I don't mind.
#✨│ai-help
wooden lotus
Idk select all of it and joint into one
wooden lotus
use audacity or sth idk
twilit gulch
Fn
Gn
so how do I use the model I just made?
Oh
Only 25 min worth of audio
:/
vagrant marsh
wide ingot
pip install python-dotenv
vagrant marsh
pip install python-dotenv
Clone the applio rvc first before to type "pip install python-dotenv"?
I mean type "cd <applio-rvc-folder>"?
wide ingot
If you have Python installed system-wide. I think there's no need to.
wide ingot
Maybe run pip as admin?
vagrant marsh
Maybe run pip as admin?
No no no, I solved it already
sand iris
That was python 3.11, don't think that'll work
wide ingot
Okay.
final stream
-colab
azure marshBOT
twilit gulch
What’s inference
sand iris
converting voice
wide ingot
Basically models trying to predict what the target voice will sound like given the input data.
rare gobletBOT
@wide ingot has reached level 4.
twilit gulch
Basically models trying to predict what the target voice will sound like given t...
Basically, use dataset
Use the model*
A model?
MAKE THE LINE WITH THE AI VOICE
wide ingot
The dataset is the stuff you train the AI model. It can be any model.
So yes, you need a trained model to do inference.
@twilit gulch
twilit gulch
<@845696636654321665>
Basically
Inference
Is using ur model
So u can make covers
Is 600 epochs for 25 min worth of audio enough
Or do I go in the 1000s?
wide ingot
You can make just about anything, not just covers. As for epochs, there's no recommended amount. Gotta open up TensorBoard and watch for signs of overfitting or overtraining.
I trained a dataset of 90 minutes, overfitted at around 500/600 epochs.
twilit gulch
Well I cant use tensor if it error 404s
rare gobletBOT
@twilit gulch has reached level 4.
wide ingot
Install it locally.
My Linux repository (a collection of software) has TensorBoard available, so I did that.
The 403* error is a known problem involving Chrome and Colab.
It's said that switching browsers like Firefox works around it.
So, if you don't want to install TensorBoard locally, you can switch web browsers that's known to work. And if you want to keep Chrome as your training browser, install TensorBoard locally.
torn seal
Ok so who uses MEDLEYVOX? ANYONE?
twilit gulch
Install it locally.
no my specs are far too weak
wide ingot
I mean, after you stop training on Colab, download the training logs of your model and run local TensorBoard from there.
snow pasture
Having trouble with this. I'm using a light voiced model that's struggles with deep voices. Do I increase it to correct the voice from being broken?
twilit gulch
I mean, after you stop training on Colab, download the training logs of your mod...
h o w
pearl arrow
how do you access a file generated in colab? lmao i'm stupid
twilit gulch
Having trouble with this. I'm using a light voiced model that's struggles with d...
try NOT putting a whole song at once btw, so it takes less time to process
snow pasture
It's only vocals.
wide ingot
If you're using the RVC Disconnected, go to your Google Drive and navigate the "rvcDisconnected" folder. There's another folder that contains your model's name, which holds the current state of training. Select it to download. It'll be a zip file.
twilit gulch
If you're using the RVC Disconnected, go to your Google Drive and navigate the "...
I meant the local RVC tensor stuff
pulsar lake
is there a tutorial on how to use applio
rare gobletBOT
@pulsar lake has reached level 1.
twilit gulch
not where to get my data from (to input it in said tensor)
wide ingot
Hmm...that I don't know. I need more context.
twilit gulch
Hmm...that I don't know. I need more context.
how do I set up LOCAL tensorboard
twilit gulch
yes nd yes
wide ingot
Open a command prompt to type pip install tensorboard.
wide ingot
Python should come with pip.
wide ingot
You can update, but it's not necessary.
twilit gulch
oh
OH
I did pip install tensorboard.
not pip install tensorboard
it's the fullstop
wide ingot
Yeah, languages can make things a little complicated.
wide ingot
Type tensorboard to a random folder to ensure it's working.
If it says "TensorFlow installation not found", you can ignore this as it's irrelevant.
twilit gulch
Type `tensorboard` to a random folder to ensure it's working.
wym
wide ingot
wym
Like this: tensorboard --logdir YourFolder
Replace YourFolder with your folders to test.
wide ingot
Understood. I also don't know much about Python as well.
Open a web browser and type localhost:6006 in the address bar.
wide ingot
No. It's just graphs out your model's training state.
rare gobletBOT
@wide ingot has reached level 5.
wide ingot
You can, but I don't recommend it.
pulsar lake
where can i find my model
twilit gulch
You can, but I don't recommend it.
so when
wide ingot
For me using RVC Disconnected Colab, I train for 3 hours, stop, and download the folder inside rvcDisconnected.
wide ingot
It's totally up to you how long you gonna train your model.
You can do 2 hours, 1 hour, etc.
wide ingot
The folder that has your model, eg the one that has D and G and such. For Mangio, it should be inside the logs folder.
wide ingot
Should be. Click on it.
twilit gulch
wide ingot
Hmm, it seems you just started to train.
rare gobletBOT
@twilit gulch has reached level 5.
wide ingot
Either your Colab didn't download the D/G files and the events.out.tfevents files, training has to start over. Sorry.
twilit gulch
Either your Colab didn't download the D/G files and the events.out.tfevents file...
fuck this I give up man
I tried 5 times 💀
wide ingot
Switching to Firefox may help.
twilit gulch
ima just see if I can find anyone willing to assist me w training w local (since I know how to use local) and just do it with their specs or some shit
collabs I ain't a fan of man
wide ingot
Local is almost the same as Colab.
twilit gulch
Local is almost the same as Colab.
more stable
wide ingot
That too.
wide ingot
Since you don't have to deal with disconnections.
twilit gulch
Since you don't have to deal with disconnections.
so what do I do
I really REALLY need to train this shit
wide ingot
Do you have Mangio running?
What's your GPU? If it's an old NVIDIA or AMD/Intel, I'm afraid you're out of luck.
If that's the case, Colab is your only option sadly.
wide ingot
An NVIDIA RTX card is recommended for decent training, although I heard that you'll need a card with at least 6GB of VRAM to do so.
wide ingot
That's a big yikes.
formal current
how to make dataset?
twilit gulch
get voices of characters
wide ingot
And if you have some spare money to spend out of pocket, you can consider upgrading it to an RTX.
twilit gulch
just give me a recommendation
wide ingot
RTX 3060. At around $300, this card will make the most out your money.
At least in the USA.
wide ingot
You can spend that much money on a 4090.
twilit gulch
fuck no
im not wasting more money on a hobby
I ain't even over 16 yet brother, I need to learn how to use my money wisely
half cove
wide ingot
Yes, gotta do budgeting.
wide ingot
No, any motherboard is fine.
twilit gulch
any power supply specifications?
oh and also what about my cpu?
