so flux dev would work on it then you are saying
#✨│ai-help
dark ginkgo
dark ginkgo
@simple ore I dont see these options on mine
also, when ever I try to change the sd vae to ae, it changes back
as in?
yes, but it also needs the sd_text_encoder in the options
it wont let me switch
see here #✨│ai-help message
also looiking it up, I have not found any infor regarding automatic 1111 being compatible. Everyone say to use forge or comyfui but I dont have any tutorial online to setup on a1111
I changed it to that already but also still no text encoder option
like I cant find this in the list of options
can you verify you are indeed using a1111 and not forge? a1111 has been abandoned from what I read so far with zero updates
as of 2024
furthermore, it said flux is incomaptible with a1111 and had no plan to update
red imp
can anyone help me, this keeps popping up when I try to use the MMVCServerSIO in the folder, I am on a m1 silicon mac
red imp
i switched to a updated os and it worked
but now for some reason when i try to select an output and monitor for the audio inside the voice changer, nothing shows up
idk how to use it
ohh ok thanks
it works
dark ginkgo
@simple ore So I just downloaded this, used the default what it comes with. Already having issues
SD.next, never heard of it until today
I have yet to touch anything
like it just finished downloading
I'm doing my usual prompt test for flux, 1024/1024
photograph of a red apple on wooden table, red apple, wooden table, high quality, professional photograph, dark background,
thick aurora
RVC v2 disconnected is working bro!
simple ore
<@155030383648440320> So I just downloaded this, used the default what it comes ...
anyway, if you have issues with sd.next, follow the link from their github page to discord
there's a good community there
light latch
Hi, can someone help me? I know how to look for AI cover models, but I don't have the link to make the voice work in an audio? I used Google Collab, has anything changed? Does anyone have a new link? Please answer.
elder coral
RVC v2 disconnected is working bro!
finally
quasi gyro
Did a new klm model drop or am I seeing things? Cause I saw that the thread was updated
simple ore
Did a new klm model drop or am I seeing things? Cause I saw that the thread was ...
yes
simple ore
quasi gyro
https://huggingface.co/SeoulStreamingStation/KLM6_Experimental/tree/main 3 da...
And it uses the spin embedd right? I can also just use Apollo or did he use code names fork?
simple ore
if you use a custom embedder option you can use any version
or applio exp branch
or the fork
modern surge
-colab
patent trellis
modern surge
I need a model to separate the drums of a song
Just the drums
Does anyone knows how to?
-uvr
quasi gyro
if you use a custom embedder option you can use any version
Yah but is it the spin embedd? Like what did he train it with is what I mean?
jovial kraken
I mean the RVC version by W-okala
latent kettle
I mean the RVC version by W-okala
RVC is different thing and w-okada is different. W-okada is a real-time voice changer. Which use RVC as base to work
peak path
I used the Colab version of AudioSep.
https://github.com/Audio-AGI/AudioSep
https://colab.research.google.com/github/Audio-AGI/AudioSep/blob/main/AudioSep_Colab.ipynb
But I've got an error.
Is there an active maintained version out there?
import torch
from pipeline import build_audiosep, separate_audio
device = torch.device('cpu')
model = build_audiosep(
config_yaml='config/audiosep_base.yaml',
checkpoint_path=str(models[0][1]),
device=device)
audio_file = 'zand.wav'
text = 'water drops'
output_file='separated_audio.wav'
# AudioSep processes the audio at 32 kHz sampling rate
separate_audio(model, audio_file, text, output_file, device)
simple ore
Yah but is it the spin embedd? Like what did he train it with is what I mean?
yes, spin7-12 layers
simple ore
depending on the version of torch, it is either a warning, or it simply did not load the weights at all. All torch.load calls need to be provided with weights_only=True
hazy eagle
i use firefox how to fix this
everytime i choose my microphone on client that pops up
simple ore
everytime i choose my microphone on client that pops up
if you're running it locally, use sever mode
or change the mic settings in the sound contol panel
hazy eagle
if you're running it locally, use sever mode
i don't want to use server because i think client is better
i tried this on many browsers but chrome gives best quality, i cant test it on firefox because of that error
knotty moth
i don't want to use server because i think client is better
server mode allows using WASAPI devices which have less latency than MME devices that are used in client mode
beside that the built-in noise suppression (Sup2) can only be used in client mode
quiet axle
hi was wondering what crossfade setting u guys usin, or just using the default?
knotty moth
hi was wondering what crossfade setting u guys usin, or just using the default?
the default
quiet axle
oh thanks, btw i was wondering. So i just reinstalled my windows cause of a certain problems. For some reason my vc doesnt sound like it used to
got some advice you reckon i could do?
i feel like the voice seems to abruptly end on a sentence more so then usual
knotty moth
i feel like the voice seems to abruptly end on a sentence more so then usual
increase chunk settings according to gpu performance
quiet axle
which protocol is better sio or rest? whats the different anyway
knotty moth
which protocol is better sio or rest? whats the different anyway
Protocol: rest (Use SIO if you want less delay but if you encounter any issues with SIO switch back to rest. Rest has slightly more delay than SIO)
quiet axle
oh i see, thanks for information
simple ore
how to make my index file have added_
you dont need to rename it.. usually a proper .index is included, regardles of what it is actually named
elder coral
you dont need to rename it.. usually a proper .index is included, regardles of w...
because my model will get rejected
knotty moth
because my model will get rejected
not really, we know Applio names it YourModel.index
elder coral
yes
knotty moth
and it works in any rvc applications
simple ore
very old apps called index files like 'added_IVF1406_Flat_nprobe_1_modelname_v2.index' for no good reason
knotty moth
the mainline and older versions name it like that, also spawns total_fea.npy and trained_*.index
summer reef
Read this guide first before making models
I can't understand this document. I want to find a current video and download it locally. When I tried it on the cloud before, there were problems like discornnect error.
elder coral
I can't understand this document. I want to find a current video and download it...
-colab
atomic goblet
I was thinking of getting a dedicated gpu for the voice changer, does anyone think a 5060 Ti would be good to handle that?
remote steeple
how to make voice changer sound not robotic
simple ore
I was thinking of getting a dedicated gpu for the voice changer, does anyone thi...
you can even get somethig cheaper if you wanna use a dedicated card
atomic goblet
you can even get somethig cheaper if you wanna use a dedicated card
What GPU were you thinking?
simple ore
What GPU were you thinking?
what's your main gpu?
atomic goblet
3090 Ti
simple ore
you can find like 3050 used for $100-150
just need to make sure you have a free 8x pcie slot and 2 slots for the card itself
atomic goblet
Hmm, that can work. I can try to find one then
forest jolt
hi whats the latest fork or program for rvc gui
i tried to use rvc ai cover maker and that its not sounding as good as rvc gui
hazy eagle
server mode allows using WASAPI devices which have less latency than MME devices...
so its better to move on server mode?
sleek sand
I need to make a Russian girl's voice out of my male one. which model should I use and how should I configure it? I use AI-Voice Changer
undone abyss
My command prompt closes automatically after opening start.bat anyone know a fix? I've downloaded PyTorch and Python
latent kettle
My command prompt closes automatically after opening start.bat anyone know a fix...
