KVR Audio

maxhodges1 · Post by **maxhodges1** » Mon Jul 09, 2007 2:53 pm

anyone know a tool to extract drum sounds (kick, snare, etc) from a music track? It seems like it wouldn't be that hard because the drums are very regular, so the software would just remove all the none regular elements from the sample. For example, if you had two bars of music from a song, couldn't you do some kind of bitwise compare and remove all the data that isn't the same? i.e. like in photoshop you could stack two images and do a color dodge or whatever...
does such a tool exist?

morphex · Post by **morphex** » Mon Jul 09, 2007 6:45 pm

You need to check out Propellerhead's ReCycle software application.

Peace - morphex

AKJ · Post by **AKJ** » Tue Jul 10, 2007 9:14 am

what you want is imo not possible. you need separate tracks to have a good result. you could try to do some filtering (eq and mid-side separation etc.), but the result will not be satisfying.

declassified · Post by **declassified** » Tue Jul 10, 2007 9:25 am

AKJ is right. The FFT/subtraction things work well on pads, vocals and other "soft" sounds, but they fail on transient sounds like drums.

Pantsdown666 · Post by **Pantsdown666** » Tue Jul 10, 2007 10:34 am

maxhodges1 wrote:i.e. like in photoshop you could stack two images and do a color dodge or whatever...

The trouble with this analogy is that all the instrument "colours" - i.e. the frequency range - are common to many of the instruments, they cross over.
So the drums have many of the same "colours" as the vocals, for example, and software doesn't recognise them as separate instruments.

maxhodges1 · Post by **maxhodges1** » Tue Jul 10, 2007 10:55 am

AKJ wrote:what you want is imo not possible. you need separate tracks to have a good result. you could try to do some filtering (eq and mid-side separation etc.), but the result will not be satisfying.

Let's say you take tripod moutned pictures of a fix object, like a parked car, that is in front a background with motion in it, like automobile and people traffic. If your camera is rock steady, shouldn't you be able to digitally compare all the images and remove all that data that isn't the SAME from frame to frame?

How is that different from analzying a drum beat that is consistent from bar to bar amid a cacophony of other noises? I guess there is a lot of interference/wave cancellation etc, but if the 'background' music isn't too heavy, it seems like the same logic would apply, you analyze the contents of the "FRAMES" (i.e. a bar or a 1/4 bar) and wipe away the stuff that isn't the same from frame to frame.

impossible?

maxhodges1 · Post by **maxhodges1** » Tue Jul 10, 2007 10:58 am

why can I separate it in my head haha
I'll post the track. I bet I can do it in Wavelab. the non-drum part is very sparse...

z15 · Post by **z15** » Tue Jul 10, 2007 11:04 am

Let's say you take tripod moutned pictures of a fix object, like a parked car, that is in front a background with motion in it, like automobile and people traffic. If your camera is rock steady, shouldn't you be able to digitally compare all the images and remove all that data that isn't the SAME from frame to frame?

but what about if the car is occasionally partially occluded by (hidden behind) some another car or there's some heavy shadows cast on the car by nearby trees and other objects? or reflections from other objects in vicinity? lens flare from sunlight? or all that combined?

that's what you usually get with music.

that isn't to say you couldn't do it, but simply that it isn't really as easy as that, you'll usually end up either losing part of the sound or leaving frequencies from other instruments there... unless you do it by hand & ear in which case it's really tedious and isn't guaranteed to work anyway.

BertKoor · Post by **BertKoor** » Tue Jul 10, 2007 11:14 am

If this was possible, then you could also make the highly sought-after polyphonic audio -> midi conversion! We're clearly not there yet. The answer probably lays in using neural networks for analysis and recognition, just like our brain does. Someone has to take a load of time to do some R&D on the subject.

z15 · Post by **z15** » Tue Jul 10, 2007 11:21 am

BertKoor wrote:If this was possible, then you could also make the highly sought-after polyphonic audio -> midi conversion! We're clearly not there yet.

people are working on it though... http://www.music-ir.org/mirex2006/index ... on_Results (that's just for recognizing and extracting the melody line out of polyphonic audio tho, but pretty impressive results already if not nowhere near perfect)

maxhodges1 · Post by **maxhodges1** » Tue Jul 10, 2007 11:31 am

ok, I gave it a shot. I discovered that WaveLab's Spectrum Editor is pretty bad ass ( http://tinyurl.com/3cz7vj ) and that trying to extract drum sounds from a mix is a bit like trying to get the egg back out of a cake.

HelgeG · Post by **HelgeG** » Tue Jul 10, 2007 12:08 pm

excellent analogy, max! you had me rolling on the floor.

timobrien · Post by **timobrien** » Tue Jul 10, 2007 2:59 pm

maxhodges1 wrote:why can I separate it in my head haha

Human brains are far more powerful than any computer in existence.
Plus your brain can easily be fooled (ever see any optical illusions that look impossible??)
And your brain is hard-wired to look for patterns that may - or may not- be there like the "Face on Mars" photograph.

Too many people think that just 'cause a song is store digitally that the elements are separate.
It's just a right-speaker-wave and a left-speaker-wave and nothing more....

loachm · Post by **loachm** » Tue Jul 10, 2007 10:35 pm

maxhodges1 wrote:
AKJ wrote:what you want is imo not possible. you need separate tracks to have a good result. you could try to do some filtering (eq and mid-side separation etc.), but the result will not be satisfying.
Let's say you take tripod moutned pictures of a fix object, like a parked car, that is in front a background with motion in it, like automobile and people traffic. If your camera is rock steady, shouldn't you be able to digitally compare all the images and remove all that data that isn't the SAME from frame to frame?

How is that different from analzying a drum beat that is consistent from bar to bar amid a cacophony of other noises? I guess there is a lot of interference/wave cancellation etc, but if the 'background' music isn't too heavy, it seems like the same logic would apply, you analyze the contents of the "FRAMES" (i.e. a bar or a 1/4 bar) and wipe away the stuff that isn't the same from frame to frame.

impossible?

...you're sure you're feeling well?

maxhodges1 · Post by **maxhodges1** » Wed Jul 11, 2007 5:46 am

timobrien wrote:
maxhodges1 wrote:why can I separate it in my head haha
Human brains are far more powerful than any computer in existence.

My Commodore 64 kicks my ass at math. And Kasparov got his ass handed to him by Deep Blue. Human brains are for shit for a lot of tasks: trying memorizing a 12 digital number after hearing it just once.

Plus your brain can easily be fooled (ever see any optical illusions that look impossible??)
And your brain is hard-wired to look for patterns that may - or may not- be there like the "Face on Mars" photograph.

Too many people think that just 'cause a song is store digitally that the elements are separate.
It's just a right-speaker-wave and a left-speaker-wave and nothing more....

True, true. The human mind is great at pattern recognition. I still think what I'm seeking is probably fairly attainable i.e. human brain power augmented with computer tools. I'm sure there is some highly specialized audio forensic software with algorithms that might be useful for this... but in general, we'd probably get much better results synthsizing our own drum sounds rather than extracting them. There are a lot of bass drum and snare drum sounds in a lot of tracks that I listen to which just sound better than anything in my library (mostly GURU). I guess it's just the treatments (reverb, eq, dynamics). Just need to refine my craft a bit because it looks like sampling them ain't gonna work.

extracting individual drum sounds from music track