-gui
#✨│ai-help
knotty moth
median island
I was suggested to use this, but I don't know what else I should be using
knotty moth
What should I be using?
-rvc
AI HUB Docs
median island
Thx
median island
I'm now on the new one, but I'm getting this error and I have no idea what this could mean
AttributeError: 'NoneType' object has no attribute 'dtype'
median island
If anyone could help me with resolving the error I'd be grateful
simple ore
full error message
median island
if data.dtype in [np.float64, np.float32, np.float16]:
AttributeError: 'NoneType' object has no attribute 'dtype'
Is that good?
simple ore
i mean the full error stack
simple ore
the entire error message that shows the entire chain of the failing modules
hallow thistle
Are you sure your RAM is large enough to handle that?
knotty moth
depending on your ram capacity, try processing a shorter file, or split it first
primal tree
Hey guys! I'm looking to find/create an rvc voice that is free to commercially use. Idc if I have to collect massive sets of text-to-speech voice lines in order to create the rvc, I just want to know what options are out there.
knotty moth
Hey guys! I'm looking to find/create an rvc voice that is free to commercially u...
the best bet is to use your own voice or your friend's and have him/her consent
simple ore
Are you sure your RAM is large enough to handle that?
it was a regular RAM allocation, running on CPU
primal tree
the best bet is to use your own voice or your friend's and have him/her consent
So I do already have my own voice, but is there any other options? Like I know you can use the voicevox characters as an example.
analog obsidian
So I do already have my own voice, but is there any other options? Like I know y...
voices have copyright, and people can sue you if you use them without their permission
knotty moth
So I do already have my own voice, but is there any other options? Like I know y...
u should try asking the VAs for permission
analog obsidian
it's safer to pay some random fiverr vas
primal tree
it's safer to pay some random fiverr vas
is there an easier way to search for vas that would be down for that? or do you just gotta reach out to each one and hope they say yes pretty much?
analog obsidian
is there an easier way to search for vas that would be down for that? or do you ...
basically the second one, most of vas hate ai tho
good luck
🦈 🤙
primal tree
thank you.
knotty moth
is there an easier way to search for vas that would be down for that? or do you ...
if you don't tell them for AI use even though you pay them, you could risk getting sued
primal tree
I understand
Like I said, I just know there are options out there like voicevox or elevenlabs voice to voice stuff
crude flame
voices have copyright, and people can sue you if you use them without their perm...
you can just merge a bunch of voices and bam og voice
primal tree
was wondering if anyone else had other options
analog obsidian
you can just merge a bunch of voices and bam og voice
true lol
analog obsidian
was wondering if anyone else had other options
u can merge rvc models to create a original voice
buuut i know that merging too much can cause unstability
knotty moth
you can just merge a bunch of voices and bam og voice
with some notes that it could only be done on the same sample rate and could be hit or miss in terms of quality
analog obsidian
yeah
crude flame
with some notes that it could only be done on the same sample rate and could be ...
yea 😔
analog obsidian
hallow thistle
my error is VB cable not working for me
VB-Cable cannot be used with regular RVC program. Unless you mean by W-Okada, which you're supposed to ask in #🔍│help-w-okada.
knotty moth
VB-Cable cannot be used with regular RVC program. Unless you mean by W-Okada, wh...
already answered there
hallow thistle
That's why I told him to go there.
hollow thunder
Is the Hina mod colab not working anymore?
hallow thistle
Is the Hina mod colab not working anymore?
Yes, it is bugged from now on. There's another better one available, that is Detris' W-Okada, but using this one on free Colab can get you terminated from using the service.
For W-Okada, go to #🔍│help-w-okada.
hollow thunder
I'm trying to make covers, not ai voices
hallow thistle
Has Hina ever made the "RVC" Colab? Never heard of this information. I've only heard of Hina making W-Okada Colab notebook.
Still. If you mean by RVC that can do AI cover, there's another better one available.
hollow thunder
Which is what RVC
hallow thistle
What does that mean? You only have a phone and no PC?
hallow thistle
Laptop is still a PC.
hollow thunder
Oh, thought they were considered different
knotty moth
I have a laptop
there's no reason not to use the laptop over the phone
hallow thistle
Using a Colab notebook on mobile phone is much harder than using it on desktop/laptop PC, you know.
hollow thunder
I know
But most of the time I'm not at home, and when I am I pass out from exhaustion
So I multi task and do what I can on my phone, so I was using hina mod
Since it was easier, but it's not working now
low shard
Is the Hina mod colab not working anymore?
There's various different Hina mod colabs, Hina is just the name of the creator, mod means modified
Which are you talking about? Link?
And what's your PC GPU?
And what do you want to do?
hallow thistle
He said he has a laptop, aside from using phone, but never tell the name of GPU of it.
hollow thunder
Which are you talking about? Link?
It's Hina_Mod_AICoverGen_colab.ipynb, I was using it to make AI covers. As for the GPU, idk the name of it, I'd have to look
knotty moth
It's Hina_Mod_AICoverGen_colab.ipynb, I was using it to make AI covers. As for t...
check at task manager
or the laptop package with spec details
low shard
It's Hina_Mod_AICoverGen_colab.ipynb, I was using it to make AI covers. As for t...
that’s never coming back
@hollow thunder #📰│dev-updates message aicovergen got replaced forever, and it’s abandoned
always check that channel
low shard
It's Hina_Mod_AICoverGen_colab.ipynb, I was using it to make AI covers. As for t...
Also You can check your pc gpu via:
ctrl+shift+esc (task manager) -> Performance tab -> GPU
Cloud )remote good pc) services like google colab should be the 2nd option only if you got a bad pc, not the first
broken tulip
Now that I messed around with the voices a bit, how do I train my own voice?
hollow thunder
<@385170368530219009> https://discord.com/channels/1159260121998827560/115938024...
Ok
low shard
Now that I messed around with the voices a bit, how do I train my own voice?
what’s ur pc gpu
broken tulip
what’s ur pc gpu
GeForce RTX 3060 Ti
low shard
GeForce RTX 3060 Ti
Good
knotty moth
GeForce RTX 3060 Ti
it's recommended to train locally, you should check out this guide: https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/
broken tulip
it's recommended to train locally, you should check out this guide: https://docs...
Thank you.
broken tulip
As you got a good PC, you can use RVC locally, you can choose between:
- [Applio...
