there wouldn't be any need for such...
My apologies
glacial dagger
I'm trying to make a model by following all the steps from the doc but after training and trying to use my model with w-okada it produces only static. What am i doing wrong?
glacial pollen
" If I trained a model on custom pretrains then stopped ( pretrains not available for the colab I use ) and if I want to resume the training, Should I redownload the pretrains "
would have been enough really
For future for anyone seeing this. Your model's folder ( in log ) has all that's needed in terms of G/D. For the resume to work, you also have to have included most recent epoch / small model from first training, else it'll start from scratch
brittle wing
Am I understanding well? Is KLM 4.3 for speech and KLM 4.2 for singing?
Yes you are
brittle wing
" If I trained a model on custom pretrains then stopped ( pretrains not availabl...
I could've worded it better
glacial dagger
I'm trying to make a model by following all the steps from the doc but after tra...
when trying to upload my model in applio it errors. list index out of range
it errors upon uploading the index
brittle wing
-guides
dense marsh
unique rock
What effects should I put on the voice to make it sound more real? So far I know that I should put a reverb, a Graphic EQ and an De Esser.
flint solar
What effects should I put on the voice to make it sound more real? So far I know...
this is a channel for rvc help
not for vocal mixing
latent kettle
What effects should I put on the voice to make it sound more real? So far I know...
An EQ and tone shape your vocals (very important), a compression, a reverb (if needed) saturation (if needed)
unique rock
An EQ and tone shape your vocals (very important), a compression, a reverb (if n...
thank you
silk helm
im trying to find the latest voice changer client but i cant seem to find it do u guys know where i can find it because mine is really old?
hallow thistle
im trying to find the latest voice changer client but i cant seem to find it do ...
W-Okada?
-realtime
hallow thistle
The first link for the forked version of W-Okada runs best.
silk helm
oh!
silk helm
The first link for the forked version of W-Okada runs best.
which one do i dowload cause i have winows 11
low shard
im trying to find the latest voice changer client but i cant seem to find it do ...
what’s ur pc gpu btw
low shard
which one do i dowload cause i have winows 11
that depends on ur gpu
silk helm
im geeting a new pc coming in today acually
low shard
im geeting a new pc coming in today acually
whats its gpu?
silk helm
my gpu is going to be a 4080 super
low shard
my gpu is going to be a 4080 super
then Nvidia yea
silk helm
yee
hallow thistle
Damn. That's a pretty decent GPU you'll get. 
silk helm
yea i upgraded everything to the latest for my new pc can i show specs here or?
low shard
yea i upgraded everything to the latest for my new pc can i show specs here or?
i mean sure u can if u want
anyways it’s good enough to run it ofc
silk helm
thats god
lavish lintelBOT
gentle pivot
-realtime
hallow thistle
Oh wait, this channel is #✨│ai-help not #🔍│help-w-okada. 
silk helm
should i mover down then
silk helm
need assistance on okada i have the okada voice changer open and running now im trying to figure out how to input vac lite or how to use it
knotty moth
need assistance on okada i have the okada voice changer open and running now im ...
Line 1 (VAC) as voice changer output and discord/game's input
silk helm
thats the thing i cant find line one
brittle wing
which one do i download guys
i have a rtx 3050 laptop gpu
onnx directml cuda
or onnx gpu cuda
silk helm
which one do i download guys
vac 4.70 lite download
hallow thistle
which one do i download guys
Please download the normal CUDA one for your NVIDIA GPU. Unless you have AMD/Intel dedicated GPU, you can download that DirectML one.
brittle wing
what?
brittle wing
Please download the normal CUDA one for your NVIDIA GPU. Unless you have AMD/Int...
wdym normal cuda
oh
ok
directml
got it
low shard
rtx 3050 ti laptop gpu
that's good
It's named Wokada
There's also a fork (modified version), the deiteris fork, which gives better performance
-rt
low shard
1st link is the wokada fork
2nd link is original wokada
I'd suggest the fork
Ofc u gotta download the Nvidia version said in the written guide
hallow thistle
wdym normal cuda
What do I mean? Yeah, ok. The regular CUDA version is for NVIDIA GPU, while DirectML CUDA is for AMD/Intel GPU.
low shard
And don't follow yt tuts, there's no updated one
hallow thistle
vac 4.70 lite download
That's not what he asked for. 
silk helm
sorry replyted wring person
viscid moss
vac 4.70 lite download
restart ur PC and W Okada after Install the VAC
hallow thistle
low shard
also for wokada help it's:
#🔍│help-w-okada
silk helm
restart ur PC and W Okada after Install the VAC
in the vac file do i need to run anything?
viscid moss
in the vac file do i need to run anything?
after install no
silk helm
after install no
im in th vc if u dont mind
hallow thistle
also for wokada help it's:
# <#1159290161683767298>
I've already said it so here, but still.
low shard
I've already said it so here, but still.
smh this channel is always misused
low shard
in the vac file do i need to run anything?
btw next time ask in #🔍│help-w-okada not here
silk helm
after install no
ive restarted my pc i still dont see it
low shard
This channel is for RVC, which is not the same as wokada
viscid moss
oh wrong channel mb too
hallow thistle
Unless you wanna run RVC the audio converter on your PC, you can go here for it. 