Intel(R) Core(TM) i3-8100 CPU @ 3.60GHz 3.60 GHz
pulsar lake
any power supply specifications?
i5 12400 rtx 3060/3060ti should be enough to do most things
wide ingot
CPU doesn't matter.
pulsar lake
ur right
twilit gulch
also
will I be overloading my CPU or ram or someshit
I ain't know how computers computer
I jsut know some shit tends to break
wide ingot
For power supply, I recommend one over 500 watts.
twilit gulch
can I check that?
wide ingot
If an RTX 3060 is too much, you can get a 3050 instead. About $100 cheaper but lower VRAM.
twilit gulch
any upgrade my pc needs
pulsar lake
what are ur specs rn
pulsar lake
u should be fine witha 3060 and that cpu
rare gobletBOT
@pulsar lake has reached level 2.
wide ingot
A GPU upgrade is all you need.
twilit gulch
what if im playin roblox or sum shit
because yes I am a little kid and I play roblox
pulsar lake
that means its working yea
twilit gulch
or do I have some kind of virus 💀
wide ingot
RAM may not be enough.
pulsar lake
yea
twilit gulch
AC115v-230v
twilit gulch
do I fr have to open my pc case
pulsar lake
u can prolly just search it up
twilit gulch
like uh
pulsar lake
is it prebuilt
twilit gulch
3 years old
wide ingot
do I fr have to open my pc case
Yes, for real.
rare gobletBOT
@wide ingot has reached level 6.
twilit gulch
(prebuilt would've been so much better but it was a learning curve)
pulsar lake
u cant dig around and find out what psu u bought
twilit gulch
Yes, for real.
welp, I'll try to find my toolbox
twilit gulch
I've opened my CPU fan and shit several times so I know how to do it
na ima just do it the old fashioned way
I'll get to that give me like 5 mins
pulsar lake
what should the number on the left be
aboyut
its been running for about 10-20 minutes now
pulsar lake
probably
twilit gulch
As in the one w the glass
wide ingot
Is your PSU at the top or bottom?
twilit gulch
Bottom
wide ingot
I think it's other side with the storage drives.
wide ingot
The ones with SSDs/HDDs.
twilit gulch
rare gobletBOT
@twilit gulch has reached level 6.
wide ingot
It's not clear from the picture.
wide ingot
650W
wide ingot
That should be enough.
twilit gulch
That should be enough.
So all I need is just to remove the gpu
And place in the new one?
wide ingot
Yea
twilit gulch
so which one?
wide ingot
Replace the GT 1030 with the new card.
Or, you can use the new card just for training and keep the GT 1030 for gaming.
That is, if your motherboard supports a dual GPU setup.
twilit gulch
or can i keep them as .mp4 files
Import to audacity via ffmpeg
Then make it mp3
Mp4 only supports mp3 codec
half cove
:0
wide ingot
My motherboard does. I have two AMDs. One for general tasks, the other for realtime RVC.
loud hazel
there are new rvc projet out
wide ingot
You mean 3050/3060?
loud hazel
pulsar lake
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/releases/t...
is there a vid on how to use
loud hazel
is pre-made
wide ingot
Oh that's news. A better RMVPE model, cool.
loud hazel
download ndivia or amd/intel
pulsar lake
ill try it out
loud hazel
2023-10-06
We have created a GUI for real-time voice change: go-realtime-gui.bat/gui_v1.py (Note that you should choose the same type of input and output device, e.g. MME and MME).
We trained a better pitch extract RMVPE model.
Optimize inference GUI layout.
Future Plans:
Features:
Add option: extract small models for each epoch save
Add option: export additional mp3 to the specified path during inference
Support multi-person training tab (up to 4 people)
Base model:
Collect breathing wav files to add to the training dataset to fix the issue of distorted breath sounds
We are currently training a base model with an extended singing dataset, which will be released in the future
luv this one : Support multi-person training tab (up to 4 people)
twilit gulch
You mean 3050/3060?
shi man idk what do I buy
wide ingot
What cards do they offer there?
twilit gulch
The ropes?
pulsar lake
if u cant afford the 3060
wide ingot
Search for rtx and see what they have.
loud hazel
i got mine a asus tuf 3060 12g at 320e
twilit gulch
?
Wait
What the
Isn’t this worse?
What’s a 3060ti
I don’t
I give up
My brain boom
pulsar lake
the 3060ti is better
wide ingot
In terms of VRAM, yes. But I think the Ti is a better performing card, that's why is pricier.
pulsar lake
thereisnt a huge difference
loud hazel
only 3% of compute on ti vs normal
wide ingot
In other words, 97% of computing power of the 3060 Ti against normal 3060.
pulsar lake
3% difference most likely
rare gobletBOT
@pulsar lake has reached level 3.
twilit gulch
What
loud hazel
looks on nvidia dev website u can finds compute on each cards, i lost url...
Geforce RTX 3060 Ti 8.6 compute
Geforce RTX 3060 8.6 compute
lol
fossil glen
-colab
azure marshBOT
loud hazel
just buy what's u like
wide ingot
I would've just get a 3060 and forget about it.
wide ingot
It's that simple. It doesn't have to be difficult.
twilit gulch
3060 it is
also,
With this I will be able to train any voice I want?
(Locally ofc)
wide ingot
Spot on. Though might need RAM upgrade too.
wide ingot
Less than that of GPUs.
wide ingot
A 64GB DDR4 kit costs $100 US dollars. But I highly doubt you need that many.
A 32GB costs half of that.
loud hazel
16go is oki aslo
twilit gulch
Let me check my ram rw
faint bronze
twilit gulch
Rq
wide ingot
16GB DDR4 is about $20 to $30
worldly oasis
Can I somehow test how well is my model trained ? I thought it should be really good because I had 30min of high quality audio, but when I used it for AI covers, it sounds like shit.
wide ingot
This is where you open up the side panel with the motherboard.
wide ingot
So 2x4GB DDR4.
wide ingot
You can use a page file in Windows, which means it'll use your drives as RAM.
twilit gulch
Fr?
wide ingot
More page file size, that is.
wide ingot
Drive speeds are often slower than RAM speeds, including SSDs.
twilit gulch
Drive speeds are often slower than RAM speeds, including SSDs.
But will it work for model making
wide ingot
It may work.
brittle wing
how can i not hearing myself in the voice changer while im using it in a game?
wide ingot
But it's gonna slow to a crawl once memory runs out and uses your drive.
wide ingot
how can i not hearing myself in the voice changer while im using it in a game?
@brittle wing #🔍│help-w-okada might be the channel you're looking for.
