What's new
By:
Filters
LiteRECORDS

Register a free account today to become a member! Once signed in, you'll be able to participate on this site by adding your own topics and posts, as well as connect with other members through your own private inbox!

  • Guest, before your account can be reviewed you must click the activation link sent to your email account. Please ensure you check your junk folders.
    If you do not see the link after 24 hours please open a support ticket.

Powerful New Vocal Remover AI - Instructions

rkeane said:
compared to other programs and based against the earlier version..........v2 is smoking
only draw back is you defo need a top notch PC to get things done fast......
if this is based on 300 odd pairs...........the 1000 pair edition that Anjok is maybe gonna release is gonna be immense
think i maybe need to rob a bank for a new pc
Click to expand...

I'm going to attempt to train a 1000 pair model with the same amount of layers so you guys won't have to worry about PC drawbacks. With that being said, if the performance doesn't improve (or improves only marginally) I may have to add more layers to the model which will make it less friendly to older/slower PC's.

I'll do my best though!
 
NewAgeRipper said:
Personally I'd like to see a GUI with anything current and then a way to simply add the updates later by either certain file types or a simple update command tied to the git-hub.
Click to expand...

I'll still be working on it concurrently with the models. However, I won't be dedicating as much time to it until I get the models to perform their best. Once I have something decent mocked up, I'll throw it on Git-hub if we have any coders here who want to contribute!
 
Anjok said:
I'm going to attempt to train a 1000 pair model with the same amount of layers so you guys won't have to worry about PC drawbacks. With that being said, if the performance doesn't improve (or improves only marginally) I may have to add more layers to the model which will make it less friendly to older/slower PC's.

I'll do my best though!
Click to expand...

lol, my laptop will explode when you release the 1000 pair version. But hey! You're the only one who still works in improving the AI. Everybody else just release the first version and then, nothing. Thank you so much for you work!
 
hahaha, I say bring on the beast xD Actually is kinda exciting that you are upgrading your PC.
Mileage may vary on this but I have had great success under-clocking my GPU to manage heat. ...
Not that I absolutely need to since I'm running a GTX 1070, but I've always got MSI Afterburner running.

For example, one of the previous models had me at 80-85C,
so I under-clocked from 1700MHz to 1200MHz Core and from 4,000 to 3,500MHz memory.
Presto, cool and comfy 68-72C.

I do the same when I'm gaming. I can still run Witcher 3 on my laptop at 60FPS under-clocked and hold a steady 70C, rather than the 85C it would otherwise run the game at.
 
ChrisCall said:
hahaha, I say bring on the beast xD Actually is kinda exciting that you are upgrading your PC.
Mileage may vary on this but I have had great success under-clocking my GPU to manage heat. ...
Not that I absolutely need to since I'm running a GTX 1070, but I've always got MSI Afterburner running.

For example, one of the previous models had me at 80-85C,
so I under-clocked from 1700MHz to 1200MHz Core and from 4,000 to 3,500MHz memory.
Presto, cool and comfy 68-72C.

I do the same when I'm gaming. I can still run Witcher 3 on my laptop at 60FPS under-clocked and hold a steady 70C, rather than the 85C it would otherwise run the game at.
Click to expand...

My current PC is pretty tough and was definitely getting the job done, as I managed to keep it under 98F CPU and my GPU under 86F during some intense training sessions. The issue is it's 5 years old and my MB is capped on the amount of RAM I can install (32GB). I also can't upgrade my CPU or GPU to what I really need. I'm pretty sure if I kept training on it, the lifespan was going to decrease a lot quicker than if I had hardware optimized for deep learning.

My new set up will have 64GB's of RAM (expandable to 128GB), a RTX 2080Ti with 11GB's of V-RAM, and a far stronger processor. Definitely going to be spitting out more models in a shorter time frame, and save time determining the quality of my datasets. It takes a least 15-20 epoch runs to know if my dataset is effective. My new PC will get there in maybe 3 or 4 hours with a dataset consisting of 320 pairs. With my current PC using my GTX 1060 GPU, it takes a little over 24 hours to get to 15-20 epochs with a 320 pair dataset...

I'd rather use my current PC to code the GUI, analyze models, and run test conversions, which take 15 seconds per track. All while my new one trains.