What you are using?
undone abyss
Windows 11 I have an intel gpu
latent kettle
Applio, w-okada?
undone abyss
w okada
latent kettle
Have you downloaded correct version from guide?
latent kettle
-realtime
latent kettle
i can try to check which one i downlaoded
undone abyss
thank you
undone abyss
https://github.com/deiteris/voice-changer/releases/download/b2332/voice-changer-...
do i delete the one i have rn
and install this?
latent kettle
Probably if it is wrong version
undone abyss
okay thank you
what do i do with the file you gave me
@latent kettle
do i extract
or like
what do i do
latent kettle
do i extract
Yes extract it
undone abyss
OMG ITS DOWNLOADING THANK YOU SO MUCH
youre the best @latent kettle
appreiciate it
latent kettle
Yw glad to help you 
undone abyss
@latent kettle one last question, how do i add a voice to the voice changer. I can't find a tutorial on YouTube to help me because the tutorials show w okada via app and im on a website
latent kettle
<@1174561195102056459> one last question, how do i add a voice to the voice chan...
See the edit button. Click on it and add model to any slot
@undone abyss
viral mason
pth files goes to model and index goes to index
undone abyss
tyty
undone abyss
pth files goes to model and index goes to index
alr thank ypu
i appreiciate both of you guys' help
viral mason
no problem ^^
my dms are always open if anyone has questions about setting up the okada voice changer
undone abyss
@viral mason I just dmed you and discord took away my permissions to dm, do you know why?
viral mason
<@1023278814752677918> I just dmed you and discord took away my permissions to d...
idk did you block me?
I couldn't message you either
undone abyss
i think someone went on my account and blocked ppl
i reset my cookie'
idk what happened it was weird
i texted you
and then discord logged me out of everything the second i did
luckily i reset the cookie in time before stuff happened
but that was rlly weird
undone abyss
https://github.com/deiteris/voice-changer/releases/download/b2332/voice-changer-...
was it bc i downloaded this?
undone abyss
ok do I can't dm u for some reason but
okay ty
latent kettle
was it bc i downloaded this?
Voice changer for intel gpu
latent kettle
I mean w-okada for intel
viral mason
I can't send pictures
latent kettle
The one you was using before, it was designed for Nvidia
viral mason
I'm gonna hang myself
undone abyss
I can't send pictures
thats so weird
undone abyss
I'm gonna hang myself
donttt
latent kettle
Yep
viral mason
I need image perms?
latent kettle
Delete that
viral mason
or somthing
latent kettle
I need image perms?
Nop
viral mason
wtf is wrong with discord rn then
undone abyss
discord being weird
latent kettle
wtf is wrong with discord rn then
Idk. I'm able to send images. We have Same roles
viral mason
ugh
viral mason
w okada is delayed when i talk, anyone know how to fix?
that's normal
viral mason
latent kettle
Congratulations 🎊
viral mason
@inner pivot
undone abyss
top one?
undone abyss
i cant reopen mmvcserveriosio it says failed to excecute script 'client' due to unhandled exception
nvm i got it
ill look at setting explination now @latent kettle
latent kettle
ill look at setting explination now <@1174561195102056459>
Good luck
undone abyss
thank you
viral mason
ill look at setting explination now <@1174561195102056459>
good luck ^^
I'll try helping if I can but I'm about to burst into a ball of angry juice
undone abyss
I'll try helping if I can but I'm about to burst into a ball of angry juice
yeah, i feel you. voice changer is lowkey complicated. and idk why you cant send images 😭
undone abyss
good luck ^^
thank uu
full marsh
same kaggle showing firbase problem
full marsh
colab dead now kaggle ..nice
full marsh
it's not saving tho
yeah that's the firebase issue
full marsh
it loads for me too but won't work
no unfortunately until the creators themselves fix the issue
viral mason
do they fix it quickly usually orrrr
bc I kinda need to keep training the model I'm working on
full marsh
do they fix it quickly usually orrrr
u need to try after few hours cause that issue is firebase...so kaggle might be doing it
full marsh
see if it's fixed after few hours or wait for the notebook creators fr a bettr solution
full marsh
u can always train locally if u have gd hardware
viral mason
I don't like local training, it messed with my vr
full marsh
i only use local fr inference training takes toll on my pc
viral mason
that's fair
kaggle was having issues like not too long ago was it not?
is someone attacking them or
viral mason
what
full marsh
what
gotta wait b4 getting GPU
u can get free immediate only CPU tho
i'm waiting too
full marsh
<:misc_cry:1176674698629750975>
going slow 😦
tough fiber
its stuck like this is this collab broken ? i used this 1 hour ago but now
full marsh
its stuck like this is this collab broken ? i used this 1 hour ago but now
which colab is this
full marsh
Finalllyyyy
winter dew
is https://applio.org/ the thing to use to make models??
snow vine
crystal pine
How do you make modles??
astral jungle
viscid moss
what AI stem do you guys use
it depends on what u want
viscid moss
Does anyone know how to use the voicechanger?
-rt
terse halo
<@723551098681688085>
can you Dm?
viscid moss
can you Dm?
I'm about to sleep tbh. U just need to download the version according to u GPU
viscid moss
Do it after removing instruments
umbral breach
Hey there, is the 9070XT capable of running inference or training on windows? Ive been trying to get it running on Applio with Zluda with no luck. Or should I just give up? lol
viscid moss
Reverb, then backvocals and finally denoise
simple ore
Hey there, is the 9070XT capable of running inference or training on windows? Iv...
there is a fresh experimental built of native pytorch for rocm that supports 9070 on windows, zluda is not required
unfortunately by default applio install comes with python 3.10 and the rock requires python 3.11 or 3.12 (applio does not not work with 3.12)
to use zluda you can see https://github.com/IAHispano/Applio/issues/1005
umbral breach
there is a fresh experimental built of native pytorch for rocm that supports 907...
sweet
simple ore
to use python 3.11 with Applio you basically need to nuke env folder and make a fresh venv using separately installed python3.11 and use pip install -r requirements.txt
without conda
or maybe change the install script to use python3.11 conda
that's another possibility
once applio is installed, install the experimental wheels
umbral breach
to use zluda you can see https://github.com/IAHispano/Applio/issues/1005
i've tried this but still no luck
simple ore
i've tried this but still no luck
"no luck" does not give me anything to investigate
umbral breach
"no luck" does not give me anything to investigate
ah sorry, I'll try the 3.11 method that u mentioned when i get some time, I'm still new to all this so thank you for the heads up
simple ore
ah sorry, I'll try the 3.11 method that u mentioned when i get some time, I'm st...
I have not tried the rock, you may need some beta of HIP SDK 6.5 that is not available yet.... never mind, should work with 6.2.4
but the method from the applio ticket should work
umbral breach
oh okay, so it is possible its just me missing something...