Anyway, it's more worth it to upgrade RAM than is to use additional page file.
16GB should be fine in your situation.
And if your motherboard supports 4 RAM sticks, you can install alongside, increasing it to 8GB + 16GB = 24GB.
You can buy RAM and GPU at the same time.
I bought a CPU and motherboard together to replace bad components, leading to the conclusion that my old board was faulty.
ebon wraith
yo need
sent like 3 messages
then you will ahve the abillity
I have this problem whily tryed to start up the proggram
is there any way i can fix it?
i mean programm didnt start up after that
wide ingot
That looks like a question for #🔍│help-w-okada.
rare gobletBOT
@wide ingot has reached level 7.
wide ingot
I saw your post in that channel.
ebon wraith
ye
wide ingot
So you're good.
brittle wing
Anyone else having an issue with voicemy.ai with training where it gives an error message when attempting to train with a voice file?
*uploaded file
wide ingot
If not, then it's a skill issue.
ebon wraith
U need to get lvl 1
serene citrus
how
do
i
do
that
im
just
typing
so
i
can
get
up
level
this
rare gobletBOT
@serene citrus has reached level 1.
proper shale
Refresh and select from the dropdown
zinc anchor
@serene citrus It won't work, the level bot has a time cooldown
serene citrus
Refresh and select from the dropdown
ye doesnt work
earnest ruin
yo can someone help me in dms i tried using applio but my ai cover sounds a little crispy with noise pops every now and again. i don't wanna post the mp3 here tho
or do i just need to use audacity to iron those out
serene citrus
serene citrus
@proper shale tell me im doing something wrong?
ebon wraith
Is there any way i can fix it? the programm wont start up
rare gobletBOT
@ebon wraith has reached level 2.
stuck zealot
someone help me please
rare gobletBOT
@stuck zealot has reached level 1.
proper shale
Select from the "Auto detect audios" thing dropdown
proper shale
its just using my regular voice
disable passthrough
stuck zealot
merci
is there any way i can deepen the voice of the voice changer to make it sound more like dr phil ?
proper shale
decrease tune
stuck zealot
alright thanks, another question, what is the function of the chunk setting ?
proper shale
basically, how much time it'll take to make the change in voice
stuck zealot
is there any extra way i can improve how the voice sounds?
crimson basalt
I got bored so made a few changes. 🙂 My question is... Is there any themes other than The standard one and Acat?
proper shale
I got bored so made a few changes. 🙂 My question is... Is there any themes othe...
LMAOOOOO THATS
I never thought you could do that
Huh...
I think there are really
crimson basalt
Pretty easy to do. Just change the strings in the theme file. Acat had an anime girl in the bottom right so I changed it to my logo.
lean owl
is there a way to prevent colab from disconnecting
wise jolt
How do I turn google drive zip files into huggingface links?
rare gobletBOT
@wise jolt has reached level 1.
lean owl
How do I turn google drive zip files into huggingface links?
upload the zip file to a huggingface space
wise jolt
upload the zip file to a huggingface space
which sdk do i choose?
lean owl
whats sdk
wide parcel
is there a dockerized version of Applio
sage dagger
ok i made a finneas model but the pth file isn't in the "weights" folder
lean owl
how do i import an audio file into this?
honest cobalt
why do I keep getting no-feature-todo
acoustic shale
how many epochs for a 2:19 dataset
wide ingot
why do I keep getting no-feature-todo
That may happen if you set the number of threads too much with a small dataset; some threads won't get a chance to extract features.
glad zealot
how many epochs for a 2:19 dataset
As much as you want, also depends on you dataset quality
honest cobalt
That may happen if you set the number of threads too much with a small dataset; ...
ok thanks
rare gobletBOT
@honest cobalt has reached level 1.
wide ingot
So, I think it's safe to ignore this message.
nocturne mural
how do i import an audio file into this?
/datasets/ name random
rare gobletBOT
@nocturne mural has reached level 1.
nocturne mural
nocturne mural
how do i import an audio file into this?
Although, you're in a hugginface environment, so it's impossible to do it.
lean owl
Although, you're in a hugginface environment, so it's impossible to do it.
i guess thats true
honest cobalt
That may happen if you set the number of threads too much with a small dataset; ...
how do I change the number of threads
wide ingot
The "Number of CPU processes" slider
honest cobalt
ok
wide ingot
If it still says "no-feature-todo", perhaps you haven't processed the dataset yet.
acoustic shale
As much as you want, also depends on you dataset quality
welp already finished training on 250 but thanks ig
mortal pawn
Hey my computer is without internet because I'm having issue with the internet provider and Now Applio is not opening despite being able to before
wide parcel
pip3 install -r assets/requirements/requirements.txt
ERROR: Invalid requirement: 'gdown+https://github.com/wkentaro/gdown.git' (from line 22 of assets/requirements/requirements.txt)
Hint: It looks like a path. File 'gdown+https://github.com/wkentaro/gdown.git' does not exist.
i just changed it to git and its downloading fine 🤷♂️
wide ingot
Hey my computer is without internet because I'm having issue with the internet p...
Applio should work without Internet access, as the network traffic only goes through your computer.
sage dagger
ok if the pth isn't in the weights folder where is it
mortal pawn
Applio should work without Internet access, as the network traffic only goes thr...
When I try to run the program it gives me this error regardless of what option I pick
winter pier
how do i import an audio file into this?
What is that interface? Where can I download it?
rare gobletBOT
@winter pier has reached level 1.
lean owl
What is that interface? Where can I download it?
its on huggingface
winter pier
its on huggingface
Do you got link pls?
proper shale
urban latch
will rvc run if i pay for colab pro?
sage matrix
no
craggy goblet
anyone have a "disconnect prevention method" for rvc disconnect
sage dagger
ok got my dataset
sage dagger
but it keeps saying
['infer/modules/train/preprocess.py', '', '48000', '11', '/home/user/app/logs/Finneas ', 'False', '3.7']
Fail. Traceback (most recent call last):
File "/home/user/app/infer/modules/train/preprocess.py", line 119, in pipeline_mp_inp_dir
for idx, name in enumerate(sorted(list(os.listdir(inp_root))))
FileNotFoundError: [Errno 2] No such file or directory: ''
grizzled cliff
which file should i open. i only ran it once and i cant run it again
grim needle
how do i copy and paste a huggingface link like this one https://huggingface.co/EvanTheToonGuy40/RVC_Models/blob/main/BanditHeeler.zip
wide parcel
apparently applio doesn’t work at least on stock linux and i believe the website says that..but either way does mangio have a large enough performance gap to where it would be better to run a windows emulation layer on linux or should i just set up windows server or a minimized windows 10 machine?