Just letting you know.
low shard
Unless you wanna run RVC the audio converter on your PC, you can go here for it....
ig some think that RVC is Wokada 😭
I seen some calling wokada as:
- voice changer demo
- voice changer client
- RVC
- MMVCsio
low shard
Wokada is woke!
Wokada crypto coin
simple ore
low shard
ran by e girls
hallow thistle
W-Okada uses RVC voice model. I always assume people here don't read more than one line of a guide, so they'd understand as that.
low shard
W-Okada uses RVC voice model. I always assume people here don't read more than o...
people needa read
it's the way to go
analog obsidian
W-Okada uses RVC voice model. I always assume people here don't read more than o...
in older versions (youtube tutorials) there's some big letters in the gui naming wokada "realtime voice changer"
pic from google pls don't use any of the settings here, they're horrible
hence why the confusion
so people assume rvc > realtime voice changer
this is wok's fault
low shard
so people assume rvc > realtime voice changer
absolutely not
hallow thistle
Here goes the outdated YouTube tutorial videos. 
low shard
u gotta explain people that RVC = Retrieval-based-Voice-Conversion
low shard
Here goes the outdated YouTube tutorial videos. <:trol:1159397152183824405>
smh
People gotta memorize the acronym
winter garden
Can someone help me find a colab for doing rvc? I just want to upload a zip of a model and upload an audio file and get one back. Every link I’ve tried to find online or in this discord has been exactly not what I need
azure marshBOT
Can someone help me find a colab for doing rvc? I just want to upload a zip of a...
hallow thistle
Can someone help me find a colab for doing rvc? I just want to upload a zip of a...
-colab
winter garden
Which one
low shard
Can someone help me find a colab for doing rvc? I just want to upload a zip of a...
for doing rvc
I'm guessing you're talking about RVC inference, so using the models
low shard
Which one
what's your pc gpu btw? maybe u could run it locally than cloud
It's better to always know that first
winter garden
1050 ti
hallow thistle
I've never tried one of these, but first three ones are RVC.
low shard
1050 ti
Your GPU is good enough to do inference (use models) locally (on ur pc), you won't be able to train (make models) but use them
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- Cloud (remote good pc, easier and faster than ur PC but it's limited):
- Ilaria RVC Zero: fastest and simplest that you can get for free
- Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio Colab: max 4 hours, not granted, of GPU
winter garden
I tried applio on the cloud but the download kept getting stuck
hallow thistle
1050 ti
This is the minimum NVIDIA GPU series for doing AI. I mean, you can inference AI like fine, but it won't be so fast.
winter garden
I tried applio on the cloud but the download kept getting stuck
but it’s gonna do what I need?
low shard
I tried applio on the cloud but the download kept getting stuck
What download link model are u using?
winter garden
no I meant the applio colab, on the install applio snippet it kept getting stuck on downloading or something
knotty moth
iirc there were some ppl using the first wokada version coming from some yt tutorials (where it still supports svc and something)
low shard
but it’s gonna do what I need?
Applio = RVC (Retrieval-based-Voice-Conversion, Speech to Speech) Fork (modified version), it's a bit easier and faster
winter garden
I don’t remember exactly
low shard
no I meant the applio colab, on the install applio snippet it kept getting stuck...
and what model are you downloading
you need to be sure to not leave the colab page
you always need to keep open the colab page and the applio ui page
and to be connected to a runtime
low shard
Your GPU is good enough to do inference (use models) locally (on ur pc), you won...
here I gave a Tutorial about it, and also other alternatives
low shard
I’ll try it locally then
Your GPU isn't the best, will be slower than colab, but should work
-# even if I'd suggest upgrading
low shard
I didn’t expect it to be good enough to even do rvc
What's your ram
winter garden
16
low shard
Yea then it's fine
low shard
I didn’t expect it to be good enough to even do rvc
U got the minimum requirements lol
low shard
i need it for work so I can’t have it be unreliable
Yea Locally is better to have it whenever you want
Cloud is limited by time
quartz turtle
How can I solve this installation problem?
Installing dependencies...
Failed to create temp directory "C:\Users\Homura Akemi\AppData\Local\Temp\conda-<RANDOM>"
An error occurred during installation. Please check the output above for details.
hallow thistle
Did you try to install RVC to your Conda environment?
quartz turtle
Did you try to install RVC to your Conda environment?
I was just trying to install Applio on my computer, do I need to install anything specific first?
knotty moth
Installing dependencies...
Failed to create temp directory "C:\Users\Homura Akem...
delete that damn space in the path name or move to somewhere else like D:\
hallow thistle
Try install RVC in a directory like D:\RVC\.
quartz turtle
delete that damn space in the path name or move to somewhere else like D:\
What space? The username space?
quartz turtle
Try install RVC in a directory like `D:\RVC\`.
