I'm using Applio NoUI, I don't know what the pretrain type is selected. What is the model currently training as? @low shard
#✨│ai-help
low shard
I'm using Applio NoUI, I don't know what the pretrain type is selected. What is ...
i think it's just using original since you didn't turn on custom pretrained
perhaps it's better to ask @nocturne mural since he made that notebook
vapid mantle
Oh good then. It is better to use the original
nocturne mural
Original, TITAN, Ov2 Super?
if you want to use other pretraining, you must activate the custom_pretrained checkbox, otherwise the original pretraining will be used
vapid mantle
if you want to use other pretraining, you must activate the custom_pretrained ch...
I want to use the original anyway
I tried 2 of them and didn't like them
TITAN, Ov2 Super
The model says some letters roboticly
low shard
Oh good then. It is better to use the original
I mean it kinda depends, it depends on your dataset language and lenght
nocturne mural
fixed
low shard
#📰│dev-updates message @tawdry spade @unique rock @glass igloo @white bough
hallow thistle
Dev announcement about Google Colab just dropped.
tame mica
like that wasnt the message nick sent literally before yours ? 😭
hallow thistle
Here's the English-translated of it. https://multimedia.easeus.com/images/multimedia/voice-changer/resources/w-okada-client-server-architecture.jpg
This is the highest quality image I could find.
I don't know what MMVC does stand for. But for VC, I think it stands for voice changer.
I can think for the name of it: W-Okada.
low shard
Where can I download the newest rvc voice changer?
realtime for calls? and what's ur pc gpu
low shard
Or is w Okada better?
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
simple ore
I think this happens when you resume training
you did not stop at the actual saved epoch, so it stepped back
river trout
Is applio main better or codename fork better?
simple ore
Is applio main better or codename fork better?
same base, codename's has some power user features for training
analog obsidian
I think this happens when you resume training
you have to stop the training at the moment rvc saves the epoch
so you can resume from that epoch avoiding that problem
but its just a visual bug
the model itself is fine
tawdry stirrup
how to overcome this Welcome to ColabMod
Timer: 00:00:23DEPRECATION: omegaconf 2.0.6 has a non-standard dependency specifier PyYAML>=5.1.. pip 24.0 will enforce this behaviour change. A possible replacement is to upgrade to a newer version of omegaconf or contact the author to suggest that they release a version with a conforming dependency specifiers. Discussion can be found at https://github.com/pypa/pip/issues/12063
Timer: 00:00:24warning: The --system flag has no effect, a system Python interpreter is always used in uv venv
Using CPython 3.10.12 interpreter at: /usr/bin/python3.10
Creating virtual environment at: .venv
Activate with: source .venv/bin/activate
Timer: 00:00:45DEPRECATION: omegaconf 2.0.6 has a non-standard dependency specifier PyYAML>=5.1.. pip 24.0 will enforce this behaviour change. A possible replacement is to upgrade to a newer version of omegaconf or contact the author to suggest that they release a version with a conforming dependency specifiers. Discussion can be found at https://github.com/pypa/pip/issues/12063
Cloning the repository...
edgy tangle
the model itself is fine
oof, i thought that the model died, thx
edgy tangle
you did not stop at the actual saved epoch, so it stepped back
yes, I already suspected it
analog obsidian
as long you resumed the training using the same batch size, yeah nothing to worry about
edgy tangle
Well, I leave it at 4, since my RTX 3050 only has 4GB of VRAM
river trout
same base, codename's has some power user features for training
Are those features effective? I tried to check this channel, but I couldn’t find much information about them
analog obsidian
Are those features effective? I tried to check this channel, but I couldn’t find...
warmup epochs were proven to give better results than without using it
but only for adamw if i remember well
currently both applio and the fork uses radam
and that already does warmup by itself (don't enable warmup epochs in the UI atm, its meant to be used with adamw)
besides that uhh
its just applio
ah and the fork has the mel spectogram similarity metric, i forgot about that 🦈
distant hamlet
oh I'm not the only one then..
yaaa 😔
dusk sphinx
Hina's rvc on colab errors like this also
Timer: 00:03:28/content/voice-changer/server/HVoice.py:3: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
from distutils.util import strtobool
Traceback (most recent call last):
File "/content/voice-changer/server/HVoice.py", line 10, in <module>
from downloader.SampleDownloader import downloadInitialSamples
File "/content/voice-changer/server/downloader/SampleDownloader.py", line 12, in <module>
from voice_changer.RVC.RVCModelSlotGenerator import RVCModelSlotGenerator
File "/content/voice-changer/server/voice_changer/RVC/RVCModelSlotGenerator.py", line 4, in <module>
import torch
ModuleNotFoundError: No module named 'torch'
WARNING:pyngrok.process.ngrok:t=2025-02-27T21:20:22+0000 lvl=warn msg="Stopping forwarder" name=http-46499-3f12fb39-2175-40bc-8140-83858962dbee acceptErr="failed to accept connection: Listener closed"
--------- SERVER STOPPED! ---------
simple ore
uv is broken on colab, so it does not install nothing
fervent rover
So y’all are saying that the mainline Colab is not working?
Okay, so I guess I just wait until you guys fixed the mainline Colab y’all
knotty moth
how to overcome this Welcome to ColabMod
Timer: 00:00:23DEPRECATION: omegaconf 2...
you dont
brittle wing
-colab
karmic olive
ashen badger
Is there any guide to making ai covers and stuff that’s updated and is there anything I can use for text to speech (I think this is the correct channel, correct if wrong)
river trout
ah and the fork has the mel spectogram similarity metric, i forgot about that 🦈
thanks bro
low shard
Is there any guide to making ai covers and stuff that’s updated and is there any...
what’s ur pc gpu?
weary peak
FIxed old issue on collab if you run out of time but it trained enough epochs how do you download the file because all of them are not visible even to the download script
quaint radish
Hina's rvc on colab errors like this also
Timer: 00:03:28/content/voice-changer...
same here
fervent rover
Okay Guys, I’m going to test out RVC Mainline since I saw those guys saying that the RVC Mainline Colab Is Not Working, so I’m going to tried it out for myself, I will kept you guys posting for updates to know if it worked or not
Okay NeverMind
quaint radish
Okay NeverMind
its fixed?
which one is this?
jaunty mason
-rvc
AI HUB Docs
fervent rover
its fixed?
which one is this?
RVC Mainline Colab
quaint radish
RVC Mainline Colab
https://rentry.co/RVC-Mainline-Colab
this one?
fervent rover
It hasn’t working according to one of em on discord
fervent rover
https://rentry.co/RVC-Mainline-Colab
this one?
Yes
abstract flame
-colab
low shard
check #📰│dev-updates
Also Womada isn't rvc
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
Use #🔍│help-w-okada
cerulean hedge
gt 1030
i want to convert an audio file into someone else voice
i dont need training
low shard
i want to convert an audio file into someone else voice
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- 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.com: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio UI Colab: max 4 hours daily, not granted, of GPU
Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero
easiest local: Applio
simple ore
howw
tts with 0-shot voice cloning
fish speech, f5-tts, xtts v2 from coqui
depends on the language though, most tts are just english and chinese
jade rose
how do i make the ai voice more expressive in Applio? Cuz now it only reads the text like a robot. Should I use something else?
low shard
how do i make the ai voice more expressive in Applio? Cuz now it only reads the ...
it's normal, RVC is natively Speech to Speech (STS) not Text to Speech (TTS)
the way applio uses it for TTS is because they actually generate an audio first with Microsoft Edge TTS API, then, use that audio as an input in rvc
edge tts is multilingual and good quality, but not emotional
low shard
then what should i use
There are different Text To Speech (TTS) AIs:
GPT So Vits: RVC isn't as good as GPT So Vits for tts, but gpt so vits (few shot tts, which means needs just a lil training for models) can't use rvc models (and viceversa), and its only limited to: english, chinese & japanese, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/
Freemium 11labs: Easy way to do TTS is https://elevenlabs.io/, you can't use RVC model on this but its a mostly premium easy way for good quality TTS
FishSpeech: FishSpeech is a 0 shot (no explicit training needed) TTS, if you got a good pc you can use it locally else use their site
You can check TTS in our tts index
With RVC Models:
RVC is natively for Speech To Speech, but forks such as ilaria rvc mainline & applio have built in tts (using Microsoft Edge TTS to make a generated tts audio, which i suggest you to choose a tts model that is the same gender and language of the rvc model you wanna use, and then convert it with rvc)
If you wanna do tts locally with RVC Voice Models (if you got a good pc):
If you don't got a good pc you can do tts with RVC Voice Models on cloud:
-
Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide
-
Use Applio UI Colab (with google colab T4 free daily limit gpu)
-
You could try another tts from our tts index and use the output as an input in rvc
The best way would be using 11labs tbh, but it's paid
else you could give gpt so vits, f5 tts, fish speech, a try
jade rose
Ok, thanks
weak cipher
tts with 0-shot voice cloning
i am using edge tts so can it do this
low shard
i am using edge tts so can it do this
no, edge tts can't do this
I explained it above and also sent a message about tts
check it out
simple ore
i am using edge tts so can it do this
Edge TTS is a screen reader, plain and simple, very neutral, no emotions
lost lagoon
RVC S2S
noble jay
hello guys am i at the right place to ask a question ?
dusk sphinx
Anybody got restricted by Colab ?
low shard
Anybody got restricted by Colab ?
