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

djtayz said:
I'll make a separate thread for this, thanks for bringing it to my attention.
Click to expand...

I hate to be annoying but Google Colab isn't working right for the instrumental conversion.
It worked great yesterday.

Step 2
cp: cannot stat '/content/drive/My Drive/vocal-remover/': No such file or directory
Reading package lists... Done
Building dependency tree
Reading state information... Done
ffmpeg is already the newest version (7:3.4.6-0ubuntu0.18.04.1).
0 upgraded, 0 newly installed, 0 to remove and 32 not upgraded.

Step 4
%cd /content/drive/My Drive/vocal-remover
[Errno 2] No such file or directory: '/content/drive/My Drive/vocal-remover'
/content

Step 5
!python inference.py --input "/content/drive/My Drive/Clocks.flac"
python3: can't open file 'inference.py': [Errno 2] No such file or directory

The only way I solved it was to redo everything.
 
Last edited:
NewAgeRipper said:
I hate to be annoying but Google Colab isn't working right for the instrumental conversion.
It worked great yesterday.

Step 2
cp: cannot stat '/content/drive/My Drive/vocal-remover/': No such file or directory
Reading package lists... Done
Building dependency tree
Reading state information... Done
ffmpeg is already the newest version (7:3.4.6-0ubuntu0.18.04.1).
0 upgraded, 0 newly installed, 0 to remove and 32 not upgraded.

Step 4
%cd /content/drive/My Drive/vocal-remover
[Errno 2] No such file or directory: '/content/drive/My Drive/vocal-remover'
/content

Step 5
!python inference.py --input "/content/drive/My Drive/Clocks.flac"
python3: can't open file 'inference.py': [Errno 2] No such file or directory

The only way I solved it was to redo everything.
Click to expand...

That's the way Colab works. If you don't use it for some time or it's been more than 10-15 minutes since you closed the webpage, your assigned virtual PC is no longer attached to you and you have to connect once again. Just play all steps again. That's a downside, I know, but nothing I can do about it.
 
djtayz said:
That's the way Colab works. If you don't use it for some time or it's been more than 10-15 minutes since you closed the webpage, your assigned virtual PC is no longer attached to you and you have to connect once again. Just play all steps again. That's a downside, I know, but nothing I can do about it.
Click to expand...
It's all good. All good.
 
***UPDATE***

1. I'll be releasing an update to GUI this week. It's going to include new options and a new stacked model (a stacked model is one trained on converted data). The stacked model will help clean up vocal residue left by the AI in certain tracks. This particular model also comes with special options that I will explain in more detail upon release.

2. Unfortunately, a compiled exe version of the AI has been canceled due PyTorch & Cuda incompatibility with existing compilers. I brought in some brilliant minds to help me with it and it's been determined that it will not be an option at this time. Big requests are actually in progress with the PyTorch development team on GitHub as it's on them to fix the compilability & compatibility issues. As soon the bug is fixed, I will release the compiled version. If anyone here knows anyone that would like to take a stab at it, please go for it! My code is open source and free for anyone to use or change how they see fit.

Luckily, most of the installation issues have already been ironed out. As long as the instructions are carefully followed in the main post, there shouldn't be any issues running the AI. Also, I encourage those having issues running the AI due simply to poor hardware to please try the Google Colab instructions provided. (Special thanks to djtaz for hooking that up!)
 
Last edited:
Anjok said:
***UPDATE***

1. I'll be releasing an update to GUI this week. It's going to include new options and a new stacked model (a stacked model is one trained on converted data). The stacked model will help clean up vocal residue left by the AI in certain tracks. This particular model also comes with special options that I will explain more detail upon release.

2. Unfortunately, a compiled exe version of the AI has been canceled due PyTorch & Cuda incompatibility with existing compilers. I brought in some brilliant minds to help me with it and it's been determined that it will not be an option at this time. Big requests are actually in progress with the PyTorch development team on GitHub as it's on them to fix the compilability & compatibility issues. As soon the bug is fixed, I will release the compiled version. If anyone here knows anyone that would like to take a stab at it, please go for it! My code is open source and free for anyone to use or change how they see fit.