Thank you.
low shard
Thank you.
lmk for any issues
hallow thistle
Yo can anyone send me the link to download RVC?!?
Which RVC you're looking for? And what is your PC GPU?
!howtoask
patent trellis
solar token
Which RVC you're looking for? And what is your PC GPU?
Thats the thing.. Im lost and dont know what to do.
Is this too weak to handle?
hallow thistle
Thats the thing.. Im lost and dont know what to do.
Is this too weak to handle?
There's only just GPU 0? That's real weak.
solar token
There's only just GPU 0? That's real weak.
Does that mean its not compatible for my computer?
solar token
welp its over
knotty moth
Thats the thing.. Im lost and dont know what to do.
Is this too weak to handle?
it would run in cpu mode
solar token
it would run in cpu mode
oh
solar token
it's just much slower but still doable
I see. Thank you!
Didnt know just having to change your voice could be so demanding 😔
hallow thistle
This is Applio the RVC. https://docs.applio.org/applio
solar token
This is Applio the RVC. https://docs.applio.org/applio
So I could use this?
hallow thistle
It is best RVC program out there.
solar token
It is best RVC program out there.
"best" welp
it'll probably lag
but ill check it out
knotty moth
"best" welp
you can't compare anything else not based on rvc
broken tulip
lmk for any issues
How do I know if I did it right? XD
low shard
How do I know if I did it right? XD
Are you having any issues? If you did it all right the program should open normally and the model would sound good
primal tree
Hey guys! I made a post about this on their Reddit, but I never got a response. Wondering if anyone here would have an answer.
I was wondering if it is within elevenlabs terms of service to record a large set of audio from a voice I made on their site, to then turn it into a RVC model that I can use in real time with a program like w-okada. Other than having to stay subscribed, would this go against their TOS at all?
(also sorry if this is confusing at all, I'm still very new to AI voices and am unsure if I'm using the correct terms here lol)
simple ore
Hey guys! I made a post about this on their Reddit, but I never got a response. ...
using synthetic audio as a source for a voice model training is not ideal
if you're the one who only crosses the road at a designated pedestrian crossing, then violating ToS is very bad
same as using ablock on youtube
/s
primal tree
Haha I appreciate that.
Didn’t think about using a synthetic voice and how that might effect tho
analog obsidian
Hey guys! I made a post about this on their Reddit, but I never got a response. ...
you can use elevenlabs to earn money but only if you pay for their suscription which has a commercial license
iirc they have a non audible watermark in the results and thats how they can know if you're using the free plan
low shard
you can use elevenlabs to earn money but only if you pay for their suscription w...
That's interesting
analog obsidian
but ye, converting it to a rvc model might be against their tos, and even if isnt, the rvc results will be worse actually
jaunty shale
I love how I do my recording with my new mic with 48k sample rate and voice ends up being pixely as always.
low shard
how i can upload a voice
#🔍│help-w-okada message for realtime voice changer, that's wokada not 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
low shard
Intel(R) Iris(R) Xe Graphics
oh damn, integrated graphics, pretty bad
you got any other gpu?
hollow thunder
I told you I have a laptop, not a desktop
hollow thunder
you got any other gpu?
No
low shard
No
well in this case it's better to use cloud (remote good pc), #📰│dev-updates message here it's explained what's the replacement for AICoverGen
you can use RVC AI Cover Maker UI instead
small timber
What is the best progeram to clone voice to then use in Applio?
low shard
What is the best progeram to clone voice to then use in Applio?
Applio can do that
tiny idol
this one good ?
low shard
Don't ask in other channels, Im talking to you in #🔍│help-w-okada
tiny idol
yh i forgot
blazing solar
-colab
shadow sparrow
anyone else getting the problem with applio google colab right now? everytime i try to click on the public url, it just takes me to a website error
blazing solar
anyone else getting the problem with applio google colab right now? everytime i ...
Same
I was just enjoying it an hour ago
But then erorr
shadow sparrow
huh, thats weird
blazing solar
-audio
white bough
The problem with Colab seems to be coming from pydantic. Before clicking on "Start Applio", run this:
!pip install pydantic==2.10.6
And it should work
rain mulch
most models I put sound really robotic... I tried so many models, even adjusted my settings over and over again
but nothing
at this point... I think its just my voice
I use the w-okada fork voice changer
knotty moth
most models I put sound really robotic... I tried so many models, even adjusted ...
try another model, and you should go to #🔍│help-w-okada for the voice changer discussion
viral mason
need help
got this error in applio kaggle
prob gonna make a new acc on kaggle to test
viral mason
idk how to do all that
viral mason
can the kaggle space be updated so it doesn't do that, literally didn't do that yesterday
als remove the old bugfix code
abstract gulch
are hugging face links still currently used by users or is there a newer way to?
knotty moth
are hugging face links still currently used by users or is there a newer way to?
I don't quite get what u mean, but it should be publicly available unless set to private
abstract gulch
I don't quite get what u mean, but it should be publicly available unless set to...
nono, it says a voice model should have a hugging link with it right? but i didn't find any voice model with a link of it
hasty idol
im available to commission for a voice model or smth, need it for my vtuber project
where do i look for artists?
simple ore
can the kaggle space be updated so it doesn't do that, literally didn't do that ...
yes, one of the libraries that gradio uses got updated and it broke the UI
welcome to python development
viral mason
yes, one of the libraries that gradio uses got updated and it broke the UI
Sure, is it possible to fix the error or should I just wait?
dms ok?
simple ore
as I said, you make a new code cell using +Code button
paste the pip install command, then run the cell
viral mason
as I said, you make a new code cell using +Code button
see I'm slow and kinda need a visual example
viral mason
which cell do I put that before
simple ore
the one that +Code creates
viral mason
which one
unique rock
applio is working now?
knotty moth
applio is working now?
read the pydantic workaround above
unique rock
Should I create a cepda and put this?
simple ore
create what
blazing solar
The problem with Colab seems to be coming from pydantic. Before clicking on "Sta...
Can you show me where is that?
unique rock
create what
Cell*
blazing solar
I alr follow the instruction but nothing seem to work
It says TypeError: argument of type 'bool' is not iterable
simple ore
It says TypeError: argument of type 'bool' is not iterable
what's the output of the cell?
blazing solar
what's the output of the cell?