glossy grove
@wide parcel I git applio running on Arch Linux with a few tweaks to the dependencies
The main issue was pyworld
wide parcel
i was fine until the actual training part of training
glossy grove
Ah
wide parcel
lemme get my err msg
Traceback (most recent call last):
File "/usr/lib/python3.10/multiprocessing/process.py", line 314, in _bootstrap
self.run()
File "/usr/lib/python3.10/multiprocessing/process.py", line 108, in run
self._target (*self._args, **self.kwargs)
File "/root/Applio-RVC-Fork/lib/infer/modules/train/train.py", line 327, in run train and evaluate(
File "/root/Applio-RVC-Fork/lib/infer/modules/train/train.py", line 540, in train_and_evaluate scaler.scale (loss_disc).backward ()
File "/usr/local/lib/python3.10/dist-packages/torch/tensor.py", line 487, in backward
torch.autograd.backward (
File "/usr/local/lib/python3.10/dist-packages/torch/autograd/ init.py", line 200, in backward # Calls into the C++ engine to run the backward pass
Variable._execution_engine.run_backward(
RuntimeError: [../third_party/gloo/gloo/transport/tcp/pair.cc:598] Connection closed by peer [127.0.0.1]:13533]
glossy grove
I can try ot when I get home. Still using my mangio midels
rare gobletBOT
@glossy grove has reached level 1.
glossy grove
Hm.
glossy grove
For one I'm running it in python3.9
Not sure if that matters. But most of my AI stuff is stable on tyat
Gimme 20 minutes and I'll be hime
I'll try out mine. If it works I'll pass you my files
wide parcel
brother it’s 11pm for me i’m heading to bed
but yeah please do
feel free to flood my dms with info i’ll read it in about 8h when i get up
rare gobletBOT
@wide parcel has reached level 4.
grim needle
how do i construct a huggingface url
tribal badge
-colab
azure marshBOT
visual pumice
should my dataset just be 1 audio file or should i split it up
glossy grove
it;s reccomented to cut it up
visual pumice
I wonder why?
glossy grove
in my case it helps with VRAM
empty orchid
So has Google just outright banned free users from doing anything related to RVC even without a GUI? I was just trying to use the AICoverGen no UI colab & it still disconnects me after a few minutes.
urban sable
how do i fix output coming out my mic while using RVC?
rare gobletBOT
@urban sable has reached level 1.
proud elbow
should my dataset just be 1 audio file or should i split it up
For UVR processing? yes you may want to do for 15+ mins file depending on the system RAM and VRAM, and model you're using.
As training data? it doesn't make difference, but too many split files could probably make training slower.
serene musk
should my dataset just be 1 audio file or should i split it up
I suggest 1
visual pumice
i already started training hours ago
its a 2 minute dataset i just split up in 6 audio files
proud anvil
Tell me why google colab does not see the model, everything is done, I took a ready-made one and threw it into google drive, but when you add, it is empty, what to do, I saw that covers are made with this voice
empty orchid
So has Google just outright banned free users from doing anything related to RVC...
Can I get an answer for this please?
main patio
Just finished installing Replay and downloading all the necessary files but the app wouldn't open. It opens once but it refuses to open again after an error.
Anyone know how I can open it?
late raptor
-colabs_br
azure marshBOT
main patio
Does the Google Collab still work?
proud anvil
Tell me why google colab does not see the model, everything is done, I took a ready-made one and threw it into google drive, but when you add, it is empty, what to do, I saw that covers are made with this voice
rare gobletBOT
@proud anvil has reached level 1.
glad zealot
Does the Google Collab still work?
Which one? Trainer or covergen?
main patio
Which one? Trainer or covergen?
Easy GUI
calm trench
EVERYTIME I try to open it, it immediately closes, is there anything else that I need to run it?
glad zealot
Easy GUI
thats trainer just use rvcdisconnected if youwanna do colab /its in #1159949278270193734
river agate
glad isle
I get something called [VCClient] wait web server...290 http:// and some numbers
What's that
fresh oxide
-colabs_br
azure marshBOT
sand iris
I get something called [VCClient] wait web server...290 http:// and some numbers
It's having an issue
sand iris
Usually the fix is to remove the pretrain and model_dir folders that are found in the folder you unzipped
glad isle
I'm reinstalling the program
sand iris
Should download something on next launch and work again
glad isle
Will that help?
sand iris
That would also work yeah
glad isle
Thanks mate!
lapis harness
help
my voice changer is just spamming "Pipeline is not initialized" how do i fix that?
proper shale
main patio
How do I use RVC disconnected?
rare gobletBOT
@main patio has reached level 1.
proper shale
How do I use RVC disconnected?
There will be a guide for it soon - don't worry! As of right now, I recommend reading each and every step, it should be easy to continue this way.
You need to format your dataset in a different way though, so look out for that
candid dust
is it some sort of collab cause i still use local which seems to be the best method by far
candid dust
Yes! It is a Colab.
hello brother im doing good thanks 🙂 hope you are too?
proper shale
Yeah, I'm doing alright, haha. By the way - did you ever figure out how to fix the feature extraction thing with no pitch guidance models?
candid dust
good to hear bro, and i only could do it by a glitch on the old colab version of rvc v2 seems the local one is still broken, apple rvc seems to have the option to remove the pitch guidance but i cant use the newest one cause it only supports win 11
rare gobletBOT
@candid dust has reached level 1.
candid dust
so no use to me lol
so im kinda sticking with rvc 1 atm
if u use the original rvc github source that one works with pitch guidance off
proper shale
oh damn
candid dust
just means not using mangio fork
proper shale
Because I recall making a model without pitch guidance - however it kinda sucked lol
And I use Colab a lot so
candid dust
i havent tried the newest colabs but it might work now i just havent tried them
but i have noticed that with rvc v2 the none pitch guidance off compared with rvc v1 guidance off is very different rvc v2 doesnt do so well
i cant put my hand on why that is
candid dust
https://www.youtube.com/watch?v=G9T-sk5_5qs
damn lol they really cracking down on it
vague sedge
yes they are
lament eagle
What is target_sample_rate in training?
rare gobletBOT
@lament eagle has reached level 4.
half cove
<@352664453541658624> has reached level **4**.
skill issue
hallow crane
Hi
Can someone explain to me why after doing the Train feature index, the index file weighs only 31mb, on a 1GB dataset in .wav...