I'll give it a go, thanks for the help
terse halo
I tried to download it yesterday but I found it very confusing and it gave an error, can anyone help me?
latent kettle
I tried to download it yesterday but I found it very confusing and it gave an er...
Hii
Which GPU do you have
terse halo
Which GPU do you have
RTX 3060
latent kettle
Okay
terse halo
What I'm confused about is that there is a MB and a 2GB version
terse halo
https://huggingface.co/Shadicti/deiteris-Fork/blob/main/voice-changer-windows-nv...
Just download this or is there more?
latent kettle
After that you need virtual cable.
Virtual cable to connect your games or discord with Voice Changer
terse halo
Just download this or is there more?
After downloading this, where do I click?
latent kettle
Do You want real-time voice changer?
terse halo
Do You want real-time voice changer?
Yes
terse halo
After I extract, where do I click?
latent kettle
After I extract, where do I click?
After the download, you extract the zip file. You open the folders until you see an exe application called MMVCServerSIO and run that.
latent kettle
It's already downloading, and then I'll come back to ask, ok?
Maybe Okay
terse halo
link?
terse halo
After the download, you extract the zip file. You open the folders until you see...
Is it normal to say that there is a virus?
latent kettle
I think it's better to ignore it
latent kettle
Okay start and give mic permissions
terse halo
I haven't closed it yet
I already added the voice, how do I configure it so that the voice is synchronized?
latent kettle
I already added the voice, how do I configure it so that the voice is synchroniz...
Synchronized with what?
terse halo
Virtual cable to connect your games or discord with Voice Changer
link?
terse halo
donwloand
latent kettle
After installing the Virtual Cable, it changes your default audio system. Click Yes when it asks you to open the audio device settings (or press WIN+R, type "mmsys.cpl" if you closed it already), and change your Recording and Playback devices back to your usual devices. Same for communications device aswell (right click -> set as default communication device)
latent kettle
You must select virtual cable as output in voice changer and your mic as input in voice changer
terse halo
You must select virtual cable as output in voice changer and your mic as input i...
like this?
latent kettle
Yep
terse halo
like go back to original?
.
latent kettle
Check ?
ocean holly
I got a question about w okada, or well, AI voice changers in general (hopefuly this is the right channel) Is there a software, app, or a general way to let the voice changer pick up my voice, while my IRL surroundings are playing like normal? for example, I use the voice in a vc, and theres a knock on my door, people can hear the knock from my mic, but also the AI voice.
final path
Hey guys I need someone to guide me to properly train a model, I'm a bit familiar with the process but I can still use some help
viscid moss
leave it as default
olive bear
tes
vapid gorge
Una imagen de dios en el espacio
winter dew
https://applio.org/ is this the site for making models
scarlet fulcrum
how to change or fix delay?
viscid moss
try batch 8
analog obsidian
yeah the description of the batch size is wrong
the number depends in how big the dataset is
since u got 1 hour i'd recommend 8 or 16, with 8 being safer here
knotty moth
vram is just a constraint, obviously not affecting the results
analog obsidian
20 minutes or less = batch 4
30 minutes and above = 8
knotty moth
not only that, it also depends on the dataset diversity
if it's too diverse, you might better split and train separate datasets
analog obsidian
if it's too diverse, you might better split and train separate datasets
?
why lol
analog obsidian
not only that, it also depends on the dataset diversity
not really, i got a very monotone 5 hour dataset, batch 16 was worse than batch 32
knotty moth
it's still okay, unless you have multiple sources having different quality
knotty moth
the safe bet is to use a single source, or try normalizing each source
analog obsidian
the only really bad thing for datasets is inconsistent quality and whispering
8, 6 is too low for 1 hour
i train big datasets
so i kinda know what it's best for them
analog obsidian
not only that, it also depends on the dataset diversity
^ repeated words, very monotone speech, no expressions
knotty moth
still wrong
the thing is, I tried separating screaming parts for trying on metal vocals
analog obsidian
just use 8
for anything above 2 hours you may wanna try batch 16
and for 5 hours 32 is good
the more data u add, the more realistic the output, just sayin
analog obsidian
the thing is, I tried separating screaming parts for trying on metal vocals
i already told u singing is not the way to test stuff
knotty moth
not really, i got a very monotone 5 hour dataset, batch 16 was worse than batch...
so you tried bs 32 with checkpointing on a 16 GB gpu?
analog obsidian
so you tried bs 32 with checkpointing on a 16 GB gpu?
no checkpointing, bf16, 24gb vram gpu
yeah basically
another tip for best results: use spin, single scale loss
viscid moss
knotty moth
i already told u singing is not the way to test stuff
not everyone expect ideal results for that, if you can combine inference results to remove robotic sounds, it just works
knotty moth
no checkpointing, bf16, 24gb vram gpu
I thought tf32/fp32 are preferrable to it
but well bf16 as well as fp16 allow using AMP
analog obsidian
so applio has this new branch named f0_spin, it introduced two game changing stuff: a new embedder, and they brought back the original's rvc way to calculate mel
spin handles breaths better than cvec (the default embedder)
back then applio dev added a new way to calculate mel named multi-scale, which is great but not intended to be used in rvc/hifigan, so it was found that adds ringing to the models due to that single scale was brought back, using it should give you a model with very little ringing/no ringing at all
analog obsidian
I thought tf32/fp32 are preferrable to it
bf16 works just fine
you need this pretrain in order to use spin: https://huggingface.co/Aznamir/spin/blob/main/f0G32k_spin7-12.pth https://huggingface.co/Aznamir/spin/blob/main/f0D32k_spin7-12.pth (download g and d, and place them inside the "custom pretraineds" folder)
rvc > train > train.py
multiscale_mel_loss = False
disabling multiscale mel slightly reduces the vocal range of your model so remember that in case you wanna train a singing dataset
rvc > models > pretraineds > custom
i forgot it wasn't named custom pretraineds anymore lmao
if this is too complicated for u, you can just ignore it anyway
yeah actually it's somewhat easy, finetuning in rvc is not really a hard task
uh weird, redownload the zip
analog obsidian
rvc > train > train.py
multiscale_mel_loss = False
disabling multiscale mel sli...
forgot to mention this
d_ste_per_g_step = 2
rvc's discriminator is pretty piss, this was added to make it less bad
should give a less robotic model and better breaths
website? you mean colab?
dont run it as admin
just double click it
knotty moth
rvc > train > train.py
multiscale_mel_loss = False
disabling multiscale mel sli...
looking forward to seeing it come to the stable release
rn I'm still rather conservative against the new stuffs including spin and things like that, as my current one "just works"
and I don't think you should recommend it to commoners, yet
analog obsidian
if they dont like it they can use the old stuff
just because u dont like it doesnt means is bad
it's actually a great update
literally what rvc-boss intended to do back then
a new embedder
no lmao that aint a virus relax
training is so light on the gpu
u can literally play games while training
xD
u gotta need python 3.10 or 3.11 tho
this more advanced approach do need a few gigs of space tho
knotty moth
why being paranoid of it if you have good cooling system
knotty moth
I assume you don't play with any overclocking yet, so it should be safe, even in the furmark stress test
analog obsidian
did u open it as admin again?