Elaborate:
- ur PC GPU
- what google colab are u using
- what restriction? Show a screenshot
dusk sphinx
This account has been blocked from accessing Colab runtimes due to suspected abusive activity. This does not impact access to other Google products. If you believe this action was taken in error, review the usage limits and appeal . @low shard
low shard
https://colab.research.google.com/github/deiteris/voice-changer/blob/master-cust...
@wispy lodge you have to check your colab code, this is the 2nd report now :nails:
low shard
This account has been blocked from accessing Colab runtimes due to suspected abu...
Sorry but I can't do much about it, it's better you don't use wokada deiteris' fork on colab rn
U could try the Kaggle or do it locally if u got a good PC gpu
Also next time use #🔍│help-w-okada
viral flame
how to fix
/content/voice-changer/server/HVoice.py:3: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
from distutils.util import strtobool
Traceback (most recent call last):
File "/content/voice-changer/server/HVoice.py", line 10, in <module>
from downloader.SampleDownloader import downloadInitialSamples
File "/content/voice-changer/server/downloader/SampleDownloader.py", line 12, in <module>
from voice_changer.RVC.RVCModelSlotGenerator import RVCModelSlotGenerator
File "/content/voice-changer/server/voice_changer/RVC/RVCModelSlotGenerator.py", line 4, in <module>
import torch
ModuleNotFoundError: No module named 'torch'
WARNING:pyngrok.process.ngrok:t=2025-03-01T01:16:00+0000 lvl=warn msg="Stopping forwarder" name=http-40611-85bb3119-0fa0-4dba-a7f3-4e73487e3dc0 acceptErr="failed to accept connection: Listener closed"
--------- SERVER STOPPED! ---------```colab is fked
flint seal
does anyone know how to get one speaker from an audio file? the speakers aren't overlapping but i just want a file with one person talking and don't want to do it manually
simple ore
who's gonna sort the speakers out if you dont want to do it manually? some magic?
rain urchin
Is The RVC Mainline Colab Fixed?
nope i have the same issue and it isnt fixed yet :/
fervent rover
nope i have the same issue and it isnt fixed yet :/
Oh Okay
paper pasture
guys idk which voice changer im using but im assuming its this one
start_http is taking too long to load
anyone know why
rain urchin
how to fix
```mv: cannot stat 'MMVCServerSIO.py': No such file or directory
/co...
do you see also this warning? if you do we cant do anything cus the server owner needs to update the python version if im right
simple ore
uv is messed up
also there's no compatible version of faiss-cpu for python 3.11
you either need to downgrade the environment to 3.10 or change the version to install to the one supporting 3.11
1.7.4 supports 3.11
@rain urchin
fading vault
how to start in kaggle
tame mica
ctrl + k
hallow thistle
This account has been blocked from accessing Colab runtimes due to suspected abu...
Your Google account has been terminated from using services in Google Colab, unfortunately. All you can do is to try another cloud service like Kaggle or wait for the better PC you can run locally.
low shard
how to fix
```mv: cannot stat 'MMVCServerSIO.py': No such file or directory
/co...
This is the wrong channel
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
Tell: your PC GPU, the google colab link you're using in #🔍│help-w-okada
low shard
start_http is taking too long to load
Wrong channel
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
Tell your PC GPU in #🔍│help-w-okada
hallow thistle
A lot of people keep mistaken RVC for realtime voice changer.
low shard
do you see also this warning? if you do we cant do anything cus the server owner...
Wrong channel
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
Tell your PC GPU, and google colab link in #🔍│help-w-okada
And the colabs are made by engineers, it's not like the server owner can own like 20 colabs
RVC IS NOT REALTIME VOICE CHANGING, WOKADA USES RVC FOR REALTIME, SO USE #🔍│help-w-okada
hallow thistle
I've been telling people to go to #🔍│help-w-okada if they wanna talk anything about W-Okada, for many times. Sure, I get it not everyone knows what RVC and W-Okada the realtime voice changer even are. But if they read more enough words instead of just one line, they should've been able to figure it out by themselves.
weak cipher
Have any of guys tried Livekit?
halcyon barn
Hi I'm getting this error "ModuleNotFoundError: No module named 'gradio" on chrome browser, how to fix this?
low shard
Hi I'm getting this error "ModuleNotFoundError: No module named 'gradio" on chro...
tell:
your pc gpu
what guide link are u following
and what u want to do
You can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab -> GPUYou can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab -> GPU
halcyon barn
I am using Hina_Mod_AICoverGen_colab on chrome browser windows 10
low shard
I am using Hina_Mod_AICoverGen_colab on chrome browser windows 10
it's broken like most colabs, check #📰│dev-updates
also I asked for the PC GPU not the Operative System nor browser
halcyon barn
any replacement of that?
low shard
if you got a good pc gpu you could run RVC locally (runs on your pc) instead of cloud (runs on remote good pc) services like colab
halcyon barn
Cool
brave swallow
guys no matter what I keep getting that noise in the end !
I'm using applio rvc inference
brave swallow
what could be I'm doing wrong
simple ore
what could be I'm doing wrong
what did you use for training? was it 24-bit audio?
did you clean it up?
did you clean the inference audio?
brave swallow
what did you use for training? was it 24-bit audio?
16 bit
brave swallow
did you clean it up?
both cleaned up
you can notice here this part fixed but the other broken after changing Pitch extraction algorithm
simple ore
post the source audio for inference
simple ore
brave swallow
is it fixable or nope
simple ore
it is separated vocal or something?
simple ore
well, bad separation then
use mvsep
and if the source before separation was mp3 then it is even worse
brave swallow
and if the source before separation was mp3 then it is even worse
the source was wav
simple ore
holes say it was lossy compression
brave swallow
this the original
brave swallow
holes say it was lossy compression
from where can I view this
simple ore
well, whatever method you used for cleaning did mess it up and added that blip at the end
knotty moth
why only the last part is aligned?
simple ore
but why would you remove silences for inference?
brave swallow
or what is the best example for good cleaning
brave swallow
but why would you remove silences for inference?
I thought this could make static noise
brave swallow
you could just instead generate silence on parts to "delete"
where is that ?
simple ore
denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt
simple ore
I use this repo https://github.com/ZFTurbo/Music-Source-Separation-Training
just place it into models
and then select on UI
swift patio
how to make it compatable
brave swallow
how to make it compatable
to make what ?
low shard
that is voice.ai, we don't offer support to that, it's literally paywalled wokada
also, this isn't the right channel, I literally explained oyu everything 2 mins ago in #🔍│help-w-okada
knotty moth
why do u still use that garbage
brave swallow
denoise_mel_band_roformer_aufr33_aggr_sdr_27.9768.ckpt
where do I find config.yaml for this one
brave swallow
and if the source before separation was mp3 then it is even worse
I figured out it was mp3
congrats 😄
simple ore
where do I find config.yaml for this one
brave swallow
almost 2H
brave swallow
https://huggingface.co/jarredou/aufr33_MelBand_Denoise/tree/main
yeah done that
idk if there's way to make it use gpu too
do I need to run this clean up on both , training dataset and inference audio ?
so I have to train my module in RVC over ?
brave swallow
for some reason it keeps running over cpu
simple ore
for some reason it keeps running over cpu
check which torch got installed by default
may need to replace it with cuda version
requirements does not have cuda index, so it most likely got 2.0.1 cpu installed
weak cipher
Guys I want anime tts to have a free api, is that possible?
brave swallow
simple ore
for some reason it keeps running over cpu
torch installed for denoise
brave swallow
import torch
torch.cuda.is_available()
#True
torch.cuda.device_count()
#1
torch.cuda.current_device()
#0
torch.cuda.get_device_name(0)
brave swallow
activate the environment
conda ?
simple ore
whatever that is
simple ore
pip install torch==2.3.1 torchaudio==2.3.1 --upgrade --index-url https://download.pytorch.org/whl/cu121
brave swallow
oh cool then
simple ore
but now you have cuda torch and torchaudio
brave swallow
yep its seems it works
cuda:0
thanks alot for helping
I hope it works and fix my issues
I been trying for long time
does this looks better ?
meh same issue still
I'm at https://discord.com/channels/1159260121998827560/1307339969743818852 if someone can help I can share the screen
crude bolt
where can i find pretrains
low shard
where can i find pretrains
analog obsidian
found : Rigel
this is undertrained
crude bolt
you got a better 1 ?
analog obsidian
you got a better 1 ?
for hifigan the original pretrain or https://discord.com/channels/1159260121998827560/1339155300720054316
crude bolt
hm ... not sure if refinegan is at its current state actually usable. No experience with it
analog obsidian
hm ... not sure if refinegan is at its current state actually usable. No experie...
results are comparable to hifigan results, got better harmonic reconstruction and less mirroring but people have been experiencing a buzzing sound in their models
refine has better singing range than hifi
analog obsidian
buuuut i don't personally use it because of the electric/metallic sound it gives to models
crude bolt
maybe in future those kinks will be ironned out.
simple ore
buuuut i don't personally use it because of the electric/metallic sound it gives...
perhaps not any more 😛
brave swallow
perhaps not any more 😛
so atm it doesn't has electric sound anymore ?
analog obsidian
perhaps not any more 😛
brave swallow
interesting
analog obsidian
i'm running another model test right now
iron cobalt
How can I prevent Ngrok from exceeding Data Transfer Out monthly limit using Applio on Kaggle? I'm facing only problems with training my voice models via this
idk how to decrease inbound connection volume without having to upgrade my account plan for additional capacity on ngrok
low shard
How can I prevent Ngrok from exceeding Data Transfer Out monthly limit using App...