Ok
hallow thistle
simple ore
I was just trying to install Applio on my computer, do I need to install anythin...
download the compiled .zip
and just unzip, ffs
no need to "install"
quartz turtle
download the compiled .zip
Which zip file? Sorry for my small capacity 😦
quartz turtle
Try install RVC in a directory like `D:\RVC\`.
I keep getting the same problem
quartz turtle
unzip to C:\Applio
yes, and then? what is FFS?
simple ore
and then run-applio.bat
hallow thistle
-rvc
quartz turtle
and then run-applio.bat
Traceback (most recent call last):
File "D:\RVC\Applio-3.2.8-bugfix\app.py", line 1, in <module>
import gradio as gr
ModuleNotFoundError: No module named 'gradio'
simple ore
wait until it finishes unzipping
hallow thistle
Please never download the uncomplied source code one, unless you wanna try to make them your own fork program. This image is the compiled one.
hallow thistle
There's a dirty way. Try put both .pth and .index files together into logs folder, and refresh your voice model list on browser.
winter garden
have to be same name?
hallow thistle
Every voice model should have the same name, otherwise you'd get confused which .index should belong to which .pth file.
winter garden
nah this is normal tbh
simple ore
unless you're working with colab UI version then UI is the only way to do it easy way
hallow thistle
Huh? Do you have any problem trying to let it to export audio to a specified directory?
knotty moth
have to be same name?
no need to rename the index and pth file if you have trained it
hallow thistle
Look at \Applio\assets\audios on your Explorer, and see if there's anything other than some original audio files.
winter garden
nvm the format what the issue
m4a didnt work
mp3 works
idk why because it has ffmpeg
shouldnt be an issue
winter garden
i just went to the voice recorder app on windows to get a sample of my voice
just to see if it works
flint solar
rvc will automatically convert it to wav for inference
hallow thistle
No way. A .wav file is the only file it can work on. This Applio can be tricky sometimes.
winter garden
alright ill keep that in mind
any settings you think i should tinker with for better results?
simple ore
m4a is not a supported format
simple ore
No way. A `.wav` file is the only file it can work on. This Applio can be tricky...
wav, mp3, flac... whatever librosa supports by default
grand pecan
Hi, is there a guide to train a model and make it able to laugh and scream? I have a 4070 ti super btw, thanks!
flint solar
m4a is not a supported format
but it still loads then gets converted to wav
north willow
ok i am in the right channel now, what do i install to begin, have rtx 3070 ti laptop
like the first step
simple ore
but it still loads then gets converted to wav
it loads it as a data array 🙂
north willow
come on guys someone help me
azure marshBOT
come on guys someone help me
north willow
!howtoask
north willow
I would greatly appreciate your help in this enticing task super. I have an RTX 3070 ti graphics card and would want know what is the first step to take to get access to an sts ai voice changer. Now i ping a helper @pastel oak
odd shale
https://rentry.co/Forkvoicechangerguide
🐢
north willow
i have done two steps that are presented in the picture however i do not know what to open now
if i need to open the zip file i dont know what to open it with
so now i need help
pastel oak
its as simple as extracting, then opening the folder and running the file
its nothing complicated
like with any other program youd do
this is already the file you extracted
north willow
How exactly do I upload models I know there is a section in the guide but I struggle to understand it
low shard
@pastel oak @north willow please use #🔍│help-w-okada
north willow
Ofc
charred quail
I have a hoarse voice when choosing a voice model, all the settings are by default, can anyone help me?
azure marshBOT
I have a hoarse voice when choosing a voice model, all the settings are by defau...
low shard
I have a hoarse voice when choosing a voice model, all the settings are by defau...
use #🔍│help-w-okada
RVC is different
charred quail
👍
unborn sleet
can someone help me with finding an rvc thats beginner friendly or similar to the vids in guides im finding different ones
azure marshBOT
can someone help me with finding an rvc thats beginner friendly or similar to th...
low shard
can someone help me with finding an rvc thats beginner friendly or similar to th...
All yt tuts are old
there is no single RVC Video tutorial updated
What's your pc gpu?
unborn sleet
its a laptop but its got a 3050
low shard
its a laptop but its got a 3050
What's the VRAM?
unborn sleet
What's the VRAM?
12gb
low shard
12gb
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- Cloud (remote good pc, easier and faster than ur PC but it's limited):
- Ilaria RVC Zero: fastest and simplest that you can get for free
- Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio Colab: max 4 hours, not granted, of GPU
Easiest Local: Applio
Easiest Cloud: Weights.gg
There's also Replay which is an RVC local fork made by Weights.gg but idk much about it
unborn sleet
You can:
- Locally (runs on your pc so the speed depends on that, you will have ...
alright thanks for the help
low shard
yw
blazing solar
-colAB
strong shadow
hey can i get some help?
I want to turn this text so the signature looks like diamonds
does anyone know how?
low shard
I want to turn this text so the signature looks like diamonds
?
This doesn't seem related to RVC nor AI
strong shadow
?