Im sory it worked when click the public link again
prisma kettle
-realtime
prisma kettle
Do I need to set cable as default for realtime? or no?
nothing is coming thru cable atm
prisma kettle
use different virtual cable
I just realized that after doing some searching 💀 me and VB Audio Cable have a long history
Glad to be rid of it so soon again. What cable should I be using?
Nvm I see which one to use
They need to update the guide. I was wary of using VB because it’s been pretty shitty in the past
hallow thistle
Do I need to set cable as default for realtime? or no?
For W-Okada, go to #🔍│help-w-okada or stay here for a regular RVC program.
prisma kettle
I'm not using Okada
Do I need to restart PC after installing muzychenko virtual cable
I'm not getting any voice through it rn
hallow thistle
Do you wanna use the virtual cable with RVC or something?
prisma kettle
Yea, realtime rvc
I have the gui
I'm just getting no sound thru it
wait
im dumb
hallow thistle
Realtime mode of old RVC is so far outdated than W-Okada.
hallow thistle
W-Okada is literally the realtime voice changer.
hallow thistle
prisma kettle
the guide didnt say that!!!
analog obsidian
rvc realtime gui is pretty outdated
hallow thistle
You were too confident in yourself on this one.
prisma kettle
Cut me some slack...
prisma kettle
this is all I got.
hallow thistle
There's a better one available. That's why I'm telling you to go to #🔍│help-w-okada for the realtime voice changer.
signal mauve
how do i test out my model using sample?
verbal oasis
hi guys any alternative to applio to generate text to speech and lbe able to download and load custom models from huggingface?
is applio not working also for you on colab at the moment?
jaunty shale
what settings are recommended for FL Studio Export for dataset (I used it to fix harsh frequencies).
jaunty shale
dunno which sample rate to use
low shard
hi guys any alternative to applio to generate text to speech and lbe able to dow...
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
low shard
is applio not working also for you on colab at the moment?
Elaborate the issue
flint solar
Elaborate the issue
pydantic library updated it’s not working for everyone
low shard
pydantic library updated it’s not working for everyone
Perfect
@simple ore another day, another colab broken (applio)
Google colab breaks soooo much
flint solar
the fix I gave yesterday should work
!pip install pydantic==2.10.6 ?
simple ore
yes
err... I guess it needs to be !uv pip install pydantic==2.10.6
since that was changed to uv
flint solar
U should ping dev updates do that everyone sees it
unique rock
err... I guess it needs to be ``!uv pip install pydantic==2.10.6``
Friend, I still get this error
verbal oasis
since that was changed to uv
hey is this the fix for applio?
simple ore
for UI colab, yes
verbal oasis
for UI colab, yes
so need to run that after installation and before starting?
simple ore
yes
low shard
U should ping dev updates do that everyone sees it
It got already fixed anyways, and there's already a text about broken colabs
orchid flint
This isn’t a question about RVC so much as it is about the weights bot. All of the models that I have uploaded to the server have never been credited to me on weights. Am I doing something wrong? I have no idea why or what to do.
unique rock
show the full output
brave garnet
I need to help, I have been using cover AI for 1 month and now when I use it again, it is a mistake
pastel oak
I need to help, I have been using cover AI for 1 month and now when I use it aga...
Screenshot of your interface + error log on cmd
hallow thistle
!howtoask
hallow thistle
That still doesn't give enough context of the problem you encountered.
brave garnet
Sorry but my laptop can't send pictures, does it matter if I send it to you separately
hallow thistle
No need to hop into my direct message to send that. A mod here will give you the image permission.
brave garnet
Can I write?
hallow thistle
Can you tell me the name of RVC program name you're trying to use?
brave garnet
the rvc i use is Hina_Mod_AICoverGen_colab.ipynb
The error reads as follows: "Traceback (most recent call last):
File "/content/Hina_RVC/src/webui.py", line 8, in <module>
import gradio as gr
ModuleNotFoundError: No module named 'gradio'"
hallow thistle
This one is guite too abandoned at this point. What is your PC GPU?
brave garnet
GPU is my pc or gg colab's?
hallow thistle
GPU = Graphics processing unit. It is a graphic card in your PC.
hallow thistle
To check your GPU name, open Task Manager, go to Performance tab, and spot where GPU 0 or GPU 1 is in the left side.
brave garnet
My GPU is 0
hallow thistle
My GPU is 0
So what is the name of it?
low shard
the rvc i use is Hina_Mod_AICoverGen_colab.ipynb
That one is outdated and replaced, check #📰│dev-updates to know what replaced it
hallow thistle
It's called Intell HD Graphic 5500
Ok. An integrated GPU isn't so good for AI. So let's go to online option.
low shard
It's called Intell HD Graphic 5500
It's bad integrated graphics so yeah you should use cloud
Check what I told u
brave garnet
That one is outdated and replaced, check <#1159380240271953940> to know what rep...
ok thank
yes, I know my GPU is not suitable so I use online data for example gg colab
low shard
yes, I know my GPU is not suitable so I use online data for example gg colab
Yeah lmk
brave garnet
brave garnet
Yeah lmk
Can I ask one more question?
I really like AI training but the AI training website I often use is gone. Is there any solution for the server?
low shard
I really like AI training but the AI training website I often use is gone. Is ...
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)
- 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 Colab: RVC Fork with some extra features like TTS
- RVC AI Cover Maker UI: Automatically Separates the Vocals and Instrumentals, converts the voice and mixes them back
brave garnet
wow
I struggled all day and when I entered this server everything was easy, thank Nick088 by Weights and the server very much
mystic tangle
how many epoch should i do if i have 55 minutes of talking no silence only words and no background noise
low shard
I struggled all day and when I entered this server everything was easy, thank Ni...
You're welcome and lmk
low shard
how many epoch should i do if i have 55 minutes of talking no silence only words...
No one can know
Use the tensorboard
mystic tangle
wheres that
mystic tangle
local mainline
the one i found in guide channel
first one
it doesnt mention how many epochs i should do for my dataset
low shard
the one i found in guide channel
Share a link and tell your PC GPU
mystic tangle
yea im trying to train my own
but im wondering how much epochs i should use for my dataset
i dont wanna overtrain it or anything
low shard
a link to the guide im using?
Yup
hallow thistle
or just more simple
More like easier to use.
amber skiff
Starting preprocess with 16 processes...
100%|████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:38<00:00, 38.59s/it]
Preprocess completed in 38.60 seconds on 00:23:34 seconds of audio.
Starting pitch extraction with 16 cores on cuda:0 using rmvpe...