I think the only ways are either using another tunnel (if the kaggle has the option for) or deleting and registering the account for 'refreshing' ur limit or making another acc
iron cobalt
For unknown reasons for me more than 300 MB data transferred out.
And this is just in early March 2025
pastel oak
will it work for training and inference?
Yes
reef pier
btw for those that have a gtx 1650 (like i do): inference will work. very slowly, though.
simple ore
btw for those that have a gtx 1650 (like i do): inference *will* work. very slow...
you can run inference on CPU, slow but it is what it is
reef pier
btw for those that have a gtx 1650 (like i do): inference *will* work. very slow...
at least, the very first time you infer, as it has to load the model. the rest of the times will be fast.
low shard
for example here
Tbh you can directly use that model on weights.com
You can just click create
Weights.com uses RVC in an easier user interface
summer mirage
hi, i'm training some models but my results are bad. can someone help me? this are my settings. i also tried to edit my pitch but that don't make sense.
compact tapir
cual es el mejor voice changer?
summer mirage
no
low shard
hi, i'm training some models but my results are bad. can someone help me? this ...
that RVC is extremely outdated
Mangio Fork is a fork (modified version) of Mainline RVC (the original project) which has been discontinued since 2023
summer mirage
oh
low shard
absolutely delete that, and never look youtube tuts for RVC/Wokada
what's your pc gpu?
summer mirage
what should i use then?
low shard
cual es el mejor voice changer?
We speak only english for this server, if you mean realtime voice changing, tell your pc in #🔍│help-w-okada
low shard
what should i use then?
what's your pc gpu?
summer mirage
a nvidia rtx 3060
summer mirage
what is the difrence?
low shard
what is the difrence?
between the 2 I said?
summer mirage
yes
low shard
Mainline is the original project of RVC
Applio is a fork, with an easier user interface that gets more updates
basically Applio is more maintained
summer mirage
Okey! i will try that! thank you
analog obsidian
what is the difrence?
mainline havent got a real update since 2 years (recent ones has been dependency fixes and not training related stuff)
applio in the other hand has new updated training code which can give faster and better results than mainline, its constantly getting new updates
summer mirage
a post says that i need to install it on my ssd, why is that?
analog obsidian
a post says that i need to install it on my ssd, why is that?
faster loading times and faster writing times if you want to train
low shard
SSDs 🙏
analog obsidian
rvc has to write two big files during training, if you train in your HDD is going to take a couple of seconds to write them (around 4 seconds)
but on a ssd is almost instant
it doesn't slow training speed but slows down the process a bit (basically the training will pause everytime the two big files are overwritten)
bc it has to wait until the files are written
besides that, it works just fine in a hdd
low shard
explained poorly 
summer mirage
i got this error after installing succesfull
tight beacon
Hi there, so I have trained some voices for TTS in the past, but I was thinking of trying to train the same voices for use in w-okada to use in DnD games. However the guide seems to say you need hours of clean voice samples for it to work... is this still the case?
(Most of my recent decent TTS ones have been done with about a minute of audio or less, mostly cos there isn't more than that available haha)
unique rock
how do i use kaggle?
analog obsidian
Hi there, so I have trained some voices for TTS in the past, but I was thinking ...
Speech to speech (RVC) is more complex than TTS hence why you require more data to have a good model
but anyways, you don't need hours worth of data
for realism you need around 30 minutes to 1 hour max of data
for ok results you need minimum 10 minutes
low shard
i got this error after installing succesfull
First of all, be sure to download it on the C drive
Also, you're using the precompiled from the guide I sent right
low shard
how do i use kaggle?
Which kaggle
tight beacon
Speech to speech (RVC) is more complex than TTS hence why you require more data ...
thanks, im going through the guides atm so i'll see what I can manage
unique rock
Which kaggle
colab, applio
low shard
colab, applio
You mean Kaggle applio UI? Bc u just said Kaggle then now you're saying colab
If so, check https://docs.aihub.gg/rvc/cloud/applio-kaggle/
crisp lynx
my output doesnt sound like the models. i have a 4070 super and i7 14700k. idk why its doing this
are my specs good enought to run this
knotty moth
are my specs good enought to run this
why do u blame the specs for that result? it's more of the model itself or pitch setting
knotty moth
explained poorly <:misc_trolley:1159468147133395025>
120gb in 2025
low shard
120gb in 2025
It was just an example, and 120gb are still used in 2025
Just in phones
-# more correctly, 128gb on phones
knotty moth
Just in phones <:misc_trolley:1159468147133395025>
phones dont use ssd, and not even the transfer speed
low shard
I was just talking about storage capacity lol
Anyways, it's just a reddit meme I googled 😭
knotty moth
it sounds like mid 10's
unique rock
Well, I'm using the Kaggle interface, haha, to use Applio within it, this GPU T4x2 thing.
low shard
it sounds like mid 10's
No, around 2020 https://www.reddit.com/r/memes/comments/j3ssp3/ssd_go_brrr/
-# yes I use reddit not only to search random shi
knotty moth
No, around 2020 https://www.reddit.com/r/memes/comments/j3ssp3/ssd_go_brrr/
500-1 tb ssds exist in 2020 (perhaps it was considered high end tier)
low shard
500-1 tb ssds exist in 2020 (perhaps it was considered high end tier)
I just thought of it as a silly meme showing ssds are faster but more expensive while hdds are great for larger memory but slower
knotty moth
No, around 2020 https://www.reddit.com/r/memes/comments/j3ssp3/ssd_go_brrr/
btw that means SSD 15GB and HDD 125GB lmao
hallow thistle
With the similar price, you'll get either a SSD with lower capacity but faster speed or a hard drive with larger capacity but lower speed. 
brittle wing
-colab
brittle wing
-rvc
brittle wing
-colab
carmine siren
worthy quartz
Can anyone else with the update to https://colab.research.google.com/drive/1mHKTGH5e3SAyDSBss1KtiYRbDdQzwSMs#scrollTo=9qpCkSOUCkFr and where im supposed to put everything that is required because im genuinely confused lollll
hallow thistle
Can anyone else with the update to https://colab.research.google.com/drive/1mHKT...
Are you trying to do RVC on Google Colab?
cold cave
I've been waiting for 6 minutes and it didn't even start training
worthy quartz
Are you trying to do RVC on Google Colab?
Yes I guess it hasn’t been updated in a while so I’m trying to use applio but it’s still pretty confusing
spice siren
If I cancel my annual membership, will I get my money back?
knotty moth
I've been waiting for 6 minutes and it didn't even start training
until you realize it said that the sample rate or architecture do not match the model settings applied since the preprocessing step
cold cave
until you realize it said that the sample rate or architecture do not match the ...
The samples are at 48K and Datasets are 48000
worthy quartz
Does anyone know how to use applio if so can I pm you because I’m a bit confused lol
cold cave
It's not that, its the custom_pretrained button and had to turn off pretrained
hallow thistle
If I cancel my annual membership, will I get my money back?
Cancel what? I don't think you would get refund for canceling a subscription, but the service would just let you use premium service you once paid in a year until it expires.
knotty moth
It's not that, its the custom_pretrained button and had to turn off pretrained
pls read carefully the pretrain sample rate and whether using refinegan or the default one
cold cave
pls read carefully the pretrain sample rate and whether using refinegan or the d...
the refinegan one
knotty moth
the refinegan one
make sure to use the latest applio repo or if not sure use the hifigan one and pretrain
worthy quartz
Does anyone know how to use applio if so can I pm you because I’m a bit confused lol
hallow thistle
Does anyone know how to use applio if so can I pm you because I’m a bit confused...
You sent your same message for the second time now.
worthy quartz
Thought no one saw sorry 😬
hallow thistle
spice siren
Want to talk to the website owner
worthy quartz
Okay so I’m trying to upload an existing voice model that I already made to applio so I can make an ai cover of a song but I do not know where to put everything at.
hallow thistle
Want to talk to the website owner
An owner of which website? And who are you asking to talk to the website owner?
distant turtle
-colab
lethal shale
Heloooo! I finally made my own female model and it sounds natural and amazing. But then I made singer's voice and it is metallic. I researched and found out it is because sample frequency rate.
There was a website to learn the frequency rate of an audio file. Can someone send me the link?
mint yew
what is good chunk and extra settings
lusty sun
Can anyone guide me how to install the voice model once I have downloaded it from voice-models .com ? I am using tortoise tts btw
low shard
Can anyone guide me how to install the voice model once I have downloaded it fr...