knotty moth
well, have you done manual install with the latest torch and cuda 12.8?
the current compiled release one only works on RTX 40-series/older
analog obsidian
i think they already added support for 50xx in the branch
its fine bro
ignore the error
librosa being cringe for no reason
open the url in ur browser
analog obsidian
open the url in ur browser
dis
well there u go
congrats u installed applio yay
🏆
lmaooo
enjoy super fast training speeds now
yea
just train locally, it's better
kaggle is piss bad
u want a tutorial on how to use applio locally
?
ok so u did the two steps, applio is installed
download the pretrain i gave u, place where i told u to place it
place ur datasets inside the assets > datasets folder (or you can do it somewhere else, it doesnt matter lol)
knotty moth
at this point you might better sell your 5090 to any folks wanting it so bad and knowing what to do
knotty moth
that was way overpriced, recently it has quite dropped
analog obsidian
then manually place the location of the dataset like this
yeah
use auto slice if you haven't truncated the silence of your dataset
yea but like
your audio still has silence
rvc kinda hates that
so use auto slicer
yes
multiscale thing?
more natural results
knotty moth
if you want step by step guidance, this convo should be continued in a new thread in https://discord.com/channels/1159260121998827560/1192011222023950368
analog obsidian
so like simple words
applio by default added a thing that boosts your dataset voice range at the cost of ringing (a static sound while singing high notes)
if u dont want that
u can disable it
inside your model's log folder
everything should be there, index, g, d, ur epochs
graphs
and finally, save every 10 epochs if u wanna save some disk space
nah just make a thread here > #1192011222023950368
void holly
anyone know how remilia bandxz makes his voice like that?
crude flame
Erm actually wavlm may be better than spin (ignoring the breaths)
analog obsidian
Erm actually wavlm may be better than spin (ignoring the breaths)
breaths are crucial for realtime
crude flame
breaths are crucial for realtime <:yt_nails:1159569314848972891>
Just don't breathe 
analog obsidian
astral void
@left sentinel
forest jolt
how to de harmony a track with uvr5?
manic olive
I have an applio error saying that no api found and I am using version 3.2.3
knotty moth
I have an applio error saying that no api found and I am using version 3.2.3
try getting the latest version 3.2.9
manic olive
try getting the latest version 3.2.9
I installed the new version and it shows me the error file not found requirements.txt
polar grail
yo, ive been tryna setup this ai shit for years on a shitty laptop that couldnt run all the downloads. Im back with an actual PC, someone want to help me set it up an explain how it works? im a dumbass and would prolly need help in a voicecall
latent kettle
@simple ore sorry to bother you but can you explain why this is happning like my g avg loss in decreasing but the G total loss is increasing.
simple ore
<@155030383648440320> sorry to bother you but can you explain why this is happni...
dont use search, just expand loss_avg_50
what batch size?
latent kettle
what batch size?
13 minutes of dataset and 4 batch size
elder coral
why isn't mainline working
flint anvil
chat i am looking at the available pitch extraction and i asked gpt deepresearch to figure out what's better for realistic m2f voice conversion and it says crepe-full is better but it gives it more delay - and i notice the deiteris fork and i think wokada also have crepe-full without onnx so i assume its gpu bound
I have a really good gpu and i only have to run the voice changer, so no games and stuff, so is crepe-full better than rmvpe for pitch extraction if i can afford to run it?
urban dune
help, Why only works Beatrice, how to make it work and rvc
agile kelp
Guys from your experience , i tried many models , can anyone suggest me a model that cannot look like its an AI talking ? i tried many couldnt find the perfect one
urban dune
only one installed when launched, the other one doesn't download, I don't know why when the program starts
knotty moth
help, Why only works Beatrice, how to make it work and rvc
ye the og wokada is good for beatrice only
better get one of these:
https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/
https://docs.aihub.gg/rvc-voice-changer/local/vonovox/ (Nvidia support only yet)
urban dune
I mean, everything worked for me before, after resetting Windows, when I start HTTP, only Beatrice is installed, without RVC, and nothing works
maybe it is possible to download RVC separately and everything that should be downloaded automatically
digital cairn
hey guys with experience, what is better, deiteris w-okada or vonovox? in terms of delay and quality of output
neat meadow
hi Could you let me know what the most current/latest version is right now? Please send me the link
brittle wing
Can anyone tell me how to add beats to an acoustic song using ai 😢
novel radish
Does anyone know of a model for cloning a very expressive voice with a lot of vocal voice! NOT LIKE ANUEL AA, BAD BUNNY. They don't have a voice! SOMETHING FOR MANELE FOR AN ARTIST NAMED FLORIN SALAM! HELP! I NEED SOMETHING BETTER THAN HIFI-GAN AND MORE REALISTIC WHEN CLONING SINGING VOICES
low shard
Does anyone know of a model for cloning a very expressive voice with a lot of vo...
elaborate:
- ur pc gpu
- what u want to do (pre-record or realtime)
also, remindn yourself that it depends on how the model was trained, and you need to kinda voice act yourself
low shard
hi Could you let me know what the most current/latest version is right now? Plea...
of what? what do you want to do?
what's ur pc gpu?
low shard
hey guys with experience, what is better, deiteris w-okada or vonovox? in terms ...
kinda similar, heard @crude flame say it was slightly better for nvidia gpus in terms of delay
elaborate:
- ur pc gpu
- what u want to do
- what u exactly mean
novel radish
I have a 1070 Nvidia, I want something better than Hifi-gan.. my cloned voices with this method are good but I want something better that feels more real when cloning these expressive voices, look for FLORIN SALAM AND YOU WILL SEE THAT HE HAS A LARGE VOICE! not like Anuel AA or Bad Bunny
crude flame
kinda similar, heard <@673327878288703519> say it was slightly better for nvidia...
I love how my phone autocorrected vonovox to bonobo. Anyway yeah vonovox is better in delay and same quality
low shard
I have a 1070 Nvidia, I want something better than Hifi-gan.. my cloned voices w...
a gtx 1070 isnt that great, are you looking to train models or use them in realtime?
want something better than Hifi-gan
RVC has limits, it can't 100% pass for human, for example it sucks at laughing
crude flame
I have a 1070 Nvidia, I want something better than Hifi-gan.. my cloned voices w...
Vits1 problem not hifi
novel radish
I'm looking to train models! Better than Hifi-Gan! I train singing artist models.
I have applio rvc wit HIFI-GAN
neat meadow
of what? what do you want to do?
what's ur pc gpu?
5600 and rtx3060
low shard
I'm looking to train models! Better than Hifi-Gan! I train singing artist models...