This doesn't seem related to RVC nor AI
isnt this channel supposed to be intelligent people? how I do it x please
low shard
isnt this channel supposed to be intelligent people? how I do it x please
isnt this channel supposed to be intelligent people
What? This server is for AI, and this channel specifically is for RVC
AI is completely different than Graphic Designing
Maybe u could try some free image editors but i don't really know, I'm not a graphic designer
-# Or google it
Tech has many and many sub categories 
storm holly
on the applio colab
magic_number = pickle_module.load(f, **pickle_load_args)
EOFError: Ran out of input
hallow thistle
isnt this channel supposed to be intelligent people? how I do it x please
Intelligent people of what?
tame mica
I want to turn this text so the signature looks like diamonds
Illusion Diffusion is probabky the thing you're looking for :p
knotty moth
isnt this channel supposed to be intelligent people? how I do it x please
RVC is an AI architecture for voice conversion (STS), which is not related to your topic at all. you could consult here #🔍│help-ai-art
hallow thistle
brittle wing
I hope everyone is doing well. How do I add custom pretrainers like KLM into Applio, I keep getting errors maybe I'm not placing it in the right spot?
knotty moth
I hope everyone is doing well. How do I add custom pretrainers like KLM into App...
Applio\rvc\models\pretraineds\pretrained_custom
then click refresh custom pretrains
frail vortex
what's the newest rvc rn
brittle wing
Applio\rvc\models\pretraineds\pretrained_custom
then click refresh custom pretra...
Thank you, you're the best!
brittle wing
what's the newest rvc rn
Applio but they all sound the same.
frail vortex
ty, got the new link
low shard
what's the newest rvc rn
What's ur PC GPU
flint solar
was abt to ask lol
low shard
was abt to ask lol
🔥
U understood how it works 🔥
Ai hub 1st rule: we ask your PC GPU first
Ai hub 2nd rule: we do not talk about furry rp
knotty moth
<@681186927151546397>
resample to (go find out some best resampler with minimal aliasing) and set target to either 40k or 32k sample rate, and again it must match the pretrain's sample rate
brittle wing
resample to (go find out some best resampler with minimal aliasing) and set targ...
Okay, I resampled to 32k. How's it looking?
knotty moth
Okay, I resampled to 32k. How's it looking?
it is called brickwall filter, not resampling smh
and not a right way to do
brittle wing
it is called brickwall filter, not resampling smh
and not a right way to do
Oh my bad 😥
What programs would you suggest a better way?
blazing solar
-colab
safe sparrow
i want to run AICoverGen locally, but when I try to install the requirements.txt, it gives me errors
grizzled heart
hello, i downloaded this... what the hell is this? how can i use this to generate ai voice
knotty moth
i want to run AICoverGen locally, but when I try to install the requirements.txt...
you should follow the instructions carefully, usually in the repo's readme
probably like downgrading pip to 23.1 first
brittle wing
guys is the okada voice changer safe or not
reef tusk
hello guys, how can I smooth out the vocals so they don't sound so robotic?
ripe granite
it feels like I have an old version of tensorboard installed... how do I update it or something? on the doc page there's nothing written about it... pls help
safe sparrow
you should follow the instructions carefully, usually in the repo's readme
proba...
The all working local rvcs in guides here has one called AI-Cover-Maker-UI, by Shirou. How do I install it?
Wait, maybe i didnt let the run.bat run for long enough cuz it wasnt updating and i thought it finished whatever it was doing
knotty moth
it feels like I have an old version of tensorboard installed... how do I update ...
wdym old? I see no difference at all
ripe granite
well, i can't follow these instructions
cuz i just don't have these options
maybe im dumb and I don't understand something
knotty moth
you mean you haven't started training and made into 1 epoch?
ripe granite
you mean you haven't started training and made into 1 epoch?
oh, so I need start the training first?
hallow thistle
brittle wing
guys is this safe pls help
low shard
guys is this safe pls help
We don't give Wokada support into this channel
use #🔍│help-w-okada
modern surge
-colab
lavish lintelBOT
hard crystal
i know this is the wrong channel but does anyone know whats good software to make your own tts voice thats decent, because i tried to setup tortoise however theres nothing for web inference nor training
silent stratus
i know this is the wrong channel but does anyone know whats good software to mak...
fish.audio
bitter tiger
hey i need rvc for pc
silent stratus
hey i need rvc for pc
applio
odd valve
https://huggingface.co/IAHispano/Applio/tree/main/Compiled/Windows
thanks a lot
odd valve
@simple ore I was thinking this would be like the rvc v2 where it could do it in real time and direct it straight through your microphone
is that somewhere else now?
simple ore
https://rentry.co/forkvoicechangerguide#download-nvidia for nvidia, or scroll down for amd/intel/cpu
and one channel below this for help with that
limber canopy
help help
help
rvcs sound ass
and they dont sound like the sample
how do i fix this
fallow hearth
why is applio not letting me look at their vc changers
fallow hearth
its the voice changer
fallow hearth
yeah sometimes the voice changers ruin the pitch for the rest of them
just reset the app
limber canopy
this is the sample b
limber canopy
do you know why i cant find the catalog of voice changers on applio
idk
fallow hearth
yeah the same stuff has happened to me it changes the pitch when i switch between certain voice changers cause the apps perseption of the pitch and what it is meant to be just goes awol
and it doesnt matter what pitch i set it to it sounds off and strange
until i reset the app
fallow hearth
idk if you already know this but voice changers with higher epochs sound more realistic and sound closer to the person
limber canopy
epochs
fallow hearth
i restarted it
if the problem persists get another version of the app cause sometimes its just a bad version it pmo on how some are just low quality
fallow hearth
if the problem persists get another version of the app cause sometimes its just ...