Tortoise tts can't use RVC models, those are 2 different type of AIs, RVC is STS
The only thing you could do, is make an audio with tortoise tts, then use that as an input in an RVC like Mainline or Applio
But tortoise is pretty old
There are different Text To Speech (TTS) AIs:
GPT So Vits: RVC isn't as good as GPT So Vits for tts, but gpt so vits (few shot tts, which means needs just a lil training for models) can't use rvc models (and viceversa), and its only limited to: english, chinese & japanese, if you wanna check gpt so vits instead, read https://docs.ai-hub.wtf/tts/gpt-sovits/
Freemium 11labs: Easy way to do TTS is https://elevenlabs.io/, you can't use RVC model on this but its a mostly premium easy way for good quality TTS
FishSpeech: FishSpeech is a 0 shot (no explicit training needed) TTS, if you got a good pc you can use it locally else use their site
You can check TTS in our tts index
With RVC Models:
RVC is natively for Speech To Speech, but forks such as ilaria rvc mainline & applio have built in tts (using Microsoft Edge TTS to make a generated tts audio, which i suggest you to choose a tts model that is the same gender and language of the rvc model you wanna use, and then convert it with rvc)
If you wanna do tts locally with RVC Voice Models (if you got a good pc):
If you don't got a good pc you can do tts with RVC Voice Models on cloud:
- Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide
- Use Applio UI Colab (with google colab T4 free daily limit gpu)
- You could try another tts from our tts index and use the output as an input in rvc
What's your PC GPU btw
low shard
what is good chunk and extra settings
Wrong channel
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
Show a screenshot of your WOKADA in #🔍│help-w-okada and be sure to not follow yt tuts
low shard
There was a website to learn the frequency rate of an audio file. Can someone se...
You can just download Spek on your PC
low shard
Can anyone else with the update to https://colab.research.google.com/drive/1mHKT...
Ilaria RVC mainline colab is outdated and broken since a year, it won't be fixed
low shard
Does anyone know how to use applio if so can I pm you because I’m a bit confused...
First of all, what's your PC GPU?
Also I just noticed you got a TCOAAL pfp, I played it too lol
lusty sun
Thanks man
lusty sun
What's your PC GPU btw
Nvidia 4060
low shard
Rtx 4060? Yeah you're good then
lusty sun
Yeah
lusty sun
Rtx 4060? Yeah you're good then
Could I DM you bro? I’m not that well equipped in terms of coding experience so your help would be great
low shard
Could I DM you bro? I’m not that well equipped in terms of coding experience so ...
No need for DMS, and not even to code
All you would need is just download Applio if u want the easy way
eager crown
-colab
maiden idol
Hi Help RVC Members,
I hope you’re all doing well. My name is Vikha, and I’ve been exploring voice conversion using Retrieval-Based Voice Conversion (RVC). I encountered an issue while trying to merge two PTH models—one trained for 200 epochs and the other for 150 epochs—into a 50-50 balanced blend. However, the resulting audio quality didn’t meet my expectations, and I’m unsure why the quality degraded despite the models being quite close in training epochs.
I’ve been experimenting with various fusion approaches, but I haven’t been able to achieve the desired results. I’m reaching out to you because I came across your profile and noticed your work in this field. I believe your insights could help me understand the potential issues that might be causing the problem in my fusion process.
If you have experience working with similar models or have any suggestions on improving the process, I would be extremely grateful for your guidance. Additionally, any resources, tutorials, or techniques you could share would be invaluable as I continue troubleshooting.
Thank you so much for taking the time to read this message. I hope we can connect!
stuck crypt
Hey bro, I need help with this issue.
When I input VB-Cable into Discord or other apps, my voice becomes choppy and sounds weird.
Does anyone know how to fix it? 😭
simple ore
Hey bro, I need help with this issue.
When I input VB-Cable into Discord or othe...
I think the recommendation is to use Virtual Cable Lite https://software.muzychenko.net/freeware/vac470lite.zip
low shard
Hey bro, I need help with this issue.
When I input VB-Cable into Discord or othe...
this is the wrong channel
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
I replied u there https://discord.com/channels/1159260121998827560/1345791557063802970
formal wind
I've seen so many tips on how to detect overtraining but I have no idea what is most effective. I've read the tutorials but I just want to be sure yk yk?
analog obsidian
I've seen so many tips on how to detect overtraining but I have no idea what is ...
hear the epochs
compare them
overtraining is very easy to hear
the model starts to sound robotic
formal wind
Thats what I used to do but It never really turned out well
analog obsidian
Thats what I used to do but It never really turned out well
why so?
formal wind
I just don't have a good ear when it comes to listening for overtraining
crude mist
hi, why does like every rvc model sound so weird when laughing and whats the solution?
crude flame
hi, why does like every rvc model sound so weird when laughing and whats the sol...
thats an rvc limitation
crude mist
so thats normal and theres no fix?
analog obsidian
hi, why does like every rvc model sound so weird when laughing and whats the sol...
there's no solution to that
crude mist
damn cuh thats unfortunate
crude flame
so thats normal and theres no fix?
atm there is no fix
crude mist
I guess I'll rp as a mentally unstable egirl with missing laugh muscles
analog obsidian
lmao
crude mist
lemme ask chatgpt what the condition is that makes u not able to laugh
Akinetic Mutism yup I got that
thats me
been had that
marble forge
anyone know why rvc won't launch?:
low shard
anyone know why rvc won't launch?:
holy shit you got an ancient version lmfao
mangio rvc fork is discontinued since 2023
what's your pc gpu?
marble forge
i have the 12gb one
simple ore
can train stuff
marble forge
simple ore
do you need a realtime voice changer or trainer/voice changer for files?
marble forge
yea ik im saying what is the new one
simple ore
that changes voices for calls/discrodd
marble forge
the one i have is hella old
simple ore
marble forge
is that the official one?
simple ore
best settings for it <@155030383648440320>
marble forge
wait it isnt real time
i want a realtime one
@simple ore
you got one or no?
also the one you sent doesnt allow pitch change
low shard
wait it isnt real time
wokada deiteris fork is realtime
wokada is a program which is better than the mainline rvc realtime, which is better than the mangio fork rvc realtime
and the deiteris fork is better than the original wokada
low shard
how do i fix it
I replied to u in #🔍│help-w-okada
cold cave
Everytime I use ApplioNoUI, my storage keeps getting full instantly.
That's because all these G and D files keep duplicating
low shard
Is The RVC Mainline Colab Working?
no, it will work when it will be added as fixed in #📰│dev-updates
tbh just use Appllio meanwhile
fervent rover
tbh just use Appllio meanwhile
Umm No Thanks
I just play the waiting game, I guess
low shard
Umm No Thanks
There's nothing wrong with applio, it just got a more user friendly interface and more updates
I don't really understand what's wrong with it, but your choice
low shard
alright
outer isle
If I train a singing voice model will it contains another dataset the same as RVC2 Disconnected on Applio?
simple ore
If I train a singing voice model will it contains another dataset the same as RV...
if you mean "can I train a voice model with a pretrain?" the answer is yes
outer isle
if you mean "can I train a voice model with a pretrain?" the answer is yes
I woyld have to start FromScratch
polar sage
Hello, can anyone tell me which colab they currently use to make AI COVERS?
warm oar
-colab
weary peak
sorry i keep asking for help, trying to retrain on collab, getting this error
NameError Traceback (most recent call last)
<ipython-input-2-5294ebea29b0> in <cell line: 0>()
26 print('Paste model link and try again!')
27
---> 28 if not os.path.exists(f'/content/sample_data/{Model_Name}.tar.gz'):
29 print("File not found.")
30 else:
NameError: name 'os' is not defined
mild sleet
Imagine there was a way you could train accents with a model
like for different artists
like an ai to train not just voice but another to train accents
knotty moth
sorry i keep asking for help, trying to retrain on collab, getting this error
N...
above it add import os
btw I don't recommend using random colabs not listed as above #✨│ai-help message
as it may not be coded or maintained well
simple ore
like an ai to train not just voice but another to train accents
that's what the index is for
hallow thistle
knotty moth
<#1159380240271953940>
it doesn't always mean to relate the problem. the colab author might forget to include import os in the cell unless the previously run cell that contains import os failed (in that case yea might relate to the problem you state)
still I suggest him adding import os to see if it works or not
low shard
it doesn't always mean to relate the problem. the colab author might forget to i...
I was just saying the colabs that are verified to work
white bough
Would you guys recommend to keep whispering in the model or would it mess with the training? I am afraid that if I do that, the whispering tone/voice will come out when it is not supposed to...
cerulean cedar
Does anyone know what the "assets/hubert/hubert_base.pt" file is? When I run the command "python gui_v1.py " through cmd it writes to me that this file was not found, and it is. It's not there, where can I find it, and what is it?
low shard
Does anyone know what the "assets/hubert/hubert_base.pt" file is? When I run the...
don't follow youtube tuts
you're using rvc realtime from the mainline/original project, which is worse than wokada, which is worse than the wokada deiteris fork
for realtime voice changer, tell your pc gpu in #🔍│help-w-okada
simple ore
I don’t wanna do the pre train, I want to train fresh
do you have 50+ hour good dataset?
with multiple speakers and some amazing variety of the content?
outer isle
Yes and no
brave swallow
does 48Khz has more noise than 32Khz ?
mild sleet
that's what the index is for
can you please tell me whats that?
mild sleet
that's what the index is for
Ok just learned thats not what its for
the index stores the accent of the model you trained on RVC.
hallow thistle
can you please tell me whats that?
