Tutorials

[junh1024]

NAR: , IMO

You can spend 5 hours removing blips with ML, or run it through a gate in 5 minutes.

Again, You can spend 5 hours separating BVs with ML, or run it through AOM Stereo Imager D in 5 minutes.

[NewAgeRipper]

NAR: , IMO

You can spend 5 hours removing blips with ML, or run it through a gate in 5 minutes.

Again, You can spend 5 hours separating BVs with ML, or run it through AOM Stereo Imager D in 5 minutes.

I honestly don't know what ML and AOM is.

[djtayz]

I honestly don't know what ML and AOM is.

He's (probably) referring to using Machine Learning techniques to manually isolate background vocals, which indeed could take a long time, or use an effect to extract them. You already know how the second one works, as it was in your Colab tutorial, it's a matter of removing the mono signal (Adobe Audition does the same thing).

The first one can be done using apps like Melodyne or Inifnity. I've used them both and most of the time they do a decent job, when Audition fails. It separates the vocals as best as it can into notes so you select the ones that you don't need (in this case the notes of the lead vocal) and remove them.

[joaoasq]

Ultimate4 Vocal Remover has finally worked for me with Google Colabs method. Thanks @Anjok, @djtayz <3
Also thank you @NewAgeRipper for your amazing video tutorial!! It couldn't be posible without it.

I'd like to try demucs too, because I want to compare results myself and have some stems. Is there any similar tutorial?

[NewAgeRipper]

I'd like to try demucs too, because I want to compare results myself and have some stems. Is there any similar tutorial?

I have not had time to make a separate tutorial for Demucs. But it's basically the same thing as Vocal Remover. Just download the files needed. Put the folder needed in your Google Drive and the small file needed. Upload both the folder and the small file. Each time you open Google Colab it will show you the files used for each Vocal Remover and Demucs. You click the one you want to use and then go through each step click the play button as you go down the list. In Demucs it takes a little while for ffmpeg to install. After that you're on your way.

[Rafaelcrush]

Can i run Stacked Model again and again and again til voice complete disappears?

[djtayz]

Can i run Stacked Model again and again and again til voice complete disappears?

The overall quality will drop significantly. It's not a magic feature that only removes vocal bleed, it also can damage the synths and ambience of a song.

Not to say you can't experiment and see for yourself if you're curious then go for it.

[ChrisCall]

Can i run Stacked Model again and again and again til voice complete disappears?

The first pass is generally the best. Then the results diminish with each additional pass.
It will only strip the frequencies it identifies, so some parts of vocal will not be removed no matter how many passes.
For that reason, I'm thinking there may be benefit to developing more than one Stacking Model.

But in general, what I do, and if you're willing to put in the time... is to go through each individual vocal track created, listen to what's there, and mute out only the parts that you can really hear the vocals. Keep all the rest, as that is scattered instrumentation and ambiance. Each additional pass will take out less vocal and more instrumentation ... but certain parts of a song, like a few seconds worth, could benefit from 5+ passes, or even 10.. just not the whole song. So you mute out those few seconds in the stems and mix the rest of the sound back into the instrumental.

In general though, using the Stacking Model at least once will mostly always provide superior results to not using it. Just so long as you are willing to put in the extra time deciding what to mute out and what to mix back in.

[junh1024]

AOM Stereo Imager D is the name of a VST FX.

[NewAgeRipper]

AOM Stereo Imager D is the name of a VST FX.

I probably shouldn't ask but...anyway to get this for free? Or could you DM me with it?