0%| | 0/1 [00:00<?, ?it/s]An error occurred extracting file C:\Applio\logs\tep piseth AI\sliced_audios_16k\0_0_0.wav on cuda:0: cuDNN error: CUDNN_STATUS_NOT_SUPPORTED. This error may appear if you passed in a non-contiguous input.
100%|████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:24<00:00, 24.75s/it]
Pitch extraction completed in 31.58 seconds.
Starting embedding extraction with 16 cores on cuda:0...
100%|████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:03<00:00, 3.52s/it]
Embedding extraction completed in 10.07 seconds.
Starting training...
Loaded pretrained (G) 'rvc/models/pretraineds/pretrained_v2/f0G40k.pth'
Loaded pretrained (D) 'rvc/models/pretraineds/pretrained_v2/f0D40k.pth'
Process Process-1:
Traceback (most recent call last):
File "C:\Applio\env\lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Applio\env\lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "C:\Applio\rvc\train\train.py", line 497, in run
reference,
UnboundLocalError: local variable 'reference' referenced before assignment
An error occurred extracting the index: need at least one array to concatenate
If you are running this code in a virtual environment, make sure you have enough GPU available to generate the Index file.
@low shard
low shard
or just more simple
More simple and optimized
low shard
Starting preprocess with 16 processes...
100%|██████████████████████████████████...
Elaborate
What are you using
What's your PC gpu
What do you want to do
mystic tangle
okay thanks 👍
amber skiff
What's your PC gpu
3070
amber skiff
What do you want to do
train my own model
hallow thistle
Starting preprocess with 16 processes...
100%|██████████████████████████████████...
Are you using the latest version of Applio?
hallow thistle
Have you checked if your GPU driver is up to date?
low shard
Applio 3.2.8
Update windows and your drivers
mystic tangle
hi how do i make my model show up in tensor board
do i have click train model then it will show up?
low shard
do i have click train model then it will show up?
Tensorboard shows when u start training
hallow thistle
mystic tangle
no like
how do i get a pth file
do i have to start training for tensor board to work
ik where tensorboard is but theres nothing here when i train will something appear?
i already have a dataset and i processed it do i just train now
then it will appear in tensorboard?
knotty moth
ik where tensorboard is but theres nothing here when i train will something appe...
wait till it saves some model checkpoints during training, then click refresh
mystic tangle
oh okay
thanks
doe sit matter what i have the epoch at
i left it at 500
how many epoch should i let it run for
knotty moth
they should also appear in the logs folder
tulip cloak
how many epoch should i let it run for
Mhm
mystic tangle
oh my bad i was just finishing up on this
mystic tangle
they should also appear in the logs folder
what should i put for saves
hallow thistle
This is the TensorBoard.
fast scarab
hi i got an error with codename rvc and i have no idea what to do 😭
it says:
TypeError: argument of type 'bool' is not iterable
and also
unable to find files for the specified model(s)
simple ore
hi i got an error with codename rvc and i have no idea what to do 😭
it says:
Ty...
pip install Pydantic==2.10.6
rain kelp
So, I know this isn't exactly an RVC issue, but I've recently decided to try out "f5tts" model in the AllTalk web-ui, and uh... the actual voice just generates gibberish. Like, it sound as if it's speaking in reverse at an increased speed or something. Any idea what's going on there?
simple ore
f5tts does not support long texts
rain kelp
Well, just a short sentence produces the same result.
frail umbra
mmm my rvc webui doesnt even attempt to make an index
it just spits out a couple of these
unique rock
install normal requirements, then run the pydantic cell
It worked for me now, I had actually entered the pydantic code wrong, thank you very much for everything
frail umbra
im trying applio now, looks betta
cinder wyvern
does the applio gradio google colab not work anymore?
every time i try to run it i get an error
cinder wyvern
pip install Pydantic==2.10.6
oh wait nvm this works ty
cinder wyvern
aid
click 'show code' under the install applio section, and then add !uv pip install Pydantic==2.10.6 after all the other pip install lines, then run the install again
then when you run the start applio section it should give you the gradio links without errors
mystic tangle
i trained till 450 epoch should i keep going
knotty moth
i trained till 450 epoch should i keep going
look at avg loss graphs and you should test several model checkpoints
knotty moth
click 'show code' under the install applio section, and then add !uv pip install...
@tropic garden do this workaround
tropic garden
thanks i'll try again
weak cipher
Can the STT api do real time?
amber skiff
When creating a dataset, should I upload the audio as one file or separate it into multiple files of 10 seconds each?
@knotty moth
verbal oasis
hi guys, is there a way to make a generated audio more human?
I'm using applio, I generated my audio file and I'd like to make it sound more human after I generated it.
Is it possible?
low shard
hi guys, is there a way to make a generated audio more human?
I'm using applio, ...
Get a better model and clean the audio as much as possible
knotty moth
hi guys, is there a way to make a generated audio more human?
I'm using applio, ...
there are many things you'd have to learn, do experiment, and improve on: quality, consistency, variations, dynamic range, etc. as well as training stability and convergence
and note that the source quality and consistency may vary, and cleaning/restoration methods have their limitations
simple ore
When creating a dataset, should I upload the audio as one file or separate it in...
one file, use slicer in the preprocess.. or use simple slicer as explained in the doc
vapid mantle
Tam anlamadım 32K mı? 40K mı? 42K mı? hangisi?
simple ore
32k
vapid mantle
thx
bleak nymph
how good is laughing in a dataset?
would that ruin the model
i recorded my voice in audacity while playing cards against humanity and was laughing so much
it's very expressive this one unlike my other dataset
analog obsidian
how good is laughing in a dataset?
it's ok but doesn't really help, rvc will still output unrealistic laughing results
bleak nymph
yeah ik that i'm just wondering if it'll ruin the dataset and if i should remove it
like if it'll make the model worse
analog obsidian
as long it's not excessive, should be fine
but idk what could happen if the dataset has excessive laughing
bleak nymph
how much is excessive
analog obsidian
how much is excessive
20 mins of laughing in a 30 minute set
bleak nymph
eh i'll just experiment
bleak nymph
20 mins of laughing in a 30 minute set
well i would be laughing and reading out questions and answers so i don't think it'd be that bad
analog obsidian
what matters in the set is using the whole vocal range, i confirmed that yesterday when i trained a model of my voice
10 mins and was using my whole voice range without problems
also helps speech models have decent singing
analog obsidian
yuh, go for 30 minutes of expressive audio and you'll be fine
10 mins still felt a bit unnatural
looks like the sweestpot for a natural model is rlly 30 mins
rvc does not stand for realtime voice changer, it means retrieval-based-conversion
use help wokada for realtime troubleshooting/questions,etc -> #🔍│help-w-okada
rvc and w-okada are two separate things
rvc is an ai voice cloning software
w-okada allows the usage of rvc models in realtime
one is for training and inference of audio recordings
the other is for realtime inference
bleak nymph
WOAHHH
i started training
and it shows previous model graph too
that's so cool
bro why is the new model (blue) taking so much longer 💀
analog obsidian
bro why is the new model (blue) taking so much longer 💀
your dataset is bigger?
