Theoretical/Practical Signal Separation Question

[Shaffie]

Ok this isn't fully directly DSP related but, it has some applicability. And it's sort of a general question so bear with me.

Ok So I have two signals in one sound recording, one followed by another in some microphone recording. Each signal has it's own individual characteristics and they are each bathed in what can be presumed to be AWGN. Now for all intents and purposes these signals are impulsive in nature and my problem is to be able to separate each signal and characterize it under a certain category of event. So if i'm givne the recording and I take the FFT of it I'm given frequency components. I was wondering would it still be valid to cut the impulsive parts of the signal out of the data and then analyze each separately, or would there be some issue as far as frequency mixing and separating is concerned. Please if anyone has any info on how I can handle this please give me some feedback, I'm new at this stuff. Also I mentioned earlier that the noise signal is presumed to be AWGN but if I actually wanted to find a better noise distribution could anyone give me some statistical tests then I could transform to procedural tests to help me do this? Thanks.

[C00kie]

Have a look at the noise reduction in CoolEdit or Adobe audition. It already excells in what you're trying to do. Supposedly it's what the FBI uses.

First you select one part of audio that contains a typical piece of noise. The program analyses that with FFT, and gathers a fingerprint of the noise. Can be white noise, AWGN, 50/60Hz mains humm, OMFG or whatever.

Then you apply the noise reduction on the whole soundclip. Any parts matching the noise fingerprint are filtered out, so you're left with the rest: your recordings!

Look for a demo version of Adobe Audition and see weather it works for you. There are other programs & plugins that work in a same fashion...