1000 Epochs
peak crescent
i dont have gpu is there any way i can use the rvc?
half cove
i dont have gpu is there any way i can use the rvc?
inference or training?
wide parcel
It's 11 pm here btw
?
ember ermine
can someone help me out? i would like to edit the infer-web.py to not automatically start the browser. just launch the cmd and run on the local url but without the popup
how can i do that? what do i remove from the file?
stone dawn
hellooo is this normal in applio local?
proper shale
can someone help me out? i would like to edit the infer-web.py to not automatica...
i dont think u can tbh
only if u know rvc code a lot
ember ermine
hmmm
oh well
is there a way to drag and drop files into the webui to convert them that way?
right now i have them all in audio folder
proper shale
i think only in applio
ember ermine
i dont think u can tbh
i just did it lol
rare gobletBOT
@ember ermine has reached level 2.
proper shale
a
ember ermine
at the bottom of infer-web.py
remove the not before config.noautostart
now the cmd launches without a popup and i can click a bookmark to view the webui instead
nice
proper shale
oh
wide parcel
hellooo is this normal in applio local?
yeah
wide parcel
I'll try out mine. If it works I'll pass you my files
lmk
main patio
Where do I get model links that are compatible to rvc disconnected?
smoky knot
why cant i send any files showcasing the ai voices here
brittle wing
could someone please help me with paperspace 1:1, drop me a dm. thanks
livid mango
how to put data set in mangio engine
potent root
-colab
azure marshBOT
grim needle
how do i construct a huggingface url
tough shoal
how do i get back on posting model voice
nimble rock
Just for reference before I post my models again, are we allowed to post voice lines of us talking with the model? Or no voice lines at all to prevent copyright?
graceful obsidian
iirc, it's a matter of datasets / sources of datasets @nimble rock
and not really model or model-related content
brittle wing
Uh my dataset that's training is the stuck at 6 epochs and 0%
rare gobletBOT
@brittle wing has reached level 1.
brittle wing
@rare goblet ratio
graceful obsidian
Uh my dataset that's training is the stuck at 6 epochs and 0%
check the task manager for gpu usage
in case you use applio or whatever, I can't help. I only support legacy rvc workflow
brittle wing
Oh
graceful obsidian
However, might be you used too high batch_size or either something bugged / glitched in which case, try training again or report it to whoever handles applio ( I assume
brittle wing
The batch is 8 so probably
graceful obsidian
what's ur gpu's vram, in case it's 4, 6 or 8gb, try batch of 6 or 4
fickle moon
Traceback (most recent call last):
File "C:\Users\kevin\Desktop\Applio-RVC-Fork\Applio-RVC-Fork\infer-web.py", line 25, in <module>
import fairseq
ModuleNotFoundError: No module named 'fairseq'
i already try "pip install fairseq"```
lost bison
what is currently the best non-colab way to do covers?
graceful obsidian
I assume some paperspace or cpu-local stuff
alternatively if you're knowledgeable enough, obfuscate rvc aka, make your own fork and run it on saturn
graceful obsidian
don't you have at least gen 10 or so cpu?
lost bison
i dont even know what that means
graceful obsidian
you know, when you have say, intel i5 10400
more or less but anyway
if you feel like ur pc won't handle it, try whatever's recommended in guides over here
lost bison
i have an amd cpu
rare gobletBOT
@lost bison has reached level 1.
lost bison
i'll maybe try paperspace like you said
graceful obsidian
yea then that's problematic, don't bother locally then, same boat here
graceful obsidian
#1159513888199540817 try something from here
lost bison
ah wait says the free tier is pretty much useless :/
graceful obsidian
what's your cpu
do ctrl+shift+esc
go into performance tab
@lost bison
there you'll have your processor
proud elbow
i have a god awful pc tho
with decent 4-8 GB/more VRAM GPU it should be possible to do local inference & train. just don't do CPU/iGPU only.
graceful obsidian
you don't understand, dude's got amd
lost bison
doesn't how good your specs are affect the quality of the result tho?
graceful obsidian
back when I was still learning rvc, I tried it few times locally on bro's 1050ti, 1650 should handle it with ease
lost bison
hm ok I'll try
graceful obsidian
just you wouldn't go crazy with training sadly but for inference should be enough
lost bison
I don't plan on training
graceful obsidian
doesn't how good your specs are affect the quality of the result tho?
well no, not at all
the only thing that's about hardware is the vram
more vram ur gpu has, better models you can make, very lazily put
lost bison
I thought that was the case because Ive tried locally before but it just gave me the exact audio I put in...
graceful obsidian
as for inference, there's no difference aside of speed
you'd definitely want to use rmvpe to compensate for speed
if the output was 1:1 as input, there was some issue for sure
lost bison
I'm not to bothered by speed as long as the result is good
graceful obsidian
yeah then 1650 will be enough, even 4gb
what affects the quality is really the model and input's quality, as much as rmvpe handles poor or bad audio, it ain't magic
just a note ~
proud elbow
the f0 algorithm (rmvpe/mangio-crepe) you choose does only affect feature extraction time, not the training time
rare gobletBOT
@proud elbow has reached level 1.
lost bison
so it won't work if I sound nothing like the person I'm trying to sound like?
graceful obsidian
the f0 algorithm (rmvpe/mangio-crepe) you choose does only affect feature extrac...
the talk is about inference, not training models
dude only wants to make a cover
graceful obsidian
so it won't work if I sound nothing like the person I'm trying to sound like?
Look, the way rvc works is, it uses your voice, wav, or whatever you feed it - input
as instructions / blueprint
then it builds it all with the materials - model
Similar tone / voice can enhance poor made models that lack stuff but that's really all there is to it
rare gobletBOT
@graceful obsidian has reached level 4.
graceful obsidian
as long you set the pitch / octave properly, it'll work just fine, granted the model is good
aka, don't worry and go ahead, make covers
lost bison
okay, I'll give it another shot. thanks so much for the help.
rare gobletBOT
@lost bison has reached level 2.
faint bronze
when i press go web it shows the error message "the system cannot find the path specfied" what do i do
glossy grove
lmk
Yeah everything worked for me with no issue. I zipped it up last night but it's a big folder.
faint bronze
when i press go web it shows the error message "the system cannot find the path specfied" what do i do @whispers
rare gobletBOT
@faint bronze has reached level 2.
faint bronze
when i press go web it shows the error message "the system cannot find the path specfied" what do i do @whispers
when i press go web it shows the error message "the system cannot find the path specfied" what do i do
lost bison
jesus you're a bit pushy
graceful obsidian
when i press go web it shows the error message "the system cannot find the path ...
Control yourself if you don't want troubles.
People aren't obligated to be here n now on your call
A little bit of googling or github exploration wouldn't hurt ✨
but I guess lazy kids like you don't wanna bother.
