Solved: How to isolate common sound from two sound patterns?

WIFT · September 14, 2010, 13:05:50

Hi,
I'm trying to solve this problem and I have reached dead end:

I have two sound patterns which both contains the same drum sequence. But both patterns contains also some other sound which I don't want there (each of the pattern contains different additional sound).

The drum sequences in both patterns are perfectly aligned so when I take one pattern and mix it with the other one while inverting the wave I completely eliminate the drum sequence and get a mix of the additional sounds from both patterns. But I want the opposite situation: I want to eliminate the additional sounds and get only the drum sequence.

Please, help me: how to make it? And what program do I need to do that?

Thank you very much.

LPChip · September 14, 2010, 13:58:37

I assume you're only talking about the modplug note data?

I would tackle this problem the following way:

Copy the entire pattern to your clipboard.

Paste this in excel, and use text to cells function to get the notes data separated from the other things, like effects.

Use a formula in a new cell to identify which notes are duplicates.

The formula would look something like this:

=IF(A1=B1;A1;"")

Copy/paste this formula to all rows. If done correctly, it should give you the correct pattern.

Copy/paste special to reconstruct the clipboardpatterndata and paste back in modplug.

WIFT · September 14, 2010, 14:18:27

Quote from: "LPChip"I assume you're only talking about the modplug note data?

Sorry, I probably expressed myself incorrectly. I'm talking about data in WAV file, not patterns in ModPlug.

Saga Musix · September 14, 2010, 14:56:56

If I didn't know any better, I'd be saying now that you're cross-posting from pouet, because exactly the same question was asked there some days ago already.

http://www.pouet.net/topic.php?which=7635&page=1
have a look at my (Saga Musix) answer there. This should work if the data is aligned perfectly.

WIFT · September 14, 2010, 15:06:50

Quote from: "Jojo"If I didn't know any better, I'd be saying now that you're cross-posting from pouet, because exactly the same question was asked there some days ago already.

http://www.pouet.net/topic.php?which=7635&page=1
have a look at my (Saga Musix) answer there. This should work if the data is aligned perfectly.

That's exactly the problem I'm trying to explain :) Thanks for the link, I'll give it a try.

Inavon · September 15, 2010, 00:57:54

Actually, I don't think that way works.
1. Let's call sample 1 "a+b" in which a is the drum sequence and b is the unwanted stuff.
2. Sample 2 is "a+c" in which c is the unwanted stuff.
3. Subtract sample 2 from sample 1. a+b-(a+c)=b-c
4. Now you add both sample together. a+b+a+c=2a+b+c
5. Subtract the result of step 3 (step 4 in pouet) 2a+b+c-(b-c)=2a+2c
2a+2c is just a+c with double the amplitude, so you just went in a loop. The mistake was that when you subtract both samples, you don't get b+c, but instead you get b-c.

So you're going to need some program that can recognize similarities and isolate them.

WIFT · September 15, 2010, 07:36:45

Quote from: "Trancefreak"Actually, I don't think that way works....

That's just what I realized after several experiments with inverting and mixing :/ But I'm not giving up ;) I'll start experiment with simple small sine and manually added unwanted stuff...

LPChip · September 15, 2010, 08:32:12

An alternative is to recreate them using existing samples.

WIFT · September 15, 2010, 10:29:45

I think I'm in dead end again, because this post (Sam. D) explains it.

Saga Musix · September 15, 2010, 10:52:03

Yeah, of course it doesn't work with any kind of signal (or else, you could easily isolate your favourite synth line from any piece of music). I should have mentioned that it seldomly works on arbitrary sound material.

Rxn · September 15, 2010, 20:06:30

It will not work when trying to manipulate the shape rather than spectrum of
the wave. Too bad spectrum manipulation at the current stage of audio
software development is fairly rudimentary when it comes to working
implementations.

I was planning to do some practical research on it a while back after some
guy in the sound engineering community that I administer asked to help him
to remove some repetitive noise from his material. The task was straight-
forward and it would be simply trivial with appropriate tools, strangely
enough, I did not find any software that was able to do that.

I didn't go ahead with the research at the time since this topic did not
appear that much in demand then, but it seems that in fact there would be
more interest in it now with the problem at hand.

Rxn · September 15, 2010, 20:14:47

What we need here is:

get b-c by a+b-(a+c)

get the spectrum of b-c (it will be the same as the spectrum for b+c)

subtract the spectrum of b-c from 2a+b+c.

No software can do it cleanly because they do not take into account the
dynamics (variation over time) of the spectra, or at least I haven't seen one
that does while in theory that should be quite possible.

Saga Musix · September 15, 2010, 21:38:00

you have the same problem in the frequency domain as you have in the time domain, so it's no wonder that there is no software which can do it in the frequency domain yet.

WIFT · September 18, 2010, 13:33:37

Vocal Eraser Plugin for Sound Forge can do that. Not really affordable solution, but it's possible. It can erase the central part of the sound and it can also keep it while removing the rest. Of course, I have to first prepare appropriate "stereo samples", but that's obvious and easy to do.

uncloned · September 18, 2010, 14:14:43

hmmm - I got sound forge with sony vegas. I need to look into that.

I have some old tracks I'd like to extract the vocals from - thanks!

Can you explain the technique a bit?