RVC has limits, all you can do is try to train better and do voice acting
low shard
5600 and rtx3060
download what? what do you want to do? also wdym with 5600?
neat meadow
download what? what do you want to do? also wdym with 5600?
ryzen 5600 cpu
novel radish
RVC has limits, all you can do is try to train better and do voice acting
And how do I train better? Or tell me exactly what to do to improve as much as possible.
neat meadow
download what? what do you want to do? also wdym with 5600?
voice changer
also i use at2020 and scarlet solo 4th
low shard
voice changer
realtime for calls right?
neat meadow
yes
low shard
And how do I train better? Or tell me exactly what to do to improve as much as p...
you can try checking the suggestions in the docs https://docs.aihub.gg/, but as i said, you will NEVER get something that is 100% like a human
novel radish
Ok thanks
low shard
@neat meadow RVC = Retrieval-based-Voice-Conversion, the best Few Shots Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models. Technically, Mainline RVC does have a go-realtime.bat (aka RVC-GUI), but it's pretty messy and outdated so it's extremely not suggested for realtime.
Wokada = uses RVC for realtime inference. There's 2 main versions, Original made by Wok, and the most suggested one is Deiteris Fork (modified version)
what you'd need is wokada deiteris fork
neat meadow
what is fork version? What's the difference from the main version?
neat meadow
<@1248980356779085955> RVC = Retrieval-based-Voice-Conversion, the best Few Shot...
thx
low shard
what is fork version? What's the difference from the main version?
better quality and performance
neat meadow
better quality and performance
fork version is better?
low shard
wokada deiteris fork b2332 is the latest, there's also another program which might slightly be better but only for nvidia gpus https://docs.aihub.gg/rvc-voice-changer/local/vonovox/ and i haven't tested it personally
low shard
fork version is better?
yes
neat meadow
how to download fork version?
low shard
how to download fork version?
100% you should
share a screenshot of your current wokada so i can check the version, or just the folder name
neat meadow
appriciate
neat meadow
It corrects the accent of the model. Using it generates a lot of cpu usage
i like zero
idk formant Is it a new feature
low shard
you can decide how much index to do
higher value means more trained index is used, but can sound a bit more autotune
honest cedar
Can anyone help me with voicemeeter related issues?
I’m trying to bitcrush my rvc model
hallow thistle
High risk of damage? That ususally happens when your CPU overheats, it can happen when you overclock or your PC simply has a bad cooling system, regardless of any process running. It also depends on your PC CPU and such.
Promoting your Discord server or stuff is not allowed in this server. It's obvious.
hollow tartan
High risk of damage? That ususally happens when your CPU overheats, it can happe...
Mine is running on my cpu for a long time and theres been no problem.
hollow tartan
Promoting your Discord server or stuff is not allowed in this server. It's obvio...
What do i do? I cant take the bot outside tho. How do i find people to test it?
hallow thistle
honest cedar
@crude flame Are you able to help my voicemeeter, codeman is asking if you can help me out
crude flame
<@673327878288703519> Are you able to help my voicemeeter, codeman is asking if ...
Do you have light host installed
honest cedar
Do you have light host installed
Yes it’s mainly just voicemeeter
It’s like working with blender all over again 🥲
crude flame
Then what's your problem
honest cedar
Then what's your problem
i cant hear anything when im running voicemeeter
and it says fader grain for all the sliders which doesnt match up with the picture
crude flame
i cant hear anything when im running voicemeeter
Like can't hear game audio can't hear the voice or can't hear anything at all
honest cedar
Like can't hear game audio can't hear the voice or can't hear anything at all
everything
no audio anywhere
crude flame
Did you fix your output audio in Windows
brittle wing
Applio no UI is not working again
honest cedar
Did you fix your output audio in Windows
im not sure how to setup my windows audio since the guide doesnt show me that
and theres like 30 input output options
low shard
don't advertise please
low shard
Applio no UI is not working again
what's ur pc gpu? whats wrong?
crude flame
and theres like 30 input output options
Select your headphones for Windows output
brittle wing
what's ur pc gpu? whats wrong?
Android user
low shard
Android user
elaborate the issue
honest cedar
Select your headphones for Windows output
okay im going to attempt it
brittle wing
elaborate the issue
Wait it's Colab is the code broken again.
"Pkg_resources is deperecated as an API"
brittle wing
elaborate the issue
See...
Vedi
low shard
"Pkg_resources is deperecated as an API"
send a screenshot
!give-media-perms 1h @brittle wing
brittle wing
Ah no prob it's training through
honest cedar
Select your headphones for Windows output
still cant hear anything
crude flame
So you have your windows output as your headphones and in voicemeeter you have your hardware out a1 as your headphones
honest cedar
So you have your windows output as your headphones and in voicemeeter you have y...
yes exactly
crude flame
Did you restart your PC after installing voicemeeter
crude flame
On your virtual input try selecting a1 for both
crude flame
Idk then, that always works for me
honest cedar
damn so much for bitcrushing
crude flame
Idk then, that always works for me
You could try setting voicemeeter input and aux as default comms device and default sound device
honest cedar
You could try setting voicemeeter input and aux as default comms device and defa...
ill try it
honest cedar
You could try setting voicemeeter input and aux as default comms device and defa...
that didnt work but i switched my stereo input 1 to A1 and it worked for some reason
crude flame
Wild
honest cedar
so random
now i got to see if it even works with rvc
@crude flame and what would i set input and output as?
crude flame
<@673327878288703519> and what would i set input and output as?
Input your mic
Output line 1
honest cedar
Input your mic
Output line 1
im not hearing my model
digital cairn
im not hearing my model
Put your headphones/speakers into monitor
digital cairn
try restarting the whole client
honest cedar
okay
crude flame
this isnt doing anything
Did you click start
honest cedar
Did you click start
yeah i did click start
crude flame
Did you select a voice
honest cedar
Did you select a voice
is it supposed to be picking up my desktop audio?
crude flame
is it supposed to be picking up my desktop audio?
No
honest cedar
thats definitely a problem then
brittle wing
brittle wing
How do I decipher this?
honest cedar
someone else had told me you dont need to but dont take my word for it
latent kettle
It's your choice model will work with and without index
paper bloom
hey is there anyway to make owakada voicechanger more adapable to the english accent?
like sometimes it pronounces words diffrently
brittle wing
so how did it go
Normal like everything's fine
prime shell
how i can download applio?
simple ore
how i can download applio?
prime shell
brittle wing
https://huggingface.co/IAHispano/Applio/tree/main/Compiled/Windows
Another Applio no UI error/issue while train
Cannot load file containing pickled data
-colab
brittle wing
https://huggingface.co/IAHispano/Applio/tree/main/Compiled/Windows
Do you have any workaround for this?
Seems to be an issue with Numpy...yes
swift thunder
Does anyone know if Applio Colab is down? It stays here and doesn't advance: Starting backup loop... Files are up to date
brittle wing
Does anyone know if Applio Colab is down? It stays here and doesn't advance: `St...
Same I can't train
swift thunder
Same I can't train
must have fallen
swift thunder
What singer
Korean
swift thunder
I know but which artist
Sooin (MEOVV) was going to train but it's no good applio haha
brittle wing
Sooin (MEOVV) was going to train but it's no good applio haha
Ah okay so we're not training the same person lemme tag @simple ore for help
Applio no UI is unusable
Am I doing resuming wrong?
swift thunder
Applio no UI is unusable
I don't like it, it's very complicated haha
brittle wing
I don't like it, it's very complicated haha
It's easy 4 me
swift thunder
It's easy 4 me
Currently I like the most up-to-date, that's why I use Applio UI.