Get your python ( and system paths ) sorted out, my guy
graceful obsidian
path to rvc should not contain any spaces in names / folder names
❎ asdasdasd/asdasd asdasd/sodigjsdgsdg sodfgjsodg sddfsdf/sdfsdfsfds/RVC
✅ asdsasdasdasd/asdasdasdasd/RVC
if you fixed your path and still encounter it then like I said, it's your python installations fault, or to be exact, paths
faint bronze
im now getting this error message anyone help? @graceful obsidian
graceful obsidian
do you even have python
faint bronze
yes i do @graceful obsidian
graceful obsidian
then that must mean you're not set the path correctly
you gotta add python to path variable
google it up, many forums or such to help you out with it
faint bronze
@graceful obsidian how? any examples
graceful obsidian
- you gonna learn something new
just google it " How to add python to path "
I am too busy rn to explain it
wooden lotus
Chat,one of my model was on the database but was not reuploaded to #1159289700490694666
Shoul I reupload it or sth
mossy pasture
do i only need to run these 2 cell for resuming training? i mean ofc the training cell too
proper shale
do i only need to run these 2 cell for resuming training? i mean ofc the trainin...
Yes, you do
No wait
Run all the others except
Feature extraction and preprocessing
wide parcel
Yeah everything worked for me with no issue. I zipped it up last night but it's ...
cool
if you don’t have somewhere to send it i can open a ftp server
or a torrent
vagrant marsh
Is there a way to enable the "train" option on this Huggingface space?
proper shale
Is there a way to enable the "train" option on this Huggingface space?
No, because it uses CPU only.
However, you could duplicate the space and rent a GPU... but that costs money!
vagrant marsh
However, you could duplicate the space and rent a GPU... but that costs money!
oh, okay
I got this from #1159949278270193734
and I guess all the rvc links/local files needs an strong gpu
but thanks anyway
proper shale
RVC Disconnected can train though
glossy grove
cool
if you don’t have somewhere to send it i can open a ftp server
I'll get you a share link in a bit
@wide parcelitll be later today.... It'll take awhile to upload. its 14 GB
lost bison
Is Replay basically a frontend for local rvc?
feral plover
Is there a way to enable the "train" option on this Huggingface space?
whats the link for this hugging face space?
vagrant marsh
whats the link for this hugging face space?
rare gobletBOT
@vagrant marsh has reached level 3.
feral plover
https://huggingface.co/spaces/Clebersla/RVC_V2_Huggingface_Version
thanks alot
grizzled cliff
how do i make it so i won't cut out so much
worthy relic
ive downlaoded rvc applio and did my model how i make the cover now?
wide parcel
<@386275650119401473>itll be later today.... It'll take awhile to upload. its 14...
dAMN
glossy grove
It has the pretrain models and what not in it
rare gobletBOT
@glossy grove has reached level 2.
shut axle
trained a model and didn't get an index file. only got .pth
how do i fix this??
wispy burrow
trained a model and didn't get an index file. only got .pth
how do i fix this??
'Train Feature Index'
grim sparrow
How to update Applio?
mossy pasture
will the discontinued v2 have auto save? like if the colab just disconnected, will the training be auto saved?
tender flume
is there anyone that can help me with the installation of everything? I can vc if needed to screenshare and stuff
honest cobalt
how do I decrease the number of threads
proper shale
will the discontinued v2 have auto save? like if the colab just disconnected, wi...
I believe not, because of the amount of resources that takes.
radiant laurel
Can someone help me aka walk me throught how to use and make a model using colab
crimson basalt
How to update Applio?
Run the applio manager then choose update applio
tribal oar
bro the explanation of the applio guide video
all i see is flash
idk how to install applio
crimson basalt
idk how to install applio
rugged needle
Traceback (most recent call last):
File "C:\Program Files\Python39\lib\multiprocessing\process.py", line 315, in _bootstrap
self.run()
File "C:\Program Files\Python39\lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "O:\RVC-Training\Retrieval-based-Voice-Conversion-WebUI\infer\modules\train\train.py", line 232, in run
net_g.module.load_state_dict(
File "O:\RVC-Training\Retrieval-based-Voice-Conversion-WebUI\venv\lib\site-packages\torch\nn\modules\module.py", line 2041, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for SynthesizerTrnMs768NSFsid:
size mismatch for dec.ups.0.weight_v: copying a param with shape torch.Size([512, 256, 24]) from checkpoint, the shape in current model is torch.Size([512, 256, 16]).
size mismatch for dec.ups.1.weight_v: copying a param with shape torch.Size([256, 128, 20]) from checkpoint, the shape in current model is torch.Size([256, 128, 16]).
``` I'm using the Original RVC with a venv tried making a model and this happened any ideas on how to fix?try toggling back and forth from v1 to v2
rugged needle
alr
rare gobletBOT
@rugged needle has reached level 5.
sand iris
Had the UI do some weird shit when I did that, it showed new settings despite being on v2 before and after the switch
sand iris
It worked ?
rugged needle
yeah
rugged needle
forgot how to run tensorboard how do I do that?
sand iris
forgot how to run tensorboard how do I do that?
tensorboard --logdir logs apparently
sand iris
make sure logs points to, well the logs folder within the rvc repo
and the shell is within the venv
topaz scaffold
is there any online rvc like google collab alterntive since its banned ?
rugged needle
is g loss total supposed to spike up to 300 randomly?
and just stay there?
oh crap INFO:PathfinderB:loss_disc=12.585, loss_gen=12.308, loss_fm=188.269,loss_mel=75.000, loss_kl=-0.333
Loss fm 💀
robust needle
That repo is deleted, but I'll create another one soon. Lmk if you need it earlier and I'll send it
honest pond
That repo is deleted, but I'll create another one soon. Lmk if you need it earli...
do you know how i can get my role helper again? they gave it to me and just that day they knocked down the channel, I entered this one, they gave it to me again and they took it away.
type was given to me after answering questions
vagrant marsh
Where I can test my model?
slow summit
That repo is deleted, but I'll create another one soon. Lmk if you need it earli...
Thank you, I won't rush you. I'd really apprichate if you could @ or DM me when u do though
if not thats fine too
robust needle
do you know how i can get my role helper again? they gave it to me and just that...
Hey, things are a bit messed since the deletion, but I'm sure that staff will be soon back on their usual routine, everyone is extra busy with restoring and recovering certain things. I can't directly help you about this since I'm just helper, but I'll ask around. Try to get in touch with some of the admins if you can in the meantime.
robust needle
Thank you, I won't rush you. I'd really apprichate if you could @ or DM me when ...