the new one
bleak nymph
your dataset is bigger?
i forgot by how much
but shouldn't be THAT much
22 mins
total
analog obsidian
check model_info.json of the old model
bleak nymph
it was originally over 3 hours but i cut it shorter
analog obsidian
bleak nymph
total_dataset_duration": "00:29:53"
30 mins
the newer one is shorter
hol up
old dataset
new dataset
analog obsidian
both are batch size 8?
bleak nymph
both are batch size 8?
yup
analog obsidian
idk then lol
bleak nymph
😭
that's so weird
anyways i remember i put me singing thru my ai model cause i was curious
it made me have an american accent
like a country accent
i do not have an american accent why'd it give me one
analog obsidian
i do not have an american accent why'd it give me one
contentvec main language is english
original pretrain trained only in english
this is the voice of the original pretrain, if you're curious
your model is training using this as a base
the point of training using a pretrain is to change this voice ^ to the one in your dataset
analog obsidian
wait so how do i train a pretrain or whatever
find 50+ hours worth of audio
analog obsidian
then you have to do a pretty weird folder structure since every speaker require it's own folder
bleak nymph
right
analog obsidian
and train for like 1 month idk
bleak nymph
contentvec main language is english
original pretrain trained only in english
there should be different ones with different accents
make a british one 🗣️
analog obsidian
well contentvec was trained using multiple languages
but it had more english data
bleak nymph
oh
so the english data contains british accents too? or nah
i'm assuming it had a lot more american accents
analog obsidian
i have no idea lol
edgy tangle
Hii
Im using kaggle on applio
this is a model with a dataset of 7 minutes and 4 batch size
but it gives me just 12 steps
On local training it gives me about 43 steps
but my pc explodes 
same dataset, same slicing, same pitch extraction
simple ore
kaggle is 2 gpus
so batch 4 means 2x4
7 minutes = 7 x 20 x 3s segments... so it should be ~105 steps with batch 4
so perhaps you did not slice audio properly and a bunch of segments are too big or too small
analog obsidian
I used simple slicing
batch size 4 in kaggle it's batch size 8 in local
because every gpu runs the same batch size
so if you put batch size 4 in kaggle, the two gpus will run at batch size 4, which makes it bs 8
try batch size 2
in kaggle
thats it's the equivalent of batch size 4 in local
simple ore
it is a simple check - if the train loader has less than 3 samples it means you f'd something up
show the extract features log
and preprocess too for a good measure @edgy tangle
karmic oliveBOT
simple ore
wait me a sec
I said show the logs
edgy tangle
uh
sorry
what do you mean by logs
console or files?
im not native english speaker
anyways, ill send both
simple ore
that seems like 5s slices were used?
simple ore
the training script should handle that, but it is weird
simple ore
start anew
simple ore
maybe I messed it up without noticing it
delete the folder with the prepared files
I've tested myself
5s slices
edgy tangle
Ok
simple ore
but you should not use 5s, it is too much
edgy tangle
I didn't
Just forgot to set it at 3s
hehe
I don't know what I did
but I deleted the notebook and created a new one
And now is working
hehe
Anyways, thanks for the help!
now with 2 batch size gives me 22 steps
I think it is better(?)
bleak nymph
@analog obsidian hi
analog obsidian
now with 2 batch size gives me 22 steps
looks like the slicing is still messed up
you're still using 5s slicing
bleak nymph
<@775545133448953856> hi
is this good
edgy tangle
you're still using 5s slicing
really?
bleak nymph
or overtrained or need more training
analog obsidian
is this good
can you show me the same thing but with ignore scalars on and 0.6 smoothing
because from what i looking at, looks like it got stuck
but could be because smoothing 0 + the second box enabled
something like this
bleak nymph
analog obsidian
very flat line pattern
should be like mine
can you
disable the second box
and click the third one
maybe it's just that
if still looks like a flat line
then uhh the model got stuck
bleak nymph
and click the third one
third box?
analog obsidian
analog obsidian
bro
bleak nymph
😭
analog obsidian
yeah the model got stuck
bleak nymph
do i change settings? delete model? restart?
analog obsidian
honestly i have no idea, last time i had that issue was because my set had eq and compression filters
but yours didn't had those
i know that too high batch size may cause the model to get stuck
the dataset was too hard for rvc
probably caused by the excessive laughing
but im not 100% sure
there are rare cases where rvc really hates one voice in specific
and refuses to learn it
could be a lot of things really
bleak nymph
hmm
bleak nymph
there are rare cases where rvc really hates one voice in specific
probably that
not many voice ai's can learn my voice
i was really expressive in this dataset
analog obsidian
i was really expressive in this dataset
i believe the issue could be that the dataset is too hard for rvc
maybe you can try increasing the batch size
9, 10, 11, 12, etc
bleak nymph
hm i'll do it another day
analog obsidian
see if the dataset is damaged in a way too
bleak nymph
would i need to start from scratch
analog obsidian
maybe the denoise was too much
bleak nymph
see if the dataset is damaged in a way too
it had a lot of mic noises so i did have have to filter out
analog obsidian
things like that
analog obsidian
probably the dataset got too damaged during the cleaning process
bleak nymph
really?
analog obsidian
and that could be also the real problem
yeah
eq and compressors kills rvc training because they remove vital information needed
excessive cleaning do that as well
bleak nymph
right
bleak nymph
you can actually train a noisy dataset just fine
really?