No way, I have to wait to see if they can fix it.
brittle wing
Currently I like the most up-to-date, that's why I use Applio UI.
Idk how to train there.
crisp vault
brittle wing
Colabs are broken again...
brittle wing
'Pkg_resources is deprecated as an API'
simple ore
'Pkg_resources is deprecated as an API'
so far it is a warning
simple ore
Which colab?
brittle wing
Which colab?
Applio no UI but with workaround
I tried the no workaround one too still the same outputs in code
simple ore
what workaround?
brittle wing
what workaround?
The code fixes you suggested in winter
simple ore
that was a long time ago, fixed the install like 3 times after that
brittle wing
brittle wing
that was a long time ago, fixed the install like 3 times after that
So I shouldn't use the colab notebook I added this stuff on?
brittle wing
simple ore
the link above installed just fine, no errors or warnings
brittle wing
https://colab.research.google.com/github/iahispano/applio/blob/master/assets/App...
I get the same errors here too
simple ore
screenshot
brittle wing
the link above installed just fine, no errors or warnings
Uh wait
simple ore
brittle wing
brittle wing
How...
Warning?But my model doesn't want to train it gets stuck on "data not found?"
brittle wing
extracted features?
In what sense?I should do that before loading backup after switching accounts?
brittle wing
Im cinfused
simple ore
Did you actualy load the backup, are all the files under logs/modelname?
simple ore
my spidey sense is tingling
simple ore
why you checking mute folder instead of your model's folder... which is not visible there
you said you used the URL i gave you, but did not load the backup?
brittle wing
you said you used the URL i gave you, but did not load the backup?
Model folder is on drive
Uh I'm still installing.
simple ore
but it is not in logs
brittle wing
😭
swift thunder
Don't forget about Applio UI too, please.
brittle wing
load backup copies files from your google drive to logs
How I do that???
I know the copies are there
simple ore
swift thunder
is this the applio ui or no ui?
brittle wing
Now ?
brittle wing
is this the applio ui or no ui?
No UI one
brittle wing
Done now ...?
Last cell?
Still getting the same warning @simple ore
That's the issue!
simple ore
yes, it is warning
brittle wing
yes, it is warning
Look how do I fix that
simple ore
ugh
brittle wing
"No data left in file"
brittle wing
try `!uv pip install librosa==0.11.0 --upgrade`
Oki
simple ore
run that, then your training
brittle wing
run that, then your training
ImportError: numpy.core.multiarray failed to import
Still can't train
Honestly
simple ore
you're killing me bub
simple ore
and i'm trying to watch a movie
brittle wing
and i'm trying to watch a movie
I understand but just help me fix the error quickly cause I loaded backup & ran the LAST cell
Did everything over & over wasted 2 hours
simple ore
i'm gonna check locally
brittle wing
i'm gonna check locally
Oki
simple ore
now I'll check on colab
brittle wing
now I'll check on colab
Pls do
simple ore
after you ran 'install'
preprocess and extract featiures works fine with librosa 0.11.0
simple ore
connec to drive, clone, install, +extra cells for librosa, load models, load backup, set training variables (must use the same model name and dr), then training
simple ore
that's only to hide those annoying warnings
brittle wing
After install comes the new cell?
simple ore
make a new cell, see above
brittle wing
try `!uv pip install librosa==0.11.0 --upgrade`
Done
Numpy array Multy array failed to import.
Ah wait backup complete,files are up to date
simple ore
I have no idea.,. maybe your backup is fked
brittle wing
I have no idea.,. maybe your backup is fked
In what sense?
That could be
How do I fix it.
simple ore
I've tested restore from backup and resume training and it works for me.
brittle wing
I've tested restore from backup and resume training and it works for me.
Restore from backup...
undone abyss
im using w okada rn, and everytime i talk it gets super laggy and the ms spikes rlly high, anyone know a fix? or a way to reduce the lag
golden belfry
Hello! Im just curious as to how i could fix my issue. whenever i press start on a voice, it keeps saying “Frequent errors occur. Please check if the model of the framework being targeted is loaded.”
swift thunder
Hi, Applio UI won't load my datasets. Can you help me?
open raft
I’m new to voice-conversion and excited to explore RVC! I’ve read through the README and glanced at the code in model.py and inference.py, but I’m not sure where the “core” algorithm is implemented, and how all the pieces fit together.
What I’d love to know:
Which files or classes handle the feature extraction and model architecture?
Where is the training loop defined, and how do data preprocessing and postprocessing hook in?
Are there any papers, blog posts, or diagrams you recommend for a high-level overview?
Any in-code comments or tutorials aimed at beginners that I should read first?
I’m eager to learn and eventually contribute—thanks in advance for any guidance! 🙏
crude flame
I’m new to voice-conversion and excited to explore RVC! I’ve read through the RE...
have you tried looking in extract.py, train.py, and preprocess.py for feature extraction, preprocessing, and the training code
open raft
have you tried looking in extract.py, train.py, and preprocess.py for feature ex...
I’ve reviewed the suggested modules, but my goal is to build an intuitive mental model before diving back into the code.
quasi gyro
@simple ore for the klm v3 model, is it hifi gan?
wise token
why has every model become extremely high pitched even on 12 tune
simple ore
<@155030383648440320> for the klm v3 model, is it hifi gan?
i imagine klm v3 is old af
wise token
i imagine klm v3 is old af
do yk why?
quasi gyro
i imagine klm v3 is old af
I mean the pretrain @tight ether made
the exp 3 pretrain for klm
its hifi right?
quasi gyro
klm 6.1 v3 is hifigan spin7-12
thank u
wise token
can someone help me
simple ore
can someone help me
pitch +12 increases the input audio's pitch up
quasi gyro
@simple ore is there a way to enable fp16 on apollo
so i can have faster training with my 4090
simple ore
so i can have faster training with my 4090
faster is not better
4090 is fine crunching fp32
quasi gyro
faster is not better
Figured it out anyway, need faster for the project Im doing, doesn't have to be perfect
simple ore
fp32 is not slow
i have 55 hour set doing 24min/epoch on 4070 with fp32
with fp16 it would probably be 20?
bronze hedge
yo guys, why does my changer take like 30 seconds to process, and even after it processes it cuts off and sounds horrible?
hallow thistle
Do you? Using an index model in W-Okada is not really recommended, as doing so will use more resources to process, potentially reducing the overall performance.
knotty moth
Figured it out anyway, need faster for the project Im doing, doesn't have to be ...
fp16 has exploding gradients issue, to mitigate it switch to bf16 or fp32 (tf32)
sterile hawk
so i set up my inputs and stuff with mic and cable input but cant hear anything and when i try to test in discord the yt video is playing in my mic test. anyone know whats wrong?
hallow thistle
so i set up my inputs and stuff with mic and cable input but cant hear anything ...