but thats the last step
and usually its buggy
fallow hearth
what number is it
limber canopy
limber canopy
that explains it
fallow hearth
theres always better voice changers of the same person
fallow hearth
id say 300 and over but for me id say 600 epochs
limber canopy
id say 300 and over but for me id say 600 epochs
k k
fallow hearth
300 is probably a good epoch amount and as the epochs get higher they get better
limber canopy
okay
limber canopy
what kind of gpu do you have
4060 ti
fallow hearth
oh thats the same one i have
limber canopy
huh
fallow hearth
change the chunk to 384
fallow hearth
the chunks are how long the words your saying are processed and the quality, the larger the chunk = more delay but better quality so you just want to find the sweet spot
fallow hearth
i wouldnt go over 1.5 seconds if im talking realtime tho
fallow hearth
it lowers the chunk which means its lower quality
but its delay is faster
if you know what im saying
limber canopy
oooh
its like
more lower resolution you get more refresh you get
in overclocking terms
fallow hearth
fallow hearth
more lower resolution you get more refresh you get
yess
limber canopy
ima copy
fallow hearth
put the RVC quality to high
limber canopy
how do i get that
limber canopy
oh ok
knotty moth
and the transcate to 300
you two pls discuss your topic in #🔍│help-w-okada
limber canopy
why is your thing set to serevr
limber canopy
you two pls discuss your topic in <#1159290161683767298>
ok
glacial pollen
whats the best settings for rtx 3080 i7-14700k
Setting for wut
batch_size ?
cause if you refer to w-okada, then as it was pointed out:
Any talk about voice changers should be redirected to w-okada's help channel
rvc help channel is for rvc and applio
torpid loom
@glacial pollen does applio support m4a or flac for training?
glacial pollen
<@1239634084133601423> does applio support m4a or flac for training?
as for m4a, I am not 100% sure but it should support flac
as far as I know
but this doesn't matter much in fact because it all ends up as wave anyways
the preprocessing step which takes your dataset and does on it what it needs to, ultimately converts it to waves; 1 set for target sr and one for 16khz
So in a short, it supports any codec that the preprocessing will take
torpid loom
@glacial pollen im ripping files off tidal
so after i isolate the vocals
i can export them as wav and all good right?
i just put that into the dataset and alls taken care of
glacial pollen
<@1239634084133601423> im ripping files off tidal
Do you know what kind of codec tidal operates on?
is it aac or anything of that sort?
in any case, if I was you, I'd rip it as it is, keeping the original format
then analyze what's the frequency spectrum's response ( spek, izotope, a audition, anything with spectrogram preview will do )
for instance, if the range resembles 44.1 or 48
I'd just then export the obtained stuff to a target sr wave ( and 16 bit )
This way you avoid wrong / redundant resampling
torpid loom
im pretty sure its 44.1k flac
glacial pollen
if you're absolutely sure of it, then I suppose you can go for 44.1 32 float
if you can afford memory / space for whatever you work on, that is.
In other case, go for 24 bit
Flacs are typically 24 bit yet you can go for 32 float as that's how the set ends up anyways. You'd avoid redundant conversions
torpid loom
i have it as 16 bit though
glacial pollen
welp, rip in that case
shouldn't be an issue unless you're quality-perfectionist I suppose ( or a matter of how tidal handles it. whatev the case is, you're good )
Go for 44.1 16 bit wave and call it a day
glacial pollen
Can you elaborate?
torpid loom
i mean assuming i have pretty high quality vocals
glacial pollen
You're gonna use stock pretrains?
glacial pollen
Titan? or something?
glacial pollen
A
torpid loom
should i use those ones? or stock applio instead
glacial pollen
Well, I can't really give you any guidance for customs so it'd be a blind-road
but I can def guide you on the stock ones ( Applio, rvc, both use the same )
Another thing would be, how diverse would be your set?
is it truly something you'd call diverse or are some patterns repetitive?