analog obsidian
and the model will of course sound noisy
bleak nymph
yeah
analog obsidian
but it will work
analog obsidian
lol
bleak nymph
idk why it happens either, £5 mic stand fault ig
analog obsidian
i think probably something went wrong during the cleaning process
bleak nymph
natural plate reverb whenever i shout too 🔥
analog obsidian
oor it could be that the set was too expressive and hard for rvc to learn
bleak nymph
it was difficult to get rid of
i had to use an aggressive model for it
not many models recognised the reverb noise
analog obsidian
rvc can also train room reverb
bleak nymph
im gonna try the model anyway lmao, even knowing it won't be good
analog obsidian
rvc can also train room reverb
but it adds clicks to the output and honestly sounds bad
distorts the voice too much
it can't really handle room reverb
in my testings of training noisy datasets, natural noise also randomly adds clicks to the output
but in very rare ocassions
analog obsidian
in my testings of training noisy datasets, natural noise also randomly adds clic...
i trained that with this
and rvc outputs something like this
i recorded my dataset with my cheap 20$ usd mic
so far the only problem i noticed is that the model sounds metallic
im gonna add more samples to the dataset and if that doesn't fix it, it could be the noise thats causing the problem
bleak nymph
HOLD ON
ok some parts
it sounds really weird
but when i first clicked play i was blown away
sounded just like me
and then it sounded weird after that
like
some words are good
fast scarab
Hi everyone, I need some help please.
I'm trying to run Codename-RVC-Fork v3.0.4 on my PC, but I’m getting errors when launching it.
The terminal shows this error:
TypeError: argument of type 'bool' is not iterable
It also says:
Information: unable to find files for the specified model(s)
Then a bunch of errors follow related to Gradio and FastAPI, and the app crashes.
It also tells me: "When localhost is not accessible, a shareable link must be created. Please set share=True or check your proxy settings to allow access to localhost."
Does anyone know what could be causing this? I'm a bit stuck.
Thanks in advance.
analog obsidian
some words are good
imo the best you can do is to record the dataset while reading a script and pretending to speaking to someone else, using your whole voice range
thats what i did, and it's sounds like me in every audio i use
bleak nymph
oh woah
bleak nymph
imo the best you can do is to record the dataset while reading a script and pret...
what script?
analog obsidian
what script?
i asked chatgpt to do it for me haha
analog obsidian
what's the prompt u used
i've asked it to generate random text
and then pretending to talk to someone
maybe ask to generate a dialogue between to people
and pretend you're them lol
bleak nymph
lmao i just listened to the model
it has my mic noises
the de echo didn't remove it 😭
analog obsidian
the de echo didn't remove it 😭
uvr de echo doesn't remove room reverb
rx11 dialogue isolate does
actually now i think that the reason why the model got stuck is because the dataset had room reverb
karmic oliveBOT
bleak nymph
actually now i think that the reason why the model got stuck is because the data...
it's not room
it's mic stand
it genuinely creates reverb noise 😭
my room is tiny
example (amplified a bit)
fast scarab
Hi everyone, I need some help please.
I'm trying to run Codename-RVC-Fork v3.0.4...
Can someone help me, please?
fast scarab
?
odd shale
Why do you lo-o-ove me, when i'm walking away 🗣️ 🗣️
simple ore
Can someone help me, please?
gotta update gradio version @glacial pollen
fast scarab
gotta update gradio version <@1239634084133601423>
I’ve already updated Gradio. Is this a general issue?
simple ore
yeah, one of the libraries it depends on got updates last week and broke 5.13.x
newer gradio works fine
fast scarab
yeah, one of the libraries it depends on got updates last week and broke 5.13.x
I’ll be patient then.
simple ore
I’ll be patient then.
i mean you can just update it manually
fast scarab
i mean you can just update it manually
How do I update it manually? Can you guide me through the steps?
fast scarab
if code name is using conda, like applio
conda
fast scarab
``env\python -m pip install gradio==5.23.1`` from the app folder from cmd
ok thx
rancid jay
analog obsidian
he tested it and in fact they had a problem
knotty moth
he tested it and in fact they had a problem
nvm it's the room reverb thingy
tbh im not sure how much rvc tolerates against little reverb that may be left by anvuew mono derverb or RX dialogue isolate
knotty moth
not many voice ai's can learn my voice
it can also depend on pretrain used
weary stream
a
south orchid
I need a ai hub website so I can use the ai
hallow thistle
Pretrained RVC voice models are voice models that have been trained with many audio datas, typically larger than average RVC voice models (53MB).
hallow thistle
I need a ai hub website so I can use the ai
-rvc
quasi condor
click the second one
hallow thistle
Duh, you simply just called it "the AI" again. 
tropic garden
Talking about pretrains, for the newer Refinegan models can we use the cloud Applio?
formal wind
So is the lowest point on the graph or lowest spike?
knotty moth
So is the lowest point on the graph or lowest spike?
you should test several model checkpoints
formal wind
I can't do ts by ear. My hearings Ass 😭
knotty moth
Talking about pretrains, for the newer Refinegan models can we use the cloud App...
refinegan isn't available at this moment for some quality issues
knotty moth
I can't do ts by ear. My hearings Ass 😭
my ears are rather sensitive to robotic voices, esp through headphones
simple ore
So is the lowest point on the graph or lowest spike?
it is a relatively flat chart for a small(?) model
formal wind
it is a relatively flat chart for a small(?) model
The datasets 42 mins..
Small is just a "little" understatement
scarlet mesa
hey, so I'm having a bit of a problem
I just started this stuff up, and I'm not sure why, but the voice I imported doesn't seem to be working. It's just playing my voice through my headphones.
I'm not quite sure what I'm doing wrong, or if passtru is what's screwing me, but still
This is the error I seem to be getting on repeat
To the point where the client itself is telling me that it's getting frequent errors
[Voice Changer] VC PROCESSING EXCEPTION!!! Argument #4: Padding size should be less than the corresponding input dimension, but got: padding (0, 20) at dimension 2 of input [1, 128, 12]```It wants me to check if the model is loaded, which I'm pretty damn sure it is
Fuck it, I'll just figure this out in the morning
This is giving me a headache
knotty moth
hey, so I'm having a bit of a problem
I just started this stuff up, and I'm not ...
for the voice changer discussion, please go to #🔍│help-w-okada
buoyant frost
is it possible to even run rvc gui on mac?
knotty moth
is it possible to even run rvc gui on mac?
-gui
buoyant frost
what do you recommend i use?
knotty moth
what do you recommend i use?