Luckily, most of the installation issues have already been ironed out. As long as the instructions are carefully followed in the main post, there shouldn't be any issues running the AI. Also, I encourage those having issues running the AI due simply to poor hardware to please try the Google Colab instructions provided. (Special thanks to djtaz for hooking that up!)
Click to expand...

Absolutely. Will djtayz implement the new models as well or that's not possible to do on Google Colab?
 
NewAgeRipper said:
Absolutely. Will djtayz implement the new models as well or that's not possible to do on Google Colab?
Click to expand...

Technically, yes. However, the stacked model requires multiple conversions of a single song. If he can get the GUI to work on Colab, then definitely. While stack looping (the process of running a single track through the stacked model multiple times until the vocal pinches are gone) can be done manually, doing it via the command line would be a HUGE pain in the ass. I'm implementing a stack looping option to the GUI to make the process far less painless. The only drawback will be waiting hours for one track to complete for those with dated hardware.

I'll explain all of this in more detail with the next update.
 
Last edited:
Anjok said:
Technically, yes. However, the stacked model requires multiple conversions of a single song. If he can get the GUI to work on Colab, then definitely. While stack looping (the process of running a single track through the stacked model multiple times until the vocal pinches are gone) can be done manually, doing it on command line will be a real pain. I'm implementing a stack looping option to the GUI to make the process far less painless. The only drawback will be waiting hours for one track to complete for those with dated hardware.

I'll explain all of this in more detail with the next update.
Click to expand...

Well, I'll put it like this. The command line on my laptop was taking 5 to 7 hours pending the track length. The GUI took 45 min to whatever so that was a major time increase for my end. I'll deal with the triple time if it really means me needing the echo gone. But for the most part, my DIY karaoke tracks as long as you have a singer who is on point, one won't notice the echo or other vocal residue anyways. And if so it's more like an effect than anything to enhance the singing.
 
NewAgeRipper said:
[MENTION=39673]Anjok[/MENTION] link to source files please? I know someone who wants to take a stand at it.
Click to expand...

The source files are posted in the main post. Just have them install the AI as normal by following the instructions posted and go from there.
 
I will say that I'm running a 1070 card and running conversions in 10 seconds.

I'm not sure how immense the increase in conversion time is going down from there, like would a 1060 take 10 minutes? I have no frame of reference, but I can certainly offer my own.

Edit: Also this stacking model is making my brain explode it is so awesome.
 
ChrisCall said:
I will say that I'm running a 1070 card and running conversions in 10 seconds.

I'm not sure how immense the increase in conversion time is going down from there, like would a 1060 take 10 minutes? I have no frame of reference, but I can certainly offer my own.

Edit: Also this stacking model is making my brain explode it is so awesome.
Click to expand...

1050 here and conversions take about 20-30 seconds.
 
NewAgeRipper said:
You got early access or something?
Click to expand...

I've been lucky enough to hear it in progress. ...and It's hard to contain my excitement this time is all x)
I mean, this AI is amazing as is, but just think about the results we already have with it, and then just take it all a step further.. or several steps.

It may not be a case of every track ever being perfect, but it's another (essential) tool in the tool box. When it works, it really works.
When it doesn't, well, you're no worse off than you were with the existing best model.
 
ChrisCall said:
I've been lucky enough to hear it in progress. ...and It's hard to contain my excitement this time is all x)
I mean, this AI is amazing as is, but just think about the results we already have with it, and then just take it all a step further.. or several steps.

It may not be a case of every track ever being perfect, but it's another (essential) tool in the tool box. When it works, it really works.
When it doesn't, well, you're no worse off than you were with the existing best model.
Click to expand...

True.
 
thanks man, so does this extract a good quality acapella to or is it mainly focused on getting a near perfect inst? im using splitter ai at the moment , would this get better results with acapellas? as im more into making my own beats so no use for inst, thanks.
 
You must log in or register to reply here.