torpid loom
uhh
glacial pollen
and foremostly, what's your current total length of the set? can be approx
torpid loom
i guess you could say somewhat repetitive
i have a lot of audio to choose from
but i havent downloaded pretty much anything yet
torpid loom
what do u mean by set length
glacial pollen
total length of all samples
we call it " total dataset's length "
sort of
just go ctrl+a on all files you have, properties and then check the total length
glacial pollen
no I meant, how much files do you already have ( and their total length )
glacial pollen
ahhh, then that changes the thing
glacial pollen
Well, if you say it's rather repetitive
perhaps go for 25 to 30 minutes
and for batch_size, you'd try 8 ( and if results are meh + you made 100% sure it's not an user / dataset error, you can try 4 or 12
torpid loom
im a lil rusty on this
now i know batch size affects how long it takes to train
but how does quality play into it
glacial pollen
Not quite, it's a lil more complex than just that
torpid loom
im all ears
glacial pollen
In a short, higher batch_size means more data at once per step ( within an epoch ) is used to contribute in internal updates # historical code: 836
aka, everything's smoothed out / there's less " noise " or you can call it variance
It's the opposite with small batch, you get more noise / variance
Now, in situations where you're short on data ( or it's not too diverse), you want to " not oversmooth " the learning because then the model might overfit too fast to the data and not find the rather optimal solution
tl;dr:
If you're short on data or it's too repetitive ( and I mean it by repetitive ), go for smaller batches: 4, 6 sometimes 8 if you'e good on data and it's quite diverse where each contributes in a way, you can afford higher batch size values such as: 8, 12, 16
glacial pollen
you see, when you go for somewhat noisy but not too noisy batches
you potentially increase the chance of better generalization
as the path's not " so easy " and encourages to seek more " paths " towards the " goal "
lol
glacial pollen
and up to how many epochs would u say is enough
And that should be based on the tendencies / status you observe in tensorboard
torpid loom
oh yeah
glacial pollen
Recommend for you to get familiar with these terms
torpid loom
once it plateaus
glacial pollen
kinda but not quite
that's just one of the signs
Most observed scenarios are included in the pic
and as always, it's a good practise to pause the training once unsure
test the model, if it's underperforming, continue the training
( and with that, keep saving frequency as 1; the epoch saving frequency ofc )
torpid loom
if i run applio on my browser (locally) does it still have a tensorboard?
does it show up when i begin training
glacial pollen
uhhh, not sure how it was before
but you can easily open it up on your own
In your model's folder ( in log )
once you start the training and you see tfevents file appeared, you can open up the batch and input the path as shown on the ss
then spam the enter key few times and should open up in the browser
knotty moth
i have it as 16 bit though
most rips are in that pcm format, but for audio editing, mastering, etc. convert it first to 32 float and inspect the spectrogram if it has clipped samples
torpid loom
most rips are in that pcm format, but for audio editing, mastering, etc. convert...
ty ill take it into account
spiral notch
Disclaimer, I'm an absolute mega noob here.
How do I prevent grit, leftover noise, and ugly screams in my covers? Is it a problem with the voice models I'm using, or is there something I can fix? Cover with the problem I'm dealing with incase hearing said grit and screams helps identify what I'm talking about.
There's no way my voice could stand up to Bruno Mars' so re-covering the song with my own voice isn't in the question, sadly.
Tryna make one of those dumb memes with this song, lol
One of the worst screams is at 0:39 + the Nasally-ness of the voice is a problem, too.
tiny drum
im using MMVCServerSIO, and the voicechanger only applies to the output (headphoens) and not the microphone
i can hear the changed voice on my headphones but it doesnt actually output it on the microphone
glacial pollen
most rips are in that pcm format, but for audio editing, mastering, etc. convert...
pcm is actually an audio encoding scheme / method than a format on it's own
but good point with the clipping
It's always advisable to convert 16 bit depth to 32 float for editing or tinkering with files, just in case one would screw the dynamics or volume levels as 32 float is practically impossible to clip unless you tryhard
glacial pollen
ye, 's why it's practically impossible unless you try really hard
cause who sane ( provided they have a brain ) uses that kind of volume
ionic bison
how do i do it so i cant hear myself when using it?
glacial pollen
how do i do it so i cant hear myself when using it?
#🔍│help-w-okada is the channel you want
This one is for RVC / Applio
ionic bison
oh sorry
knotty moth
cause who sane ( provided they have a brain ) uses that kind of volume
not even our universe could reach that range
glacial pollen
trust me, some goofy ah trolls are capable of everything.. ( virtually, that is )
lmao, imagine the size of a blackhole it'd create
~1100db in a small tight space and you get a black hole lol, awesome
sharp surge
do you guys still use google colab'
low shard
do you guys still use google colab'
it's a matter of how good is your pc
what's your pc gpu and what are you looking for?
google colab is a cloud computing service for people with bad pc
sharp surge
I think I have that😹
low shard
I think I have that😹
could you tell me what's your pc gpu and what you want to do?
You can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab -> GPU
low shard
do you guys still use google colab'
#✦│chat message alr nvm i see ur tryna make ai covers
If you got a good pc you can do it locally, but you need to tell me your pc gpu
else i can give you all options to make ai covers on cloud, google colab isn't the only one
sharp surge
Okay can you??
low shard
Okay can you??
Can you tell me what's your pc gpu?
So I can help you
It's better to always check it as local is better than cloud, as cloud got time limits
@sharp surge do u wanna check or directly just want all cloud options? I wanted to check it because if it's good enough you can do it locally without time limits
low shard
I’m using a gaming labtop
and did you check what's its gpu ?
low shard
You can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab ...