I already explained to you.
grand solstice
May I use a zip file with .index/.npy/.pth(I downloaded somewhere... it's a rvc model zip) to generate modification? Also, on weights.com, if a model isn't exaggerated as I expected, should I use it multiple times to enhance the modification on voice? or is there a better way to make the modification within the website?
somehow this channel doesnt allow me to upload a image😂
patent pasture
does 48Khz has more noise than 32Khz ?
better to train 32k anyway
brave swallow
better to train 32k anyway
got it
low shard
May I use a zip file with .index/.npy/.pth(I downloaded somewhere... it's a rvc ...
the rvc model zip shouldn't contain .npy, elaborate what you mean with modification and share the model download link
knotty moth
the .npy file is an intermediate file produced during index training in mainline rvc, not the final result, so better remove it to reduce the file size
grand solstice
the .npy file is an intermediate file produced during index training in mainline...
thank you
low shard
https://aimodels.org/ai-models/rvc-models-ai-voice/xxxtentacion-ai-voice/
that's just a mirror of https://huggingface.co/QuickWick/Music-AI-Voices
i still dunno what u want to do and mean with modifications
wheat lion
what up lads, is anyone else running RVC on arch using rocm? I'm facing a few strange issues, this is my full log, from start up of the web UI to trying to process a vocals file:
simple ore
what up lads, is anyone else running RVC on arch using rocm? I'm facing a few st...
downgrade torch to 2.3.1-rocm
wheat lion
i'll give it a go
ty
yeah that's a problem, I think that version of torch is incompatible with the latest hip runtime (ImportError: libamdhip64.so: cannot enable executable stack as shared object requires: Invalid argument)
i'll try and see if I can get a docker container going
raven condor
In Applio where do i select refinegan as the architecture to train a model with? i've downloaded some pre-trained refinegan models, matched the sample rate to the dataset but when attempting to train i get: The parameters of the pretrain model such as the sample rate or architecture do not match the selected model.
simple ore
In Applio where do i select refinegan as the architecture to train a model with?...
you need to clone the main repository
grand solstice
that's just a mirror of https://huggingface.co/QuickWick/Music-AI-Voices
thanks. the modification, I meant voice conversion. changing my voice to somebody else
and i just figured out how to use weights.com/... today earilier it did not change my voice after I uploaded wav, thats why i kept wondering if there is another alternative for rvc
wheat lion
I think I discovered the issue, my GPU is supported by rocm but only windows, wtf ayymd
low shard
I think I discovered the issue, my GPU is supported by rocm but only windows, wt...
amd has pretty shitty and weird support for AI
wheat lion
I am aware, some things never change
low shard
https://docs.applio.org/applio/getting-started/installation#alternative-installation-methods maybe you could try Applio, which is a fork (modified version) which got more updates and easier user interface
or try mainline/original project guide https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/blob/main/docs/en/README.en.md
wheat lion
yeah i tried mainline, i'm trying applio on windows now
low shard
yeah i tried mainline, i'm trying applio on windows now
not sure if applio got also linux amd support, maybe @simple ore knows since he had an amd gpu
wheat lion
it does but I assume the same problems would be there, the issue seems to reside in the hip runtime just not supporting the 7800 XT properly on linux
low shard
I am aware, some things never change
I mean hopefully maybe in the next years things will change
-# Or you could use Nvidia GPUs
Nvidia is better than AMD, the only issue is their prices
low shard
and i just figured out how to use weights.com/... today earilier it did not chan...
you need to import the rvc model, then click create
or use one that is on weights.com
simple ore
I think I discovered the issue, my GPU is supported by rocm but only windows, wt...
what gpu?
7800xt has options - HIP SDK + Zluda + patched cuda torch on Windows, WSL2, ROCM on Linux
wheat lion
I've tried ROCM on linux, but I do get a random segfault when doing the actual conversion, I've found other people who make their own ML projects having random segfaults as well, so I reckon it's an issue with their stuff
simple ore
yeah, more or less AMD fully supports only their top of the line GPU
can use Zluda on windows, should be fine
I think for linux you gotta use ROCM5.7, then set environment variable HSA_OVERRIDE_GFX_VERSION=11.0.0
simple ore
not sure if applio got also linux amd support, maybe <@155030383648440320> knows...
had, not using it any more 🙂
brave swallow
Why mine's not working?
not working like what ?
modest vector
Yo,
I’m looking for a good way to create a realistic AI voice, but I don’t know what to use or how to set it up to sound natural. Any tips?
frank olive
hello link rvc?
low shard
Yo,
I’m looking for a good way to create a realistic AI voice, but I don’t know ...
What's ur PC GPU
low shard
hello link rvc?
What's ur PC GPU and what do u want to do
modest vector
What's ur PC GPU
Rtx 3070
modest vector
As you got a good PC, you can use RVC locally, you can choose between:
- [Applio...
How many epochs for a realistic voice in RVC?
low shard
How many epochs for a realistic voice in RVC?
There isn't a right amount
Look at our docs for more info https://docs.aihub.gg/
slender hearth
Please how can earn iq points in this channel
low shard
Please how can earn iq points in this channel
what
low shard
How can earn points of ai hub
ai hub isn't a point system ?
slender hearth
ai hub isn't a point system ?
Okay thanks
low shard
Okay thanks
you mean levels? you have to chat
slender hearth
you mean levels? you have to chat
Yeah
simple ore
mainline has not been fixed yet
oblique venture
What does the pitch change do? to change between male and female pitch?
Im using AICoverGen No WebUI on colab: https://colab.research.google.com/drive/1u1brjK8IZt647UsbZuGYfW29oFM2I4tk?usp=sharing
simple ore
depends on the implementation
but it either raises f0 values by some amount or also nudges them to match regular note frequencies
merry eagle
@acoustic scarab can you give me google colab web rvc links?
low shard
<@362876249032359939> can you give me google colab web rvc links?
the helper explictly says in his display name that he does local inference only
he can’t help much on cloud
first, what’s your pc gpu and what do you want to do? To check if you got a good enough pc
merry eagle
the helper explictly says in his display name that he does local inference only
i need only links to access
merry eagle
first, what’s your pc gpu and what do you want to do? To check if you got a good...
r5 5600 gtx 1660ti
low shard
r5 5600 gtx 1660ti
what do you want to do? Inference or training?
merry eagle
what do you want to do? Inference or training?
wanna change my voice into another voice
low shard
wanna change my voice into another voice
That’s inference, but realtime for calls? Or on pre-recorded audios
merry eagle
That’s inference, but realtime for calls? Or on pre-recorded audios
both btw i have the realtime one i need only pre recorded
low shard
both btw i have the realtime one i need only pre recorded
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- 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.com: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio UI Colab: max 4 hours daily, not granted, of GPU
Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero
easiest local: Applio
gave you both local and cloud ways
merry eagle
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), ...
tnx man
merry eagle
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), ...
how can i make models do you have any guide?
low shard
how can i make models do you have any guide?
Train (make) RVC Models on cloud:
- Prepare the Dataset
- Setup RVC:
Choose a cloud way to use RVC,
- Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
- Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
- Mainline (UI)
- Applio by Vidal (UI)
- Applio by Shirou (UI, no guide as of right now)
- Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
- Be sure to know about the tensorboard
Google Colab = Easier but risk of getting disconnected
Kaggle = Harder but way more gpu time
If you are looking for the easiest way and for free, is using https://weights.com which ofc uses RVC
RVC Inference (use models) on pre-recorded audio on Cloud
You can use either:
- Weights.com: Easiest Possible Ever Automatic
- Ilaria RVC Zero: Fastest free on cloud
- Applio (ui)
merry eagle
# Train (make) RVC Models on cloud:
1. [Prepare the Dataset](<https://docs.ai-hu...
that's look hard
merry eagle
# Train (make) RVC Models on cloud:
1. [Prepare the Dataset](<https://docs.ai-hu...
what if i made a model can i make money with that?
low shard
what if i made a model can i make money with that?
you can’t do paid commissions unless you become a model master
low shard
that's look hard
it’s AI
merry eagle
you can’t do paid commissions unless you become a model master
i have too much free time in these day's i'll must try
low shard
i have too much free time in these day's i'll must try
model masters means that you firstly become a model maker by having your first model approved by QC, then havng to make many high quality models
merry eagle
model masters means that you firstly become a model maker by having your first m...
what's QC? it's an organization or something?
hallow thistle
what's QC? it's an organization or something?
The quality checker. A group of people in AI HUB by Weights doing models quality checking before they being uploaded to #1175430844685484042 .
merry eagle
The quality checker. A group of people in AI HUB by Weights doing models quality...
ohh tnx
knotty moth
what's QC? it's an organization or something?