I'm pretty excited about this too!!
 
halofan253 said:
lol, my laptop will explode when you release the 1000 pair version. But hey! You're the only one who still works in improving the AI. Everybody else just release the first version and then, nothing. Thank you so much for you work!
Click to expand...

You're welcome! This has been a pretty fun project to say the least. I'm thrilled that I can use my technical skills and love of music to make this happen.
 
I have made a fascinating discovery!

This AI converts early Death tracks SUBSTANTIALLY better than the primary AI in the other topic. It's no contest, it just crushes it in every way imaginable. Take a look at Zombie Ritual from Scream Bloody Gore. Convert it with each and see side by side if you like.

This does appear to be the exception rather than the rule, but there are certainly places where this AI (or this current model of this AI) does a superior job.

Another great example to look at is The Cure - One Hundred Years
The other AI does a slightly better job overall at stripping the vocals, but this AI model does not pull the drum echo into the vocal track, making it superior in that regard. I also found that Black Sabbath - Symptom of the Universe is stronger up front (though the back half of the song is a mess) ... another heavy reverb song ...

Scream Bloody Gore is also a reverb heavy album...

I don't want to jump to conclusions because it may be more complicated than reverb, but it certainly sounds like this model is far superior at handling reverb over the other one. I don't know for sure, but identifying why in particular, Scream Bloody Gore is such a crushing difference ... would be interesting.
 
Mock-Up.jpg

Here is a mock-up of the GUI I'm working on for this AI. I'm still working on coding and configuring it. More features and options will be added! This is my first draft of the application, so the final product will look better than this.

I'll also be releasing a GUI for training too, but that will be way after the conversion GUI is released.
 
Last edited:
ChrisCall said:
I have made a fascinating discovery!

This AI converts early Death tracks SUBSTANTIALLY better than the primary AI in the other topic. It's no contest, it just crushes it in every way imaginable. Take a look at Zombie Ritual from Scream Bloody Gore. Convert it with each and see side by side if you like.

This does appear to be the exception rather than the rule, but there are certainly places where this AI (or this current model of this AI) does a superior job.

Another great example to look at is The Cure - One Hundred Years
The other AI does a slightly better job overall at stripping the vocals, but this AI model does not pull the drum echo into the vocal track, making it superior in that regard. I also found that Black Sabbath - Symptom of the Universe is stronger up front (though the back half of the song is a mess) ... another heavy reverb song ...

Scream Bloody Gore is also a reverb heavy album...

I don't want to jump to conclusions because it may be more complicated than reverb, but it certainly sounds like this model is far superior at handling reverb over the other one. I don't know for sure, but identifying why in particular, Scream Bloody Gore is such a crushing difference ... would be interesting.
Click to expand...

That's really good to know! I've noticed this model is far better at removing reverbs as well. Funny thing is this dataset actually included a lot of pop and dance music (courtesy of some users here who sent me some great pairs). It makes me even more excited to get the rest of my PC parts next week!
 
Anjok said:
View attachment 642

Here is a mock-up of the GUI I'm working on for this AI. I'm still working on coding and configuring it. More features and options will be added! This is my first draft of the application, so the final product will look better than this.

I'll also be releasing a GUI for training too, but that will be way after the conversion GUI is released.
Click to expand...

if u need a beta tester send it here
 
I can not get this work on Windows 10 64bit or Windows 8.1 64bit and would like to know what the possible issues are. I've followed the instructions and still get this error.......
'pip' is not recognized as an internal or external command,
operable program or batch file.

Can you please list instructions for how to use this on Linux/Ubuntu?
Thanks.
 
NewAgeRipper said:
I can not get this work on Windows 10 64bit or Windows 8.1 64bit and would like to know what the possible issues are. I've followed the instructions and still get this error.......
'pip' is not recognized as an internal or external command,
operable program or batch file.

Can you please list instructions for how to use this on Linux/Ubuntu?
Thanks.
Click to expand...

This should fix the "pip" issue. https://www.youtube.com/watch?v=zYdHr-LxsJ0
 
Not sure what this is about but after trying to convert a song it gives this and then just sits on a blinking cursor.