I explained how in this message
hallow thistle
The one that says GPU? What is the name of it?
hallow thistle
Now what is the name of it?
hallow thistle
That's VRAM, not what the name of your GPU is.
knotty moth
I have one that’s says GPU 0 And GPU 1
maybe GPU 1, is it really 3,9 GB?
sharp surge
Yess
hallow thistle
The GPU 0 is an integrated one, while GPU 1 is the dedicated one.
sharp surge
I am on GPU 1 now
hallow thistle
You still haven't tell any name of these GPUs, so I couldn't identify which RTX your laptop has.
knotty moth
You still haven't tell any name of these GPUs, so I couldn't identify which RTX ...
probably corporate secret
hallow thistle
Please tell the number of this GTX.
knotty moth
err.. I mean GTX 1650?
sharp surge
1050
hallow thistle
This number for GTX GPU series is the minimum to run any AI related program.
knotty moth
1050
that's the weakest minimum, but you can still run RVC inference
for realtime, you should consult in #🔍│help-w-okada but I dont think you could expect good enough performance
sharp surge
It’s ok
hallow thistle
I'm not sure why you couldn't tell what GPU your laptop has when there's the name of it on Task Manager. I don't have a GPU in my laptop, so next time this is where the GPU name is.
hallow thistle
Tutorial videos on YouTube and TikTok are outdated.
sharp surge
okay
thorn gale
how to use in amd?
hallow thistle
is thereb no vide4o¨'
I've already told you about that.
sharp surge
omg yes ik
hallow thistle
how to use in amd?
Please be more specific on which AI program you're using.
I see you asked the same question in #🔍│help-w-okada, so I'm not really sure which one you wanted to use with your AMD CPU or GPU. RVC is an audio conversion, while W-Okada is the realtime voice conversion that uses RVC voice model.
lucid creek
-colab
low shard
1050
Your GPU is good enough to do inference (use models) locally (on ur pc), you won't be able to train (make models) but use them
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- Cloud (remote good pc, easier and faster than ur PC but it's limited):
- Ilaria RVC Zero: fastest and simplest that you can get for free
- Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio Colab: max 4 hours, not granted, of GPU
yes there is no video
u gotta read
wintry torrent
is there something like rvc that has a built in Vocal extractor?
flint solar
is there something like rvc that has a built in Vocal extractor?
u can use aicovergen colab
it separates vox for u
wintry torrent
u can use aicovergen colab
Thanks
knotty moth
alright, 40k it is then
ah nope, pls change the freq scale to linear
reef tusk
how do I do that
knotty moth
and it seems 16khz (32k)
knotty moth
how do I do that
right click the scale at left
knotty moth
it seems to already have gone through UVR denoise, so you should set to 32k anyway
glacial pollen
more like sub ( 36-38~khz
reef tusk
hmm, I see
glacial pollen
hmm, I see
you still have the source audio?
analog obsidian
reef tusk
yes
reef tusk
40k
reef tusk
the other model that I tried was also 40k and it sounded alright
glacial pollen
always depends on set really so
if sources and sample rate were similar then you get the point
reef tusk
yeah, i get it
also, what should the "truncate silence" thing do, it doesn't appear to apply any effect on my audio file
analog obsidian
also, what should the "truncate silence" thing do, it doesn't appear to apply an...
removes silences from the dataset
keep in mind if u do that u have to remove the mute file from your filelist.txt
reef tusk
filelist.txt?
analog obsidian
filelist.txt?
rvc injects a mute file for a reason i cant remember now lol if u truncate silence rvc is going to train the full mute file and the model will collapse
its a bit advanced
reef tusk
would "truncate silence" remove the gaps between the audio ?
analog obsidian
would "truncate silence" remove the gaps between the audio ?
yes
ideally you want that
but sometimes if you do that rvc collapses and you'll get values like this in g/total, g/mel
because its trying to learn this
which is nothing
that is the mute file
flint solar
but sometimes if you do that rvc collapses and you'll get values like this in g/...
bro started training wit a collapse 😭
analog obsidian
because its trying to learn this
so if u got a collapse due to this mute file u can remove the mute file from the filelist.txt in your model log folder
u just load the .txt file into notepad++/notepad and search for every line that contains "mute" and remove these, there are 2 mute file lines in filelist.txt IIRC
(only do this if you notice your values went from like 30 to 5 in the graphs, don't remove them from the list if your model didn't collapse)
to confirm the mute file is the culprit of the mode collapse, go to the images tab in tensorboard and check the step number where the collapse occurred, if it looks like my pic, its the mute file and you have to remove it
never delete the "mute" folder from the logs folder, only remove the lines that mention it from the filelist.txt present in your model log folder
analog obsidian
bro started training wit a collapse 😭
wintry torrent
this happened to me last year when i tried to install aicovergen and its happening again
simple ore
fairseq says fuck you
you not gonna win this fight
install it separately ignoring requirements
insteall hyndra-core separately ignoring requirements
but do downgrade omegaconf to required version first
wintry torrent
insteall hyndra-core separately ignoring requirements
so i should downgrade omegaconf to 2.0.5 then pip install fairseq and hydra-code without versions
simple ore
you can enforce omegaconf to a specific version + fairseq + hydra-core without
hold on
wintry torrent
Can you tell me how id do that because i have barely any experience with pip and git
simple ore
generally it is just pip install omegaconf==2.0.5 fairseq hydra-core
it will try to accommodate all 3 compatibility-wise
but lemme see first
wintry torrent
generally it is just ``pip install omegaconf==2.0.5 fairseq hydra-core``
WARNING: Ignoring version 2.0.5 of omegaconf since it has invalid metadata:
i tried 2.0.4 and it gave the same output
simple ore
just a sec
wintry torrent
👍
simple ore
use this
simple ore
instead of old one, yes
wintry torrent
Okay
simple ore
wintry torrent
Ill lyk if it works thanks
simple ore
i'm not actually sure why
i mean it did install the same fairseq
I guess make sure you're using a virtual environment
wintry torrent
A virtual enviroment?