Quantum Corp.
low shard
Quantum Corp.
they are going to believe that
jaunty shale
I tried to use mainline kaggle, but it gives me the same issue like in colab.
been trying to figure this out for days now
nvm i got it
pseudo steppe
i need best rvc
low shard
i need best rvc
What's your PC GPU and what do you want to do
pseudo steppe
What's your PC GPU and what do you want to do
ryzen 5 4060 need for discord
ram 16gb
low shard
ryzen 5 4060 need for discord
So, realtime for calls?
pseudo steppe
So, realtime for calls?
yeah
low shard
yeah
Wrong channel then
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
knotty moth
ryzen 5 4060 need for discord
I have answered you before #🔍│help-w-okada message
low shard
Go to #🔍│help-w-okada
pseudo steppe
ohh i though it is real time voice changer
pseudo steppe
I have answered you before https://discord.com/channels/1159260121998827560/1159...
actually its opening in my browser
low shard
ohh i though it is real time voice changer
Nope RVC never meant that
low shard
actually its opening in my browser
And that's completely normal
The great majority of AI programs use web uis
Use that channel
pseudo steppe
The great majority of AI programs use web uis
so no application of okada now?
earlier there was one
low shard
so no application of okada now?
It runs on your PC, just it uses a Web User interface
Also original wokada was a web user interface, just it made its own window which got removed in the fork as it can cause issues
pseudo steppe
It runs on your PC, just it uses a Web User interface
Also original wokada was a...
yeah but still i would prefer app
can i get that?
low shard
yeah but still i would prefer app
It's still an app
The only difference is the User Interface
pseudo steppe
The only difference is the User Interface
alright
low shard
Only the old original wokada which is way worse uses the web user interface in it's own window
Also use #🔍│help-w-okada
Don't use this channel
knotty moth
yeah but still i would prefer app
it shouldn't affect performance, but has actually better performance than the original one
pseudo steppe
it shouldn't affect performance, but has actually better performance than the or...
okay
hallow thistle
yeah but still i would prefer app
To get a separated app version of Detris' W-Okada, download the pre-complied codebase of Detris' W-Okada from GitHub, then try code the GUI by yourself and let the author know.
pseudo steppe
To get a separated app version of Detris' W-Okada, download the pre-complied cod...
💀
hallow thistle
That's what I can say if you want the app version of it.
merry eagle
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), ...
can you give me real time voice changer cuz mine is not working
sudden cave
mine neither
low shard
can you give me real time voice changer cuz mine is not working
alright, let's go in #🔍│help-w-okada
low shard
mine neither
tell ur pc gpu in #🔍│help-w-okada
hallow thistle
To request someone to do voice model for you, go to #1159289738314919936. #✨│ai-help here isn't where you asking for that.
stable remnant
i cannot start server using ngrok it allways says "server stopped"
low shard
i cannot start server using ngrok it allways says "server stopped"
elaborate:
- your pc gpu
- what do you want to do
- what guide link are you using
hallow thistle
i cannot start server using ngrok it allways says "server stopped"
For W-Okada cloud, go to #🔍│help-w-okada. If you mean something else, tell me.
stable remnant
is there any guide how to run kaggle?
broken urchin
which tool should i use to make a high quality voice model?
low shard
which tool should i use to make a high quality voice model?
What's your PC GPU
broken urchin
What's your PC GPU
rtx 3050
low shard
is there any guide how to run kaggle?
You asked the same question on #🔍│help-w-okada , firstly explain your PC GPU and what you want to do
low shard
no no, desktop
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- 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.com: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio UI Colab: max 4 hours daily, not granted, of GPU
Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero
easiest local: Applio
Train (make) RVC Models on cloud:
- Prepare the Dataset
- Setup RVC:
Choose a cloud way to use RVC,
- Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
- Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
- Mainline (UI)
- Applio by Vidal (UI)
- Applio by Shirou (UI, no guide as of right now)
- Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
- Be sure to know about the tensorboard
Google Colab = Easier but risk of getting disconnected
Kaggle = Harder but way more gpu time
If you are looking for the easiest way and for free, is using https://weights.com/ which ofc uses RVC
RVC Inference (use models) on pre-recorded audio on Cloud
You can use either:
- Weights.com: Easiest Possible Ever Automatic
- Ilaria RVC Zero: Fastest free on cloud
- Applio (ui)
broken urchin
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), ...
what about mangiorvc?
i have that downloaded
low shard
what about mangiorvc?
That's extremely outdated
Don't follow YouTube tutorials
That program isn't maintained since 2023 and is one of the oldest RVCs you could use
broken urchin
so its bad
low shard
so its bad
Yes, very much
We even removed it from our docs, the creator doesn't maintain it nor fixes any bug, don't use it at all
@broken urchin just read what I told you, I gave you all the options
broken urchin
yeah i read it thanks
low shard
Yw
knotty moth
no no, desktop
the 8 GB variant is more recommended to do so
broken urchin
the 8 GB variant is more recommended to do so
welp i have the 6gb version
carmine hearth
I'm using machine translation. I apologize if the sentences are awkward.
Dear Ai Hub intelligentsia, do you have any guesses as to why people who were once using RVC mainline or Mangio fork feel that Applio sucks after using it? It's hard for me to understand what exactly is wrong with them, because most of the people who make this claim usually treat me as a brainless worshipper of applio, or are so inexperienced with Applio (they're new to it) that they blame the problem on Applio as a whole, rather than on some feature of Applio.
One thing I can be sure of is that they are having an experience that makes them feel that the output from Applio is clearly inferior to the RVC mainline. Does anyone know why? I feel bad for them that they are giving up the conveniences that Applio's developers have worked so hard to create and either going back to the mainline or giving up on using RVC altogether.
civic kelp
How do I train a voice with TITAN? I have RVC training software already installed but the models I have are until rmvpe
analog obsidian
I'm using machine translation. I apologize if the sentences are awkward.
Dear Ai...
applio inference is slightly worse than mainline, yeah
training is better tho
low shard
How do I train a voice with TITAN? I have RVC training software already installe...
rmvpe is not a model, it's a pitch algorithm extraction
what rvc did you download and what's your pc gpu?
verbal widget
-Colab
low shard
@tough fiber #🔍│help-w-okada message that's the latest precompiled mainline rvc, which is around over 1 year ago
@crude flame is there really no newer precompiled for rvc mainline?
atp shouldn't the docs explain how to do it via source?
@tough fiber if you want you could try applio, which is a more updated fork of mainline rvc
tough fiber
<@283628784396009473> if you want you could try applio, which is a more updated ...
oh thanks ill try applio
on local
crude flame
<@673327878288703519> is there really no newer precompiled for rvc mainline?
🤷 i am unaware if there are any newer ones
civic kelp
what rvc did you download and what's your pc gpu?
I downloaded the mangio-rvc one, my GPU is 8GB of VRAM (RTX 3060)
low shard
I downloaded the mangio-rvc one, my GPU is 8GB of VRAM (RTX 3060)
that fork is extremely old and not updated
It's better you get Applio
also, in the docs it will be explained how to use pretrains
fringe summit
yoo i need help
low shard
yoo i need help
!howtoask
patent trellis
low shard
yoo i need help
#🧬│ai-chat message is this what you're asking for?
civic kelp
As you got a good PC, you can use RVC locally, you can choose between:
- [Applio...
Thank you so much for the info, yeah I use mangio RVC locally so I have no problems.
But how do people make TITAN remakes of models? That's my question, and how do I make my own?
low shard
Thank you so much for the info, yeah I use mangio RVC locally so I have no probl...
mangio rvc is old
you should delete it and get something newer and supported like Applio
about 'remakes', you would just need to retrain your model using the titan pretrain https://docs.aihub.gg/rvc/resources/training/#pretrains
civic kelp
Noted. Thank you so much for the info again.
low shard
Noted. Thank you so much for the info again.
yw
valid spruce
Could someone help me make the models' breathing more natural without that robotic sound?
Sample
crude flame
Could someone help me make the models' breathing more natural without that robot...
add a ton of breathing into the dataset or cope
valid spruce
Should I do these one after the other?
crude flame
Should I do these one after the other?
doesnt really matter where the breaths are as long as they are there
valid spruce
Okay, I always remove it because they say it adds noise to the model.
knotty moth
Okay, I always remove it because they say it adds noise to the model.
if you don't mind robotic breaths as it's common in expressive talking and singing vocals
formal wind
How do I get the model master role?
crude flame
How do I get the model master role?
https://www.aihub.gg/en/dashboard/apply apply here (i think)
formal wind
I'll check it out thanks!
sonic prawn
Can I use RVC in python code? I want to automate something using python, I generate text using LLM then TTS using RVC
dull plume
why does the perf thing not appear on my voice changer client
low shard
why does the perf thing not appear on my voice changer client
wrong channel
@dull plume
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
show a screenshot of ur wokada in #🔍│help-w-okada
simple ore
Can I use RVC in python code? I want to automate something using python, I gener...
Yes, you can call inference from python code
lean hornet
https://www.instagram.com/reel/DG0cTHIyd78/?igsh=d2t3N3c1YjB4Zm9r
Is there a voice model of this voice from this reel? If there is mention me please 🙏
low shard
https://www.instagram.com/reel/DG0cTHIyd78/?igsh=d2t3N3c1YjB4Zm9r
Is there a voi...