Alrighty
vagrant marsh
HELP PLS!!!!!!!!!!!!!
radiant laurel
how long does it take to applio to instill?
quaint tusk
Hey everyone, I'm trying to use the AICoverGen that doesn't use web UI, but it shows an error everytime I try to download a model, does anyone know why? it says it's not a zip file, but the link I'm putting in leads to a zip file
daring rune
is there a link to upload datasets from google drive link for text to speech colab? Thanks
heady trail
Just asking this. Is it normal for a specific model to not produce any sound at all? It happens with the Eagle voice model, and I don't know if it is just me or not. Running RVC through my brother's pc obviously. https://www.weights.gg/models/clm72xzkk13z1cctcc3bq00ah
Here is to hope that I just have to reinstall the model and it will be done.
languid moss
someone can give me the link for the new collab?:?
brittle wing
somebody please teach me how to use rvc text to speech
rare gobletBOT
@brittle wing has reached level 2.
brittle wing
@help
errant dagger
is there any online rvc like google collab alterntive since its banned ?
there was covergen but now its banned as well because google hates fun
zinc sphinx
i can't figure out this error with rvc lol. This FileNotFoundError: [Errno 2] No such file or directory: 'C:\Program/preprocess.log'. i looked it up someone had something similar someone suggested removing spaces from the path in training and to also add quotes before and after but this error still happens anyways
dire kayak
INFO:RodrigoV1:Train Epoch: 1 [0%]
INFO:RodrigoV1:[0, 0.0001]
INFO:RodrigoV1:[loss_disc=3.657] | [loss_gen=3.398] | [loss_fm=11.826] | [loss_mel=28.494] | [loss_kl=6.679]
DEBUG:matplotlib:matplotlib data path: C:\Users\jeter\OneDrive\Ambiente de Trabalho\Applio-RVC-Fork\runtime\lib\site-packages\matplotlib\mpl-data
DEBUG:matplotlib:CONFIGDIR=C:\Users\jeter.matplotlib
DEBUG:matplotlib:interactive is False
DEBUG:matplotlib:platform is win32
INFO:torch.nn.parallel.distributed:Reducer buckets have been rebuilt in this iteration.
when training a model
it´s been like this for a while
errant dagger
whats your gpu?
errant dagger
yeah I don't think it should take that long so something is not right
however I can't provide more info since I train online
dire kayak
however I can't provide more info since I train online
how do u do it?
errant dagger
I use voicemy.ai
dire kayak
i was lookin online but only found local
dire kayak
I use voicemy.ai
is it free?
errant dagger
is it free?
yes
rare gobletBOT
@dire kayak has reached level 1.
dire kayak
thanks a lot @errant dagger
zinc sphinx
rip
brittle wing
why do i always have to choose a basic voice text to speech
dire kayak
@errant dagger how much time does it take for medium quality?
shut axle
i keep getting this error saying
ValueError: need at least one array to concatenate
can someone explain in simple terms on how to fix it??
brittle wing
Has anyone else encountered this when trying to make the Epochs?
Collab used (9/17)
light horizon
Hey people. Very new to this tool and am hoping someone could help me out in getting it set up. I'm seeing some different tools / installers / 'forks' etc in different guides here and it's a bit daunting as someone who has no clue what each one does and how they're different from one another, so would someone be able to talk me through how to get started plz? (I'd like to use others' models ofc, but am primarily interested in training my own voice model.)
zinc sphinx
anyone know much about rvc here
stone dawn
hiii applio says that nvidia needed support
what should i do if i only have the intel card...
compact wraith
Has anyone else encountered this when trying to make the Epochs?
Collab used (9...
I think Collab doesnt really work, since there is different versions,
rare gobletBOT
@compact wraith has reached level 6.
compact wraith
what are you currently using?
remote grove
I'm AMD, but I'm using hf or colab. applo is broken and doesn't work, so it's better not to try it.
lavish eagle
Why does it say Broken on the AMD installer
lime bough
hmm im getting this error while running the webui
E:\git\rvc2\Retrieval-based-Voice-Conversion-WebUI>runtime\python.exe infer-web.py --pycmd runtime\python.exe --port 7897
Traceback (most recent call last):
File "E:\git\rvc2\Retrieval-based-Voice-Conversion-WebUI\infer-web.py", line 28, in <module>
import gradio as gr
File "C:\Python39\lib\site-packages\gradio\__init__.py", line 3, in <module>
import gradio.components as components
File "C:\Python39\lib\site-packages\gradio\components.py", line 33, in <module>
from ffmpy import FFmpeg
ImportError: cannot import name 'FFmpeg' from 'ffmpy' (C:\Python39\lib\site-packages\ffmpy\__init__.py)
i've ffmpeg installed & added to PATH
also installed ffmpeg-python but still no luck
prime hemlock
hmm im getting this error while running the webui
```
E:\git\rvc2\Retrieval-base...
the audio file is wrong
rare gobletBOT
@prime hemlock has reached level 6.
prime hemlock
try changing the file type
ember lance
Anyone know if any of these methods of training can be used with Microsoft azure text to speech to make a custom voice that sounds good?
brittle wing
I think Collab doesnt really work, since there is different versions,
I have RVC fork now and it can't process my dataset when I'm using the path that it's in
I don't know if this matters but I'm using Windows 11
compact wraith
I have RVC fork now and it can't process my dataset when I'm using the path that...
Im not sure, because is better off to use Mangio RVC Fork.
since its basically automatic, it will read the dataset, and auto adds into the process
compact wraith
If you are training a model, your sample rate has to be the same number for your mute wave file
i had that error like 2 times, and figured it out it was the path file i should say
how much sample rates you are doing?
safe whale
Anyone have Mangio and Applio both installed? I've been using Mangio, but wanted to try out Applio. I just don't want to destroy my Mangio install.
rare gobletBOT
@safe whale has reached level 1.
manic marsh
might be a dumb question but how do i get the non demo version?
compact wraith
Anyone have Mangio and Applio both installed? I've been using Mangio, but wanted...
if you have both on separate folders, then i dont think it wont affect anything, unless they work the same, its going to use the files
bold ermine
Greetings! Guys, I am getting ffmpeg error in process data section , I even downgraded ffmpeg version and even tried different extension of audio file like mp3 wav but nothing worked, plz help me with this issue 🙏
FileNotFoundError: [Errno 2] No such file or directory: '/content/Retrieval-based-Voice-Conversion-WebUI/logs/mi-test/3_feature256'
The project is creating the file 3_feature256 in mute folder instead of mi-test folder
compact wraith
if both support RVC v2, then your fine
remote cosmos
Is there anyone who can extract the vocal ad-lib for nmixx's love me like this?
serene musk
Anyone have Mangio and Applio both installed? I've been using Mangio, but wanted...
I have both and both work applio just takes long to open idk why it didn't take long the first few times
bold ermine
does anyone ever created rvc api? so we dont have to visit gradio for using rvc
rare gobletBOT
@bold ermine has reached level 1.
frigid wasp
Is anyone else struggling with RVC on HuggingFace?