C:\Users\zensh\Documents\vocal-removerV2>python inference.py --input Made Of Stone.mp3
C:\Program Files\Python37\lib\site-packages\librosa\util\decorators.py:9: NumbaDeprecationWarning: An import was requested from a module that has moved location.
Import requested from: 'numba.decorators', please update to use 'numba.core.decorators' or pin to Numba version 0.48.0. This alias will not be present in Numba version 0.50.0.
from numba.decorators import jit as optional_jit
C:\Program Files\Python37\lib\site-packages\librosa\util\decorators.py:9: NumbaDeprecationWarning: An import was requested from a module that has moved location.
Import of 'jit' requested from: 'numba.decorators', please update to use 'numba.core.decorators' or pin to Numba version 0.48.0. This alias will not be present in Numba version 0.50.0.
from numba.decorators import jit as optional_jit
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

And then I try running the pip install command again just to be sure and displays I'm still missing files.

C:\Users\zensh\Documents\vocal-removerV2>pip install torch==1.3.0 torchvision==0.4.0 -f https://download.pytorch.org/whl/torch_stable.html
Defaulting to user installation because normal site-packages is not writeable
Looking in links: https://download.pytorch.org/whl/torch_stable.html
Requirement already satisfied: torch==1.3.0 in c:\program files\python37\lib\site-packages (1.3.0+cu92)
Requirement already satisfied: torchvision==0.4.0 in c:\program files\python37\lib\site-packages (0.4.0+cu92)
Requirement already satisfied: numpy in c:\program files\python37\lib\site-packages (from torch==1.3.0) (1.18.4)
Requirement already satisfied: pillow>=4.1.1 in c:\program files\python37\lib\site-packages (from torchvision==0.4.0) (7.1.2)
Requirement already satisfied: six in c:\program files\python37\lib\site-packages (from torchvision==0.4.0) (1.14.0)
Could not build wheels for torch, since package 'wheel' is not installed.
Could not build wheels for torchvision, since package 'wheel' is not installed.
Could not build wheels for numpy, since package 'wheel' is not installed.
Could not build wheels for pillow, since package 'wheel' is not installed.
Could not build wheels for six, since package 'wheel' is not installed.

C:\Users\zensh\Documents\vocal-removerV2>
 
Last edited:
NewAgeRipper said:
Not sure what this is about but after trying to convert a song it gives this and then just sits on a blinking cursor.

C:\Users\zensh\Documents\vocal-removerV2>python inference.py --input Made Of Stone.mp3
C:\Program Files\Python37\lib\site-packages\librosa\util\decorators.py:9: NumbaDeprecationWarning: An import was requested from a module that has moved location.
Import requested from: 'numba.decorators', please update to use 'numba.core.decorators' or pin to Numba version 0.48.0. This alias will not be present in Numba version 0.50.0.
from numba.decorators import jit as optional_jit
C:\Program Files\Python37\lib\site-packages\librosa\util\decorators.py:9: NumbaDeprecationWarning: An import was requested from a module that has moved location.
Import of 'jit' requested from: 'numba.decorators', please update to use 'numba.core.decorators' or pin to Numba version 0.48.0. This alias will not be present in Numba version 0.50.0.
from numba.decorators import jit as optional_jit
Click to expand...

djtayz said:
Happened to me too. I wrote pip3 install numba==0.48.0 (or change it to 0.50.0) and I think that fixed it for me .
Click to expand...

djtayz's suggestion worked for me when I tried it. Maybe it'll help you as well.
I should also mention that for whatever reason, for me anyway, MP3 files don't work, but wav or flac do.
 
ChrisCall said:
djtayz's suggestion worked for me when I tried it. Maybe it'll help you as well.
I should also mention that for whatever reason, for me anyway, MP3 files don't work, but wav or flac do.
Click to expand...

OK so exactly how do you fix it? Sorry I'm still not sure how to apply the change.
 
ChrisCall said:
djtayz's suggestion worked for me when I tried it. Maybe it'll help you as well.
I should also mention that for whatever reason, for me anyway, MP3 files don't work, but wav or flac do.
Click to expand...

I haven't tried with MP3s but sometimes it glitches with FLAC/WAV too, because of the naming. So instead of the original file name "Artist - Song Name" and etc. I simplify it to one word only. Maybe try that but I'm not sure if it would work for you.
 
You must log in or register to reply here.