!howtoask
hallow thistle
Which W-Okada version are you using? Are you using VB-Cable or Virtual Audio Cable? And what is your PC GPU?
sterile hawk
Which W-Okada version are you using? Are you using VB-Cable or Virtual Audio Cab...
it says v.1.5.3.18a, i am using VB-Cable and i have a NVIDIA GTX 1070
hallow thistle
it says v.1.5.3.18a, i am using VB-Cable and i have a NVIDIA GTX 1070
Download and use this better W-Okada instead. Yours is old and outdated. And make sure to try Virtual Audio Cable lite instead of VB-Cable. https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#download-nvidia-on-windows
sterile hawk
Download and use this better W-Okada instead. Yours is old and outdated. And mak...
okay, thank you very much
flint anvil
how do i edit the deiteris fork on my windows pc? i install it but it fails to install faiss-gpu because its linux only and i cant even build this on gh actions to get a working source (no idea how to reproduce)
im trying to change the source code
hallow thistle
okay, thank you very much
Let me know if you have issue about delayed audio and low quality. 
hallow thistle
how do i edit the deiteris fork on my windows pc? i install it but it fails to i...
Which Detris fork W-Okada did you install? I don't think this fork W-Okada would do that.
flint anvil
Which Detris fork W-Okada did you install? I don't think this fork W-Okada would...
the latest b2332 for nvidia cuda on win
hallow thistle
the latest b2332 for nvidia cuda on win
If you wanna modify fork W-Okada, there's a GitHub for that, if you know how to do so. https://github.com/deiteris/voice-changer
flint anvil
If you wanna modify fork W-Okada, there's a GitHub for that, if you know how to ...
i think you're not understanding - i downloaded this and tried to start it from source
but python wont install faiss-gpu on win
hallow thistle
However, there's a compiled one there in this guide, this one doesn't need to be installed from start. https://docs.aihub.gg/rvc-voice-changer/local/deiteris-w-okada-fork/#download-nvidia-on-windows
flint anvil
However, there's a compiled one there in this guide, this one doesn't need to be...
yeah thats not useful to me, i want to write my own better fork as its not being updated
but i cant work from source on win11 because faiss gpu doesnt exist for w11
hallow thistle
but i cant work from source on win11 because faiss gpu doesnt exist for w11
Do you mean you want fork W-Okada to work on Linux? Sorry, but I still can't figure out what you want to do with this.
flint anvil
Do you mean you want fork W-Okada to work on Linux? Sorry, but I still can't fig...
i want to edit the deiteris fork on windows
i want to edit the code for it
but its not possible to build a working version for gpu (i would explain why but itll just confuse us more)
hallow thistle
So what makes you think b2332 is outdated? Sorry, I don't do coding, but this version is the most stable out there. Any attempt to upgrade one of Python components can cause some other components to conflict each other, I've tried it with other Python-related programs.
flint anvil
So what makes you think b2332 is outdated? Sorry, I don't do coding, but this ve...
i need to write postprocessing sfx, fix the wokada gui as its confusing in parts, remove outdated stuff like fcpe, and clean up the code
it has nothing to do with the quality itself
but if you cant help me because this is a programming question can you point me to someone who can?
hallow thistle
but if you cant help me because this is a programming question can you point me ...
Anyone with the "Engineer" role. But @wispy lodge is the author of the fork program.
quiet axle
so which one is better vonovox or deiteris fork?
hallow thistle
so which one is better vonovox or deiteris fork?
Deteris' fork W-Okada.
quiet axle
cool thx for the fast response
do any of you guys use voicemeeter or just use VAC lite and call it a day?
simple ore
do any of you guys use voicemeeter or just use VAC lite and call it a day?
virtual audio cable
voicemeeter is lil tricky to route
quiet axle
but is it better?
hallow thistle
What does this mean? Although you can use index file in "regular RVC program", Applio for example, using index in W-Okada is still not really recommended anyway. As what I said.
quiet axle
oh okay, i was reading the realism section of the docs and they recommend using voicemeeter
hallow thistle
do any of you guys use voicemeeter or just use VAC lite and call it a day?
Some might use Voicemeeter as a second virtual line to Virtual Audio Cable lite, but that's it. VAC lite is still recommended, I use this one.
quiet axle
oh okayy thanks for the info
knotty moth
do any of you guys use voicemeeter or just use VAC lite and call it a day?
voicemeeter is only for audio routing, though there is lighthost as alternative
but ffs not the virtual cable
quiet axle
Whats lighthost?
knotty moth
the guide should have mentioned it
tawdry matrix
yo i need help with a voice changer
hallow thistle
yo i need help with a voice changer
!howtoask
brittle wing
what f0 det do you use for tesla T4 gpu?
hallow thistle
what f0 det do you use for tesla T4 gpu?
NVIDIA T4 in which website? Is it Kaggle or Google Colab? And which notebook are you running?
Also, read the "How To Troubleshoot" above before you ask anything here.
brittle wing
NVIDIA T4 in which website? Is it Kaggle or Google Colab? And which notebook are...
kaggle, im running https://www.kaggle.com/code/suneku/voice-changer-public
My current f0 works (rvmpe_onnx) but i was wondering if theres a better one
the guide doesnt say anything about t4 gpu
knotty moth
the guide doesnt say anything about t4 gpu
it is Nvidia gpu
hallow thistle
My current f0 works (rvmpe_onnx) but i was wondering if theres a better one
The F0 Det on W-Okada, if you're using NVIDIA GPU there, always select regular "rmvpe". The rmvpe is for NVIDIA GPU, while rmvpe_onnx is for non-NVIDIA GPU (AMD and Intel).
knotty moth
what f0 det do you use for tesla T4 gpu?
the performance is around RTX 3050 but with 16 GB vram
kaggle has dual gpu but it will use a single gpu anyway
brittle wing
it is Nvidia gpu
Oh I didn't know, thank you guys
I thought by "tesla" they meant like elon musks tesla and I was so confused
knotty moth
I thought by "tesla" they meant like elon musks tesla and I was so confused
it is Nvidia Tesla lineup
hallow thistle
NVIDIA Tesla GPU and Tesla the company are two different things, the NVIDIA Tesla is now called NVIDIA Data Center GPU for newer GPUs within series.
hallow thistle
Almost every RVC model has index file alongside with it, but some only provide just pth file. Not really that surprising. An index file stores the accent of that voice model.It can be created during voice model training in RVC program.
candid osprey
403 ERROR
The request could not be satisfied. for links lol.
I tried vpn it doesn't work.
tawdry shore
so i have the app on my phone can i still do the contest?
ancient fable
WHat are good AI voice apps
hallow thistle
WHat are good AI voice apps
Is it pre-recorded audio converter or realtime voice changer? There are plenty of AI voice apps available. One of them being RVC.
ancient fable
What are good RVCs?
hallow thistle
What are good RVCs?
I use Weights.com mostly for fast-accessing AI cover. But there's Applio, which is available as locally (PC) and online.