Applio https://docs.applio.org/
buoyant frost
I thought that applio is only for windows
knotty moth
I thought that applio is only for windows
I suppose it should be similar to the installation for linux
simple ore
I use pyenv on linux, so it is pretty straightforward
install pyton 3.10 or 3.11, activate local env, pip install requirements
buoyant frost
finally got it to work thanks to chatgpt and terminal
hallow thistle
Only this time, but you can't expect ChatGPT to help you everything about the RVC program itself. 
buoyant frost
trying to get applio to work, finally got a voice model and an audio file uploaded but the conversion seems to have a problem. How long does it normally take to convert?
knotty moth
trying to get applio to work, finally got a voice model and an audio file upload...
what problem? it mostly run in cpu mode, or supposedly in mps acceleration for apple silicon M-series
keen crescent
-colab
hallow thistle
Is this the correct channel you're asking about W-Okada?
For W-Okada, go to #🔍│help-w-okada. This channel here is all about RVC programs.
clear carbon
Eh... Thought w-okada was rvc (real time voice changer)
sorry bout that
analog obsidian
Eh... Thought w-okada was rvc (real time voice changer)
sorry bout that
rvc stands for retrieval-based-conversion, it's an ai voice cloning software that can train voice models and do speech to speech conversion (convert the voice in a recording to another voice)
odd shale
Eh... Thought w-okada was rvc (real time voice changer)
sorry bout that
Nope, it's not.
W-okada is for realtime usage of models.
RVC is for model usage with prerecorded audios/inference
odd shale
rvc stands for retrieval-based-conversion, it's an ai voice cloning software tha...
Hi Lyery!!!!
analog obsidian
clear carbon
rvc stands for retrieval-based-conversion, it's an ai voice cloning software tha...
Ah.... I see, thanks for the clarification
glass rivet
is there a colab notebook where I can input audio files to train an rvc model of a particular singer?
knotty moth
is there a colab notebook where I can input audio files to train an rvc model of...
-colab
knotty moth
(the applio link)
glass rivet
Thank you so much. omg
vale osprey
-colab
empty knoll
someone knows why the kaggle notebook doesn't work, I get some kind of error with gradio and local tunel
knotty moth
someone knows why the kaggle notebook doesn't work, I get some kind of error wit...
no one could know your problem without showing the error messages and screenshot as needed
!howtoask
hallow thistle
someone knows why the kaggle notebook doesn't work, I get some kind of error wit...
Which Kaggle link are you using?
Also make sure you screenshot the problem you encountered somewhere on Kaggle fror more context. The "Kaggle notebook" doesn't refer to a specific RVC notebook link.
low shard
@empty knoll what kaggle link? What issue? What u want to do and what’s ur pc gpu?
frail umbra
im training this green model and it just keeps getting better
knotty moth
im training this green model and it just keeps getting better
why are you pushing crazy epochs? and you seem to have messed up some training configurations 
frail umbra
why are you pushing crazy epochs? and you seem to have messed up some training c...
thats only like 500
I did give it like almost 2 hours of material
and it was all split into sentences by hand
knotty moth
thats only like 500
well you should test the model and find out what's wrong on it
frail umbra
idk how
knotty moth
what kind of misinformation did you get from?
frail umbra
I got a feeling it's gonna be good
the red one was a lil lackluster, I didnt use any pretrain though
it wasnt bad
knotty moth
how do you know it without the actual testing by inference?
you have baked a cake, how would you think it is "good" without tasting it?
shadow sparrow
Hey everyone, I'm training an AI model with Applio and ran into a TensorBoard error on the Colab page that says:
'Data could not be loaded. The TensorBoard server may be down or inaccessible. Last reload: Mar 31, 2025, 2:27:49 PM. Log directory: logs.'
I'm not sure if this is a solution or not, but since I can't access TensorBoard, I stopped the training on Applio, enabled 'Overtraining Detector,' and set it to stop after 50 epochs without improvement—just so the training doesn't run indefinitely.
Does anyone know what might be causing this issue or how to fix it?
analog obsidian
line higher = mo' betta
the graphs are named "loss" because they measure the errors in the ai calculations
so graph going down = less errors
graph going up = more errors
knotty moth
Hey everyone, I'm training an AI model with Applio and ran into a TensorBoard er...
you need to start training and reach at least epoch 1, then click refresh button
also don't lose the tfevents file in model's eval folder
flint solar
nah nah nah I can tell something's cookin
the graphs are named "loss" because they measure the errors in the ai calculations
so graph going down = less errors
graph going up = more errors
frail umbra
yeah but this isn't a conventional voice
simple ore
you need to look at other chart to see which one is messing the whole thing up
fm and mel
flint solar
its probably fm
shadow sparrow
you need to start training and reach at least epoch 1, then click refresh button
I did reach on epoch and refreshed but it doesn't work im at epoch 65 and it still doesn't work, how do I see if I haven't lost the tfecents file, plus does overtraining detector do any good?
frail umbra
idk what these are
knotty moth
yeah but this isn't a conventional voice
listening to the inference results could tell you what's wrong, so stop judging it by yourself
frail umbra
im not at a checkpoint for another hour
flint solar
why arent u using a pretrain
flint solar
yeah its fm
frail umbra
why arent u using a pretrain
because i wanted to try it raw
the pretrain has an accent
knotty moth
d/total approaching zero means it suffers mode collapse, which means the model would produce relatively same results, i.e. static noise
frail umbra
that's good
knotty moth
that's because you didn't use pretrain for the green graph model too
frail umbra
no I did
flint solar
d/total approaching zero means it suffers mode collapse, which means the model w...
why the hell is mel at 50
frail umbra
it's 2x as fast
knotty moth
it's 2x as fast
so, faster means better, right?
frail umbra
it means the pretrainer is active
knotty moth
you still haven't figured out what's wrong, what are you smoking of? 
simple ore
This is so weird.. even without pretrain fm and mel should go down
this is complete fuckery
like trying 32 batch maybe with a small set
flint solar
funny ass graphs
knotty moth
with default mainline configurations and insufficient dataset for training from scratch, it will never be good, the results will be the same static noise, causing the discriminator easily identify the generated samples as fake (hence the d/loss keeps approaching zero)
frail umbra
it has pretrain
8 batch
somewhere between 90min-120min of content
and 2000+ files separated into sentences (by hand)
analog obsidian
i see the old graphs but could be also the compiled version of applio
disabling the pretrain in mainline is a bit tricky (just dont put anything in the pretrain path 
)
knotty moth
that's why you should use an existing pretrain, and the process is called finetuning
training from scratch would need massive amount of dataset containing several speakers (for example, VCTK dataset)
analog obsidian
This is so weird.. even without pretrain fm and mel should go down
imo the option to disable the pretrain should be hidden by default because this is not the first time something like this happens lol
simple ore
imo the option to disable the pretrain should be hidden by default because this ...
