Extract percussion from song

Hi all,

I have a file that consists of Vocals + Percussion. I have another file which just has the vocals only. My aim is to extract only the percussion.

I’ve tried the following:

  1. Duplicate > Mute 2nd Track > Split Stereo to Mono on first track > Invert one track > LS Filter Duplicated track
  2. Effects > Vocal Remover

None of them worked. My audio is definitely stereo and I have a separate vocal track without percussion so am hoping to use the latter to somehow isolate the percussion in the background.
I’m not audio-savvy, so any advice in lay man’s terms will definitely go a long way. Thank you in advance.

As I said, I’m one step ahead of many methods of vocal removal in the sense that I have a high quality apple loseless file which contains just the vocals only without any percussion. Am somehow hoping for a method which will use this to my advantage.

I have a file that consists of Vocals + Percussion.

Is that an MP3 or internet download?

If it is, then that file doesn’t exactly match anything else, and an exact match is what’s needed to get cancellation. Further, Apple Lossless isn’t completely perfect, either.

You don’t need to go to a bunch of steps for this. Import both files one after the other. They will appear one above the other and play at the same time. Invert the vocal only and play the show. Something very significant should happen even if you don’t get cancellation.

Before you invert one and play the show, listen to them both on the timeline together. They should sound perfect. If there is any echo or hollow sound at all (switch between the tracks with SOLO and MUTE), then the two voice tracks are not identical.

You can make the combined track into mono if you wish, but the effect should be clearly audible with one track in stereo. You may need to slide one track sooner or later slightly to make everything match, and it’s also possible you may need to jockey Amplification.

If one track went through digital signal processing in its life, then the two tracks may actually be different length. That’s something else you can check while the two tracks are still right-side up. They’re clear at the beginning, but are they still clear at the end?

This isn’t easy even if your two tracks are uncompressed wav files and certified an exact match.


The inversion method only works if the tracks are perfectly identical and exactly in anti-phase.
Otherwise try spectral-subtraction method, e.g. RJH stereo tool or Kn0ck0ut , which does not require absolute precision , ( spectral-subtraction method does add digital artifacts :frowning: ).

Thanks for all the suggestions. I tried using the inverter, but no luck. Perhaps because the source is from a DVD encoded in 256Kbps/48Khz/AC3

A pass through AC3 compression will produce a digital celebrity not related to anything else.

We warn people never do production in MP3, but really it’s any compression format. They are end-products designed for customer enjoyment and generally contain “gotchas” making them unsuitable for post production.

For one notable recent failure, people trying to produce subliminal shows in MP3 are doomed to failure. MP3’s very reason for existence is to delete subliminal sounds.


If you post 5 seconds of each track [matching] we should be able to tell you if extraction of percussion is possible.

[ How to attach sound-files to posts in this forum ]

I’ve attached both files, vocals only and vocals and percussions. Please view. Thanks.
Vocals and Percussions.aiff (1.24 MB)
Vocals Only.aiff (1.16 MB)

Is there any portion of the mixed performance where the drummer stops, however short? Having constant drumming in the mix pretty much precludes any convenient inspection tool and no simple/quick effects tools work for me any better than you, although I should be able to get some affect. It’s suspicious that simple inversion has no effect at all.

Broadband impact sounds from the drums pollute the spectrum and the waveforms leaving blind chance which could take days. I won’t be doing that, so any short segment when they’re not constantly drumming is the answer.


I’m afraid not, the whole mixed piece has the constant drums in the background.

I can do it myself and don’t mind how many days/weeks it takes. Could you please detail the steps for the above?

leaving blind chance

This isn’t the process for finding the cancellation, this is the process for finding it by wild luck, and assuming there is a cancellation point.

The first step I did was reduce everything to mono.

Select the first track by clicking just above MUTE. Tracks > Stereo to Mono.
Do the other one.

I was going to go right for the Step and Test process, but I got another idea.
Push the bottom track very slightly late (to the right) with the Time Shift Tool (two sideways black arrows). The top track will play first by a tiny bit and then the bottom one. It should be a mess.

Select the first track by clicking just above MUTE. Effect > Change Speed. Slow it down by some small number, just enough that it sticks out to the right of the bottom track. So it starts before the bottom track and then ends after. Listen to the performance. Is there any time during the show where the chorale starts sounding wooshy like they’re in a tunnel?

Effect > Invert one of the tracks and play it again. Any luck? If one of those variations doesn’t give you an odd sounding total playback sometime during the song, then chances are there is no cancellation point.


Here’s what Kn0ck0ut does …

I did warn you they’d be plenty of digital artifacts generated.

You’d be better-off reconstructing , (substituting), the percussion, rather than trying to extract it .
There are plenty of free percussion sounds on Freesound , e.g.

Thanks for that great drum isolation. Could you please briefly tell me how I can do this for the entire track in Kn0ck0ut? It’s the first time I’ve heard of this software.

I would prefer that too, but cannot find anything close to what I have for the percussions. I don’t mind digital artifacts, as I’ll be overlaying a different song (but similar Gregorian chanting genre) on top of it.