RVC refers to AI programs that can do voice convert and voice model training, but as what I said, there are many different programs of it, which one of them being Applio.
simple ore
I tried vpn it doesn't work.
use a better vpn
brittle wing
use a better vpn
I still have the same issue as yesterday.Would you mind if I dm you my drive folder with backups and you see where's the problem
NO DATA LEFT IN FILE...AGAIN
I CAN'T TRAIN NY MODEL
Should I start all over...
With another account
hallow thistle
Do I need to explain it again?
If you don't remember what I said earlier, let me say to you again. Using index in W-Okada is not recommended, as it will cause it to use more performance. While you can use index in regular RVC program, yes, but that's all.
analog obsidian
the voice changer app is running a rvc model in realtime
rvc does not stand for realtime voice changer, they're two separate things, rvc originally only works for local conversions and don't support realtime inside the webui
every rvc model is compatible with the .index files (yeah you can use any .index file with any model), although index files in realtime cause several issues and their usage in those conditions is not recommended, pick any .index file, set the index value to 0 and forget about its existence
analog obsidian
i think w-okada forces you to select a index file but im not sure, regardless, setting the index value to 0 will disable the index
golden walrus
Anyone know if KLM is good for real time? 
It sounds so good in these sample
Spin is a breakthrough for me
analog obsidian
Anyone know if KLM is good for real time? <:misc_smoke_cry:1159570646519521363>
og pretrain is better for speech
for spin i'd recommend noobies pretrain instead, but the grads are a bit high, not sure why
spin only ^ doesn't work with cvec
remember spin is still experimental, for a more safer approach, use the original pretrain and contentvec
stone lynx
do u guys know where i can find how to create ai voice guide
low shard
do u guys know where i can find how to create ai voice guide
what's ur pc gpu?
stone lynx
what's ur pc gpu?
rx 6700 xt
sonic night
hi, i wanted to use rvc to put an audio file and then corvert it to another audio file with an AI voice, idk how
low shard
rx 6700 xt
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- Applio (AMD Windows) : A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
- Mainline (AMD Linux/Windows) : The original RVC
- Cloud (remote good pc, easier and faster than ur PC but it's limited):
- Weights.com: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio Colab: max 4 hours daily, not granted, of GPU
Easiest possible : weights.com
easiest cloud: Ilaria rvc zero
easiest local: Applio
low shard
hi, i wanted to use rvc to put an audio file and then corvert it to another audi...
what's your pc gpu?
brittle wing
Ok
i made a environment folder inside coquis ai tts repo folder
and installed it
but how do i open it?
I don't see anything
Documentation on the github says to make a python script, but i wanted a GUI of the app
i don't wanna use the command console...
How do i open coquis ai tts?
sacred marten
hi
i have a question
where do you put these voice models in?
like what software
low shard
where do you put these voice models in?
what's your pc gpu? what do you want to do?
latent cypress
does rtx 4060 8gb good enough for training image models locally? weights queue is taking way too long
median monolith
is it normal that the "Start" cell on Applio Kaggle takes a lot of time to even give me the links? its been like 10+ minutes and it still says "cell execution is queued".
median monolith
now +30 minutes
azure patio
which model is made to separate backing vocals from vocals?
brittle wing
8gb in 2025 is a big mistake
Maybe my backup got messed up cause I tried to clone the main repository
brittle wing
which model is made to separate backing vocals from vocals?
Mel and Becruily karaoke also DM me the song you want separated I can do it for you by using uvronline's backing vocals separator
azure patio
Mel and Becruily karaoke also DM me the song you want separated I can do it for ...
thank you, the thing is i have quite a lot because im getting the vocals to create a model
brittle wing
thank you, the thing is i have quite a lot because im getting the vocals to crea...
Well you can use it yourself
azure patio
Mel and Becruily karaoke also DM me the song you want separated I can do it for ...
you mean this one?
brittle wing
Well use that model on mvsep
azure patio
cause i tried it and it gave just vocals and instrumentals(which were empty since i have only vocals already)
brittle wing
Yes but use big beta 6x by unwa on the music source separation colab for Acapella first.
azure patio
i have acapella already
brittle wing
i have acapella already
Use th model on the Acapella
brittle wing
you mean this one?
Yes
azure patio
thank you 
btw i did the acapella using the thing from kimberleyjsn, is that much different?
dire zinc
median monolith
Im starting to get kinda insane, because idk wtf is wrong with my dataset for applio not being able to train it. the logs keep saying "Not enough data present in the training set. Perhaps you forgot to slice the audio files in preprocess?"
simple ore
Im starting to get kinda insane, because idk wtf is wrong with my dataset for ap...
show the log of preprocess and extract features step
golden walrus
remember spin is still experimental, for a more safer approach, use the original...
I mean, spin has better quality in real time in Vonovox. Also it can produce my language tone a bit better. 
Let me try Noobies' pretrain
median monolith
show the log of preprocess and extract features step
analog obsidian
I mean, spin has better quality in real time in Vonovox. Also it can produce my ...
yeah spin is great 
simple ore
well, it failed to extract f0 and features
for some reason
i've tested noUI colab yesterday and it was fine
median monolith
If this helps, these are the options I put.
median monolith
well, it failed to extract f0 and features
the dataset/wav file is only 1:03 minutes long
simple ore
1 minute audio wont fit into any training buckets
who gave you this idea?
start again with a new model name and use simple slicing
make sure you get your ~30 files in extract features processed
median monolith
i mean, the dataset has little to no silence already, and so I thought that it was no necessary to cut it even more (let it as it is)
simple ore
you need to slice it, it has nothing to do with silences
median monolith
i thought it was basically going to remove parts of the audio and make it even shorter if i choose any other option, my bad
simple ore
automatic slicing does remove excessive silences, simple just shreds the file into digestible chunks
median monolith
automatic slicing does remove excessive silences, simple just shreds the file in...
well, in that case, i wonder what are the recommended values thay i should put for both "Chunk length (sec)" and "Overlap length (sec)" if i choose the simple option, for such a short dataset that at best has like a second of actual silence. maybe the default values are enough and will not "cut parts of the audio and make it lose content"?
simple ore
use default 3/0.3
median monolith
use default 3/0.3
alr, then the audio should be let intact (not lose/cut/remove content/information) 👍
simple ore
if you have enough silence, actual 0 level silence, you can set mute files to 0
median monolith
if you have enough silence, actual 0 level silence, you can set mute files to 0
i suppose the "silence" on the very start and very end should be that ?
opal depot
so for kaggle w-okada, how do I have persistence for files apply without having to do save version every time? do I run in edit mode or
median monolith
make sure you get your ~30 files in extract features processed
ok, so, i guess it did went better than last time, but the logs still say the not enough data thing, and instead of ~30 files, it gave me 22. same values, just changed the model name and the audio cutting to simple with default values.
simple ore
what's the batch size?
median monolith
what's the batch size?
4
median monolith
what's in f0?
simple ore
is it kaggle or colab?
median monolith
is it kaggle or colab?
kaggle
median monolith
try batch 2
alr, i guess its finally going!
simple ore
dont expect much of anything from just a minute of audio
median monolith
ik lol
signal chasm
Not sure why