it is
simple ore
you have to go to advanced options and uncheck it
okay, so using pretrain and 2 hours of audio
frail umbra
my expert analysis just tells me we have to push this to 2000 epochs or beyond
simple ore
can you send a piece of dataset to check?
with a pretrain training to 2000 epochs would wipe most of the pretrain's knowledge, especially if you got 2hr set
flint solar
with a pretrain training to 2000 epochs would wipe most of the pretrain's knowle...
he said its 500 epochs
frail umbra
here's a random selection
flint solar
what the fuck is this bullshit
frail umbra
it's my uncle
flint solar
are these recordings from the 60's
and why is it .wav.mp3
😭
so much shit going on
frail umbra
the audiofiles all been converted a couple times...
simple ore
6khz from shitty chinese answering machine
knotty moth
bro
#✨│ai-help message I still can't believe ts 
simple ore
i mean... this is real
frail umbra
simple ore
i've not tried to run training with shitty recording like that
crude flame
💀
simple ore
no, it is from an answering machine
but it is not even 12k
i've resampled my set to 12k and it still sounds better than this shit
knotty moth
the worst thing I ever had was when trying to train a dataset from some mobile games in 96/128 kbps mp3 quality, yea it sounded robotic as hell but the graph pattern with the default mainline pretrain was not really much different
simple ore
simple ore
I think it would be better to just get good 30-60s and then use that as reference for XTTS
and make some tts outputs
frail umbra
I had a really good model but I accidentally deleted all my AI stuff
😢
it was GOOD
excellent
i dont remember how I made it really
knotty moth
I think it would be better to just get good 30-60s and then use that as referenc...
how about microsoft sam in windows xp
simple ore
i mean to clone the voice
it will be less shitty, even considering that synthetic audio is not good for training
knotty moth
rvc miku is going real
knotty moth
I did reach on epoch and refreshed but it doesn't work im at epoch 65 and it sti...
make sure tensorboard points to the logs folder, and if ur using latest applio it should log every epoch
shadow sparrow
make sure tensorboard points to the logs folder, and if ur using latest applio i...
how do i do this??
knotty moth
how do i do this??
refer to the guide as im lazy to explain rn
steep stirrup
yes
hallow thistle
Go back to #🔍│help-w-okada.
hallow thistle
This channel #✨│ai-help is about RVC programs, not W-Okada.
idle osprey
This channel <#1159290139609137264> is about RVC programs, not W-Okada.
sorry
steep stirrup
I used three fork models
Mangio-RVC
applio
RVC1006Nvidia
hallow thistle
I used three fork models
Mangio-RVC
applio
RVC1006Nvidia
Applio is the best one. The rest is outdated.
What is your PC GPU?
steep stirrup
What is your PC GPU?
3080ti
steep stirrup
ApplioV3.2.8
Yes, but the problem was that the need for embedders, which cannot be found in our language, was an example of the use of which the sound became extremely awful.
simple ore
while the embedder is mostly trained on english, your sibilant and breathing problem is not related to that
steep stirrup
while the embedder is mostly trained on english, your sibilant and breathing pro...
Do you think if I add more pronunciation of s and breathing sound to the data set, the problem will be solved?
crude flame
the sibilants sounding bad are prob due to them being overfitted or you only have those bas esses in the dataset
frail umbra
refer to the guide as im lazy to explain rn
https://www.youtube.com/watch?v=Ts_kPyK8mek
found u a video
steep stirrup
I don't know what to do, I'm confused
knotty moth
https://www.youtube.com/watch?v=Ts_kPyK8mek
found u a video
why that random video?
flint solar
I don't know what to do, I'm confused
How long is ur dataset
empty knoll
<@1044021509435445258> what kaggle link? What issue? What u want to do and what’...
https://www.kaggle.com/code/deiant/applio
I train a model since my gpu is not Nvidia and I have to use kaggle
rx 7800 xt
low shard
https://www.kaggle.com/code/deiant/applio
I train a model since my gpu is not N...
Btw you could try locally with applio on AMD https://docs.applio.org/applio/getting-started/installation#amd-gpu-support-windows
empty knoll
I had already used it a week ago but for whatever reason now it doesn't work 
low shard
Also, have you tried deleting the Kaggle notebook and refolloeing the guide steps?
empty knoll
Also, have you tried deleting the Kaggle notebook and refolloeing the guide step...
yes
frail umbra
OK U GUYS WERE RIGHT THE MODEL IS NOTHING BUT SINE WAVES
I think I mismatched the sample rate or something
flint solar
https://www.kaggle.com/code/deiant/applio
I train a model since my gpu is not N...
!uv pip install pydantic==2.10.6
Try running this command before starting the gui
flint solar
OK U GUYS WERE RIGHT THE MODEL IS NOTHING BUT SINE WAVES
Send a screenshot
flint solar
In my 2 years of making rvc models I never seen anything like this before
flint solar
@simple ore wtf is this
crude flame
I think I mismatched the sample rate or something
its bec the dataset is horrible
frail umbra
yeah it shits out between 20 and 40 epoch
crude flame
How could it be THAT ass
has to be on purpose
frail umbra
I didnt see it
flint solar
I didnt see it
U won’t because it’s censored in ai hub
I think I’m getting timed out lmfao
flint solar
ok I resampled them all at 32khz
Can u send a portion of the dataset
crude flame
ehhh I made them all wav's
flac 
flint solar
and now it's really big
Send one file
frail umbra
and I deleted the mp3s...
empty knoll
``!uv pip install pydantic==2.10.6``
Try running this command before starting th...
fixed thanks friend 
flint solar
fixed thanks friend <:steamhappy:1275251128359194647>
👍
frail umbra
flint solar
Why the HELL are u training ts
frail umbra
ts?
crude flame
just how
flint solar
4k
flint solar
Who is this racist fella
crude flame
how does someone get this bad audio
frail umbra
how does someone get this bad audio
i dont think it's even possible to re-create accurately
flint solar
how does someone get this bad audio
Noobies said it’s audio from Chinese answering machine
6k audio is funny asf