You can search rvc ai voice models at:
- #1175430844685484042
- In #🔍│find-models , Do /find with @earnest musk
- https://weights.gg/ (login required)
- https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
- https://voice-models.com/
- https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)
if there isnt one, you can:
- #1159289738314919936
- #1191429836321849435
- make it yourself with our docs guides https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/
earnest muskBOT
You can search rvc ai voice models at:
- <#1175430844685484042>
- In <#11635920...
:wave: @low shard, How can I help?
Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image
twilit forge
i was afk while my model trained the and the runtime after it finished where can i find the pth file
low shard
i was afk while my model trained the and the runtime after it finished where can...
Did it disconnect automatically?
You might have lost it, check if there's any files on Google drive
twilit forge
You might have lost it, check if there's any files on Google drive
i lost it i will just train again its aight
night rune
Help?
low shard
i lost it i will just train again its aight
Be sure to check it once in a while so it doesn't disconnect
low shard
Help?
Elaborate:
- ur PC GPU
- what guide are u using
- what u did step by step
- what do u want to do
night rune
Elaborate:
- ur PC GPU
- what guide are u using
- what u did step by step
- what...
I'm using Applio (google colab) and trying to upload a wav file for training
To train a new voice model
unique rock
low shard
I'm using Applio (google colab) and trying to upload a wav file for training
Please elaborate also your PC GPU
night rune
same bro
low shard
I also just got a GitHub issue with the same issue for my facefusion online ports....
Welp, new cloud issue, I gotta check this
nocturne mural
Help?
I will check the error although it is most likely an internal gradio problem.
low shard
@night rune btw I asked ur PC GPU because if it's good enough you can do it locally without relaying on cloud
night rune
Please elaborate also your PC GPU
That's how I use google colab, it doesn't use my pc as such
But all right, it's a gtx 1650 super.
night rune
<@488644979456081920> btw I asked ur PC GPU because if it's good enough you can ...
nah, my pc is trash for this
low shard
That's how I use google colab, it doesn't use my pc as such
But all right, it's ...
Ohh, I mean you could technically do it locally on your GPU but it could be kinda slower and limited
knotty moth
try uploading the dataset to applio/assets/datasets through the (imjoy) file manager
night rune
Ohh, I mean you could technically do it locally on your GPU but it could be kind...
I'll wait forever if I try it xd
night rune
try uploading the dataset to applio/assets/datasets through the (imjoy) file man...
Do I need to create a new folder for this?
brittle wing
Can anyone tell me which one I download from git? I don't have a video card.
low shard
Can anyone tell me which one I download from git? I don't have a video card.
Download what? You asked the same thing in #🔍│help-w-okada
night rune
try uploading the dataset to applio/assets/datasets through the (imjoy) file man...
Okay, it worked
Why does my browser say the files are infected?
pastel oak
Why does my browser say the files are infected?
Probably doesnt understand where the download is coming from in a sense
So it gives a warning in case
low shard
heavy arrow
ik hina's not working rn, is there a webui thats currently working without many problems? my pc is probably not good enough to do anything on my computer, i use a gtx 1660 ti.
low shard
ik hina's not working rn, is there a webui thats currently working without many ...
Actually, your PC can run the Wokada deiteris fork
Are you going to use wokada for games? And if so, which?
heavy arrow
im creating AI content trying to make a model for transforming instrumentals into beatboxes
i already got the model created now though, i just need to run it on my computer, its trained on applio rvc
looking at the documentation its actually remarkable thinking that i could use this for real-time voice changing. the only thing ive seen like this is that really shitty voice.ai app
low shard
im creating AI content trying to make a model for transforming instrumentals int...
ohh you were talking about hina mod mainline rvc ?
You need to elaborate always what you are using, hina created many things
nvm you mean realtime, so you were talking about the hina mod original wokada
then yeah, you got also the wrong channel
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
low shard
i already got the model created now though, i just need to run it on my computer...
Let’s talk in #🔍│help-w-okada
heavy arrow
after searching up just 'hina' i realized that a little too late haha ^^;; i was talking about the rvc one, just mentioned realtime because i saw it come up in what i was reading 🙏
assuming i cant use wokada for ai covers or changing instrumentals to beatboxes (realtime), would any rvc fork still work with my GTX 1660 Ti? @low shard
analog obsidian
assuming i cant use wokada for ai covers or changing instrumentals to beatboxes ...
for local conversion aka no realtime use applio
brittle wing
-colab
low shard
after searching up just 'hina' i realized that a little too late haha ^^;; **i w...
oh
please elaborate next time
I just tried to guess the most probable one
heavy arrow
of course 🙏 my fault original gangster. i got you confused twice in a row because i wasnt explaining it right @_@
low shard
assuming i cant use wokada for ai covers or changing instrumentals to beatboxes ...
Your Nvidia GPU is good enough to do inference (use models) locally (on ur pc), not the best to train (make models) even if still possible
You can:
- Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
- 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.com: Partnered with AI Hub, lets u do them easily but u may be in a queue
- Applio UI Colab: max 4 hours daily, not granted, of GPU
Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero
easiest local: Applio
applio ui works still if you use the localtunnel option instead of gradio #📰│dev-updates
low shard
-colab
check #📰│dev-updates
low shard
of course 🙏 my fault original gangster. i got you confused twice in a row becau...
lol 😭
heavy arrow
applio ui works still if you use the localtunnel option instead of gradio <#1159...
alright bet 🙏 thank you again for all the help
brittle wing
check <#1159380240271953940>
Saw it
low shard
alright bet 🙏 thank you again for all the help
You’re welcome, and let me know
brittle wing
I just needed applio no UI to see the code of one cell
oblique venture
Is it ok if i put most of the dataset normal speech (20-30mins) instead of singing?
analog obsidian
Is that good or bad?
bad
heavy arrow
damnn applio with my gtx 1660 ti is actually faster than it was in a collab
heavy arrow
with applio i gotta separate all the audio first, right? its been a cpl years since ive used a normal fork that doesnt split the instrumental for you
analog obsidian
with applio i gotta separate all the audio first, right? its been a cpl years si...
you can upload a long file and applio will slice it for you
heavy arrow
oh fr?? thats sick
does it keep the instrumental file?
heavy arrow
yes. im using it to turn instrumentals into beatbox
it sounds cooler than you think trust me
i got a herbert the pervert model merged for beatboxing and it goes CRAZY
analog obsidian
yes im aware this thing can clone instruments but still lol
woeful wave
guys i wanna start making songs using ai what platform or what should i download to do this and also do i put raw vocals or mixed and instrumentals or not ???
broken crane
anyone know why my voice changer dosent work??
knotty moth
with applio i gotta separate all the audio first, right? its been a cpl years si...
don't try to infer unseparated mixtures
broken crane
anyone know why my voice changer dosent work??
im using voice changer client btw
low shard
anyone know why my voice changer dosent work??
wrong channel
RVC = Retrieval-based-Voice-Conversion, the best Speech To Speech AI Models (on v2), Inferences (use models) pre-recorded audio (ai covers) and train (make) models
Wokada = uses RVC for realtime inference
@broken crane elaborate:
- ur pc gpu
- what guide did u follow (I hope not those old ass youtube tuts)
- the issue
heavy arrow
don't try to infer unseparated mixtures
i got uvr5 and vip models, if you know, which ones would be best to download for splitting instrumentals and vocals? as well as removing echo and reverb
heavy arrow
i ended up just getting what i know works from a while ago, de-echo and de-reverb by foxjoy and kim vocal 2
its a LOT faster on my computer than on a colab, i thought itd be slow since i got a gtx 1660 ti ngl but it takes 12 seconds for a 3 minute long audio file
also, is there a local version of the AICoverGen? id like to get that one as well for when im not doing the beatbox thingymabobber
dusk rock
hello, it's just me or aicovergen got error on google colab?
heavy arrow
hello, it's just me or aicovergen got error on google colab?
yeah, seems aicovergen is prettychopped ^^;;
i couldnt even get it working locally
#📰│dev-updates / #✨│announcements currently a lot of colabs and stuff that uses gradio are broken rn tho
dusk rock
i couldnt even get it working locally
yeah, I tried locally too but same result as using colab 😦
heavy arrow
i tried another one and this one works locally and has a colab available incase you wanted to try it instead:
it has youtube links working again too ^^ woot woot
dusk rock
i tried another one and this one works locally and has a colab available incase ...
Thank you, I might try this one
heavy arrow
of course gl gl
glass igloo
Hi, can you tell me what files I need to upload to share a model I have trained with other users?
low shard
Hi, can you tell me what files I need to upload to share a model I have trained ...
the .pth and .index, also look at https://docs.aihub.gg/extra/model-maker-role/
low shard
hello, it's just me or aicovergen got error on google colab?
the great majority of google colabs are broken, read #📰│dev-updates
broken crane
can someone help me, when in call and my voice changer is on, whoever is talking can hear there self through there voice
@low shard ??
low shard
can someone help me, when in call and my voice changer is on, whoever is talking...
