I am not ranting ,I can usually figure these things out ,even on the other site, sometimes it takes a while but I can figure them out
I have zero programming knowledge ,but I get there .This is the first time I have run into a situation where nothing works
The latest from Colab
/content/MVSEP-MDX23-Colab_v2
GPU use: 0
started!
usage: inference.py [-h] --input_audio INPUT_AUDIO [INPUT_AUDIO ...] --output_folder OUTPUT_FOLDER
[--cpu] [--overlap_demucs OVERLAP_DEMUCS] [--overlap_VOCFT OVERLAP_VOCFT]
[--overlap_VitLarge OVERLAP_VITLARGE] [--overlap_InstVoc OVERLAP_INSTVOC]
[--weight_InstVoc WEIGHT_INSTVOC] [--weight_VOCFT WEIGHT_VOCFT]
[--weight_VitLarge WEIGHT_VITLARGE] [--single_onnx] [--large_gpu]
[--BigShifts BIGSHIFTS] [--vocals_only VOCALS_ONLY] [--use_VOCFT USE_VOCFT]
[--output_format OUTPUT_FORMAT]
inference.py: error: argument --input_audio/-i: expected at least one argument

I tried to install your run.bat on its own ,just opens in notepad ,ran it as administrator ,black screen pop up then disappears
tried to run it in Python 3.12 always some error ,tried to run it in Colab always some error
I have run a lot of, lets say different software on my machine ,if I do not trust it I make a complete backup first ,any time the install was
doing something else ,,it would ether not run or seem to run a little messed up and infect my machine ,so you can understand my
suspicion when nothing worked
Now as to this mentality, this chastising inference, that permeates this site, I am just trying to get your program to work
I appreciates the work you put into it ,I could not do it . I wish programmers after writing a program would go and get someone living
In a refrigerator box, sit them down in front of a computer and say, run this program ,does it make sense to you ,can you go thru it in a
understandable order, if they can then your program is a success. Programmers think like Programmers and Common sense is anything
but common, If you have ever tried to teach anything to anybody you will learn that quickly ,,Common Sense, biggest load of BS ever
pushed ,just like all the other BS we have to live thru
Really wanted to try how good your work is ,,can not get it,, to work anywhere

 

How do you get that at UVR MDX23C-InstVox HQ 2? All I see is MDX23C-InstVox HQ 1.

 

You can find it in Download Center on the list of models.

 

Already figured out - there you need to enter the VIP code for additional models to appear.

 

There's no 'run.bat' in my fork:




-----------------
"inference.py: error: argument --input_audio/-i: expected at least one argument"

means you have not provided any audio input file to be separated by the script, so of course, the script can't work.


PS: I'm not a programmer, it's not a business. I'm just a musician creating, maintaining nd updating all these musical-related colabs adaptations on my free time, for free, to be used for free, by anybody, with public source code and models. Thousands of people are using the Colab notebook without any problem and are getting their separated audio with a bunch of clicks. So, once again, stop your stupid rants against other people just because you make mistakes. You are also free to not use it if you don't like it.

 

Last edited by a moderator: Mar 19, 2024

A new UVR beta has been released in Audio Separation discord with new models based on ByteDance's algo which was reproduced from scratched an open-sourced recently.

Highest SDR score we've ever seen for vocals/instrumental stem.

These models will be implemented in few days in upcoming update of my colab fork, with some other optimisations.

They are also already available on mvsep.com & x-minus.pro (UVR-Online)

 

Can you share the link?

 

https://colab.research.google.com/g...23-Colab_v2/blob/v2.4/MVSep-MDX23-Colab.ipynb
https://mvsep.com/
https://x-minus.pro/

https://github.com/TRvlvr/model_rep...UVR_Patch_3_29_24_5_11_BETA_full_roformer.exe

In UVR use overlap 2 to alleviate some issues with clicking (increases separation time).
OpenCL is not functional for BS-Roformer yet (so AMD/Intel GPUs acceleration).

For more info, check the last entries in news section of:
https://docs.google.com/document/d/17fjNvJzj8ZGSer7c7OFe_CNfUKbAxEh_OBv94ZdRG5c/

 

Is there any way to attach mel-roformer to uvr?

 

Hey guys, sorry for probable off-topic:

What is the best voice conversion AI currently? By that I mean models which you feed with some specific voice tracks, and after that it can convert some other voice into that first one

I've heard really good musical implementations of that, is there any good guide on it? Thanks!

 

@GeekedGlitch Replay is free

 

Last edited: Jun 22, 2024

RVC is surely the best Modeler at the moment.
>> Can anyone point me to a Colab where I can Train my own models with RVC and then apply them to target audio?
I dont own an NVIDIA GPU so will need online processing if there is a working Colab.

 

Jarredou released a new Drumsep model trained by Aufr33 (x-minus author/UVR co-author) based on MDX23C arch and new, better dataset. Much better SDR.

Also, I've gathered some information of the process from previous posts on our Discord for more insights:

https://docs.google.com/document/d/...bAxEh_OBv94ZdRG5c/edit#heading=h.2u19k7ty9b00

 

Last edited: Jul 4, 2024

Hey Deton I checked out ur google doc, couldnt find too much on separating electric guitar solos from backing electric guitar, i know its kind of a niche use case but any pointers here? thanks!

 

Is it hard to train a model based on orchestra parts, such as different winds, different strings, choir, etc.?

Probably will be much harder ,since in the orchestra recording director often uses main microphone pair (picking up the whole orchestra playing) not only as a reference, but also as a main source. So obtaining high quality separate stems (even of groups of instruments) without leakages would probably prove itself impossible.

 

What about Intel OpenVINO AI for Audacity? Did one use it already?

 

you could train on recordings of individual sections or soloist.

 

Is there a Colab with full Demucs4 functionality, ie --shifts, --overlap, etc sliders? That also can select the various Demucs4 models ie, htdemucs_ft, demucs_mmi, demucs_6s, etc?

Need fine control over settings so UVR online and MVSEP.com arent gonna cut it.
Big thanks to anyone who can point me to one.
@Jarredou maybe you could add this to your master collection of models Colab if youre reading this

EDIT
Found One!
https://colab.research.google.com/drive/1EgVUKYUtbCpeA1VP1zt-6_lgIA2jRHU-?usp=sharing

 

Last edited: Nov 26, 2024

It is better. Ripx just has additional functionality, but the stem separation is better in UVR. I have compared the two with various songs.

 

I'm trying to pull out a clean stem of drums from a variety of rock songs. Which combination of models would be my best bet to use in ensemble mode? I have downloaded the VIP models. I realize there will be a need for some experimentation, but what's a good starting point for a combination of models to yield the best separation of a drum loop?