wintry torrent
A virtual enviroment?
Like a virtual machine?
knotty moth
this happened to me last year when i tried to install aicovergen and its happeni...
is the pip version 24.x instead of 23?
(I remember that issue in the colab)
wintry torrent
is the pip version 24.x instead of 23?
its the default one that comes with python 3.9
simple ore
Like a virtual machine?
virtual environment is a separate folder for libraries that are only used for a specific project
you need to activate it before you run the application
wintry torrent
is the pip version 24.x instead of 23?
oh yeah its 24.3.1
I mightve upgraded it on accident
knotty moth
@wintry torrent try downgrade to 23 like in that screenshot
wintry torrent
<@730491380006387782> try downgrade to 23 like in that screenshot
Is it the same command as upgrading but with downgrade?
Figured it out
knotty moth
Is it the same command as upgrading but with downgrade?
pip install -U pip==23.0.1
knotty moth
yep
wintry torrent
I guess make sure you're using a virtual environment
How would i make one
knotty moth
ill try that if changing the pip version didnt work out
now do you also have some errors related to numpy and librosa?
knotty moth
that's good
simple ore
the requirement are frozen in a terrible old setup
but they are frozen
so likely would install correctly
that error was because of pip 24.x.x
quite interesting because installing fairseq with anything else is an exercise in futility
wintry torrent
havent gotten this far before
knotty moth
So far, no
from my experience btw, numpy 2.x mostly causes issues on my cases (so need to downgrade to 1.x), like the rvc colab, librosa itself, and also comfyui while trying to get WAS node suite working
wintry torrent
it installed 1.23.5
knotty moth
yeah, not like the recent version of comfyui on mine being using numpy 2.x
snow meadow
help
glacial pollen
Including a description of your issue greatly increases the lihelihood of someone actually responding, btw
low shard
Including a description of your issue greatly increases the lihelihood of someon...
-# never replies again
/j
charred jewel
How can you tell if a model is overtrained, or undertrained?
glacial pollen
How can you tell if a model is overtrained, or undertrained?
And read on from that point ( you can check a lil above it for more context )
Other than that uhh.. you can just hear it. ( PROVIDED IT IS NOT: user error, dataset's being small / not too diverse, dirty audio input ) The model's sounding glitchy and underperforming
glacial pollen
Erhm, reading doesn't hurt I believe
charred jewel
The graphs are kinda confusing to me
glacial pollen
I mean yea, there's no easy way around it man
charred jewel
Oh, I just didn't scroll up far enough
glacial pollen
Training AI was never meant to be easy, you gotta read for a lil bit to understand how it works pretty much
But I promise it's worth it
✨
Take your time on it, there's no rush whatsoever
glacial pollen
Here's all loss functions ( including D and G )
(( I've made it quite a while ago but should still be more or less up to date / acceptable ))
charred jewel
I mean, like what do they stand for? Bc I don't know all of the abreviations.
glacial pollen
man man man..
charred jewel
nvm
charred jewel
Yeah, I'm kinda blind lol
glacial pollen
Ah, nevermind then. In that case
charred jewel
You gotta add some subway surfers gameplay underneath
glacial pollen
G loss: Loss for Generator
D loss: Loss for Discriminator
FM loss: most important in terms of model's matching the dataset's features ( timbre, characteristics and so on )
Mel loss: Clarity / fidelity in a short.
KL loss: A more advanced one, it's distribution divergence ish loss ( this one you can ignore really )
is that any easier for you 🙂 ?
charred jewel
Ok, so the one's I need to focus most on are the KM and Mel loss?
simple ore
fm and mel loss
glacial pollen
Mel, fm and total G
If you need any other format of a guide, I suppose..
https://docs.applio.org/applio/getting-started/tensorboard
But I wasn't making it so can't verify it's accuracy or correctness
charred jewel
Ok, tysm.
glacial pollen
Best of luck with the training
simple ore
fm loss has only one meaning - whether generator does generate the spots the discriminator checks correctly
timbre/spatial characteristics and other things do not really matter here
glacial pollen
it certainly dose, if one wants to know the meaning
too much abstraction isn't always the best
That's why we have dumdums walking around, at times
simple ore
mel loss is whether the generated audio spectrogram matches the original
kl loss is whether the generator can make a correct prediction about the signal
that affects inference
glacial pollen
Some style transferring or copying I suppose, if I understand the last part of your msg right that is