I told you before, this is the wrong channel #✨│ai-help message
Also, you need to elaborate: #🔍│help-w-okada message
You can't expect me to help without any type of info
frozen ledge
-colab
formal wind
Is it hard to get model master. Because i feel like if I don't get it, its because of my dataset cleaning process.
Oh also, how will I go about knowing how I did. Like will someone message me
crude flame
Is it hard to get model master. Because i feel like if I don't get it, its beca...
you only need 3 hq models
idk how it will let you know its a new thing
formal wind
Yeah last time I tried it was through a discord bot lol
kindred eagle
i wanna know what is the best pretrain model to train a voice on.
glacial pollen
i wanna know what is the best pretrain model to train a voice on.
This answer might not satisfy you but, there's no universal winner.
Each was trained a lil differently, different people behind it, different settings and different datasets
So one might fit your model ( or rather, dataset / voice / speaker, whatever ) better
Original pretrains, klm hifigan or those experimental refinegan ones is what I personally can recommend.
Best for you to just try and see
But as always, it's a good habit to start with original ones and only try customs if the results aren't good enough ( and after you made sure you've exhausted your opnions; aka, it's not user-error )
kindred eagle
like i've tried orignal, ov2super and rigel, sonce my data has like mixed languages, i tried it on different voices and like it worked but it had tearing or artificats in it.
i dunno what do i do to make my dataset to sound better
so other than those 3 pretrain are there any which has like low artifact rate and can produce a better voice model quality
and also do I use the rvc disconnected colab (as its says it's outdated) or the mainline colab?
knotty moth
like i've tried orignal, ov2super and rigel, sonce my data has like mixed langua...
the latter two are old and rigel is a failure overall (unless they could use better training configurations and train more epochs under some powerful H100s)
knotty moth
But as always, it's a good habit to start with original ones and only try custom...
@kindred eagle u r ignoring this suggestion
glacial pollen
🤔
kindred eagle
nope nope had tried the orignal as well
will try ov2 then compare
and then do the rest............
glacial pollen
Well, you can start away with ov2 and or klm if you want
but you see, people at times go for what seem " the best " or is the most recommended
without testing stuff, and AI is, well, it ain't deterministic in that way
However, I wouldn't bother with other models than those I wrote about
Most if not all customs at the time were trained on fp16 with exploded gradients ( simply put, aren't that stable
kindred eagle
hmmmmm, alright then i will actually compare the 3 models you have mentioned, compare them and will proceed with what i think is the best among them.
glacial pollen
Yup, the right approach
kindred eagle
ty for the suggestions @glacial pollen
glacial pollen
quality needs some decent work, that's just how it is
Yea np man
best of luck and take your time
kindred eagle
quality needs some decent work, that's just how it is
true
glacial pollen
Just to encourage you to not give up, my best model took me few months ~lmao
Ofc let's not go that drastic, just saying
cause some people train 1 or 2 models and give up, quite sad seeing it happen
kindred eagle
cause some people train 1 or 2 models and give up, quite sad seeing it happen
yeah i've trained a couple of models
but like some turned out crazy good but some had them annoyiing artifacts
glacial pollen
Yea like, I know it can be exhausting to be going through various batch_sizes and pretrains, but when it works, it's worth it
lots of things contribute into artifacts to be honest
kindred eagle
it is worth it tbh
glacial pollen
yup
kindred eagle
lots of things contribute into artifacts to be honest
yup i didnt know that you could actually do a lite version of phantom centre extractor in audacity, so the reverb that was there in the audio after the dereverb process actually F'ed up the model quality
glacial pollen
oh, well
for de-reverbering I can only really recommend vx's dereverb
tho, yea, it ain't free and is a vst ( ai powered however
kindred eagle
ahh i use the uvr dereverb and denoise it works 75% of the time but yeah that remaining reverb in the audio............aaaaaggghhhhh
glacial pollen
Yea, the models aren't the best at certain reverb types, esp those minimal room ones
If you are skilled enough, you can manually yeet them, or at least tame the trails / leftovers
goes like this
kindred eagle
yup
kindred eagle
and yeah is spectralayers like good or meh
glacial pollen
That's rx
knotty moth
yup i didnt know that you could actually do a lite version of phantom centre ext...
not a good idea to directly do center extract, you should first remove post process reverb (many are stereo, some are mono), and then remove the remaining one that's mostly mic room reverb using RX11 dialogue isolate
glacial pollen
Spectra layers is decent but, if I had to choose the winner, it's rx
kindred eagle
not a good idea to directly do center extract, you should first remove post proc...
yes i do know the process
glacial pollen
Dialogue isolate can damage the audio so, better to be careful
It's far from what I'd call reliable
kindred eagle
Spectra layers is decent but, if I had to choose the winner, it's rx
alright gotcha uninstalling spectralayers
glacial pollen
well not really, if you're handy in it, you can use that with no issues
but I just prefer rx
kindred eagle
well not really, if you're handy in it, you can use that with no issues
nahh like it crashes 7 times before i can actually work on it. so its better that i uninstall it
glacial pollen
specifically, I really like working on spectrograms in rx
change the scale, zoom in, feather if needed, work repeat
kindred eagle
specifically, I really like working on spectrograms in rx
oooooooooooooooo
glacial pollen
Aaaa
knotty moth
Dialogue isolate can damage the audio so, better to be careful
it is mostly some breaths
glacial pollen
I've never had any luck with it if it comes to anime type reverb
at best, it'd always castrate the audio
decrease the fullness or screw up the respiratory range
kindred eagle
anime reverb? anime has reverb in it? since when or am i dumb to not notice it?
kindred eagle
ahh i c the toji model i trained.......... hmm
knotty moth
I've never had any luck with it if it comes to anime type reverb
the stereo ones are quite easy
kindred eagle
thats why it was kinda F'ed up but usable
glacial pollen
The issue with stereo vs my workflow, is the fact stereo has 2 channels and they are never uniform
I extract stereo, operate on 1 channel, then de-reverb it in mono
100% predictable
kindred eagle
yup i started soing that recently
glacial pollen
Another thing is, vx lets you finetune the de-reverb to your needs
kindred eagle
better quality and prediction tbh
glacial pollen
- you tame the rest in rx
and the results are perfect
much more perfect than any automation / models can give you
glacial pollen
Why is that? because it doesn't get 100% of it, it expects the user to handle a bit of it
knotty moth
the mono dereverb one is quite tougher for me, esp when there are some breaths
glacial pollen
Yea the breaths can get damaged sometimes, but you can always just layer the tracks
and manually de-reverb the breath
it's just some feathered selection yeeting / enveloping
But ye, I get that. Everyone has their own workflow they recommend.
That's why I recommend mine, which is vx's de-reverb pro mono + manual polishing in rx
kindred eagle
jaunty iris
sorry im really late to all this but if someone could point me how to use refinegan pretrains in applio im just stuck here
low shard
sorry im really late to all this but if someone could point me how to use refine...
refinegan is experimental
there’s no stable version for it yet
the only way is via using the main branch source code
jaunty iris
the only way is via using the main branch source code
where do i download pls 🙏
hallow thistle
where do i download pls 🙏
I think it's on GitHub?
jaunty iris
hmm ok
low shard
where do i download pls 🙏
code > download zip > extract > run install
@jaunty iris what’s ur pc gpu tho
jaunty iris
oh ok so not the precompiled 3.2.8 got it
i just upgraded everything im on 4070 super 12gb 
low shard
oh ok so not the precompiled 3.2.8 got it
it’s the stable release, this is the main branch source code which could be not stable
its experimental
low shard
i just upgraded everything im on 4070 super 12gb <:sunglasses:120959209293060920...
Great
jaunty iris
understood
but yea been wanting to try training again and now i can do local 😁
interested in how this goes i was gonna test w KLM5
jaunty iris
but yea been wanting to try training again and now i can do local 😁
still not working sadly 💔 @low shard i went back to make sure i had my dataset file at the same sampling rate as the pretrains but now i genuinely dont know what to do 😭
low shard
still not working sadly 💔 <@911742715019001897> i went back to make sure i had ...
you seem to be still using the latest stable release and not the source code one
low shard
code > download zip > extract > run install
@jaunty iris did you miss this step?
kindred eagle
I’m currently using RVC for voice cloning, but I’m curious if there are any better apps out there that might do a better job. Sometimes, I feel like the slider values in RVC don’t work as well compared to the online RVC forks.
Please don’t judge me, but my GPU is a 1650 mobile. Any suggestions or experiences you can share would be greatly appreciated!
simple ore
you’re welcome
we gonna make 3.2.9 release with refinegan disabled, just to push other changes
right now even the main branch has it disabled in the repo
low shard
right now even the main branch has it disabled in the repo
huh why’s that?
low shard
I’m currently using RVC for voice cloning, but I’m curious if there are any bett...
which rvc are u using
low shard
I used to use Kits.ai until they transitioned to a fully paid service. The infer...
kits.ai uses RVC too, the major diffference is its easier to use and automates processes like separating vocals and instrumentals that u can do urself
knotty moth
I used to use Kits.ai until they transitioned to a fully paid service. The infer...
they have turned greedy and shady, even since before that they have been gatekeeping models trained using the service from downloading locally
low shard
^
simple ore
huh why’s that?
just to make a build
distant turtle
-colab