On my end, it endlessly generates
serene musk
Yeah it'd be dope to have a rvc program that doesn't need chrome to use a ui
vague sedge
So uhhhh....
check your python version
covert schooner
🤦♂️ I swear I'm not as big of an idiot as I come across but...yeah I downloaded the wrong version of python 🤣
vague sedge
^^
in my video description you get all the links ^^
so you would have the right python :D
covert schooner
Yeah I just figured the newest version would work better. Egg on my face I guess 🤣
soft umbra
so i'm using the google colab for training, it's telling me that it's saving some little models but where do i download them?
INFO:rose_speaking:Saving model and optimizer state at epoch 240 to ./logs/rose_speaking/G_2333333.pth
my model is done training on colab where do i find it?
lean owl
i wanna make a voice model out of this and im just wondering how much epochs would i need
(WARNING LOUD)
burnt agate
i wanna make a voice model out of this and im just wondering how much epochs wou...
my ears..... please remove that or add a warning at least 
lean owl
my ears..... please remove that or add a warning at least <:skullsob:11593725319...
oops sorry
but anyway
how many epochs would u need for a 3 second audio file
i know it sounds ridiculous
burnt agate
i know it sounds ridiculous
true lol but there isnt really a magic number for epochs, it mostly just depends on when the model starts overtraining imo
lean owl
oh so its just trial and error?
burnt agate
every model is different, just have to see what it does
burnt agate
I usually start at 200-300 but thats with a few minutes of dataset, cant really say much about super short ones
lean owl
I usually start at 200-300 but thats with a few minutes of dataset, cant really ...
i guess it would have to be longer
burnt agate
can someone help me make a ai cover for a song?
if you just plan on making covers id recommend using replay if you have a decent gpu
burnt agate
i guess it would have to be longer
im not 100% sure tbh, last model I did had a 17 second dataset and it started overtraining at 50 epochs but take that with a grain of salt because it was a meme model lol
lean owl
im not 100% sure tbh, last model I did had a 17 second dataset and it started ov...
i guess you really have to experiment with stuff like that
worldly oasis
Hi. Im simply traying to create AI cover. I have trained model
I select the model -> upload a song -> hit convert -> get this error after few seconds.
Tried restarting the whole RVC app. Didnt help
burnt agate
i guess you really have to experiment with stuff like that
trail and error while following the documentation is the way to go imo
lean owl
im only realising because i trained everything at like 250-300 and wondering why certain audio files sound so bad
burnt agate
Hi. Im simply traying to create AI cover. I have trained model
I select the mode...
you tried using a different audio file or a different audio format?
burnt agate
give it a try, that is one of the very random errors people dont know much about from what ive seen
worldly oasis
Sometimes it works when I try hard enough and press converrt enough times 😄
but not this time
burnt agate
im only realising because i trained everything at like 250-300 and wondering why...
tensorboard is your best friend and enemy when it comes to getting the best quality out of your model lol
worldly oasis
It just worked when I selected last to second saved epoch
lean owl
tensorboard is your best friend and enemy when it comes to getting the best qual...
it either gives out the best results or just horrible results and it doesnt even sound like the original
burnt agate
It just worked when I selected last to second saved epoch
hmm care to elaborate?
burnt agate
it either gives out the best results or just horrible results and it doesnt even...
yeah it happens, sometimes you have to make changes to your dataset and retrain from my experience
worldly oasis
hmm care to elaborate?
I was using model from epoch 500. name_e500_s80000.pth this was giving me error
When I switched to file name_e490_s78400.pth it worked
burnt agate
I was using model from epoch 500. `name_e500_s80000.pth` this was giving me erro...
ooh interesting, will keep that in mind thanks 
proper shale
I was using model from epoch 500. `name_e500_s80000.pth` this was giving me erro...
Normally that error is just because of not properly selecting stuff
worldly oasis
ooh interesting, will keep that in mind thanks <:aismug:1159365471368400948>
Then switched back to e500 and also worked. But when I changed algorithm from rmvpe+ to magio-crepe it crashed again. Cant see pattern with this behaviour so far.
rare gobletBOT
@worldly oasis has reached level 3.
burnt agate
Normally that error is just because of not properly selecting stuff
last time I had that error was on applio and I am 100% sure everything was selected, this error is still a mystery to me
lean owl
whats better magio crepe or rmvpe
soft umbra
rmvpe
rare gobletBOT
@soft umbra has reached level 1.
burnt agate
whats better magio crepe or rmvpe
mangio crepe is good if you have a really good dataset but rmvpe is just more robust overall imo
proper shale
Then switched back to e500 and also worked. But when I changed algorithm from rm...
Yeah reselecting works
lean owl
how do i use the rvc disconnected colab
lean owl
i think mangio crepe is older
lean owl
mangio crepe is good if you have a really good dataset but rmvpe is just more ro...
yea ive had better results with rmvpe
even when voice cloning
burnt agate
rmvpe just gives better results most of the time imo
wide ingot
I experimented with replacing hubert_base.pt with the ContentVec weights but keeping the file name. It still seemed to work apparently.
burnt agate
I experimented with replacing hubert_base.pt with the ContentVec weights but kee...
any improvements?
havent really checked out that yet tbh
wide ingot
I mean, it should be better objectively. For me though, I hear small improvements over regular HuBERT. It's often said that ContentVec is an improved HuBERT.
proper shale
In everything
It should be noted that it your dataset has a very low amount of noise, mangio-crepe leaves less artifacts at lower hop-sizes, and is therefore higher quality.
...but at that point you'd be wasting minutes for feature extraction
plain sparrow
What can I do if I get the error that I don't have enough GPU memory?
plain sparrow
My GPU is 7.9 GB
spare karma
guys is there really nothing like colab i have amd card i can't do it locally.
plain sparrow
My GPU is 7.9 GB
It has 400 epochos. Idk how to find the batchsize tho
rare gobletBOT
@plain sparrow has reached level 1.
proper shale
guys is there really nothing like colab i have amd card i can't do it locally.
#1159949278270193734 , you can still train
proper shale
It has 400 epochos. Idk how to find the batchsize tho
Check the training log
brittle wing
Only I installed google colab rvc v2 pls
rare gobletBOT
@brittle wing has reached level 1.
plain sparrow
Check the training log
Where can I find the training log?
proper shale
So you need a Colab?
plain sparrow
train.log
I used this model:
https://www.weights.gg/models/clm73dfn126c2cctcu44d3d49
proper shale
Oh, so you weren't training, I see
plain sparrow
Yes exactly
proper shale
Try restarting the fork if you haven't yet
And close other apps that could be using VRAM
brittle wing
So you need a Colab?
Yes, Show me how to use it
plain sparrow
And close other apps that could be using VRAM
It also tells me that I'm out of CUDA memory
proper shale
Yes, Show me how to use it
Ehhh I don't know if I got that much free time on my hands