Sound analysis

[Qacer]

Hi!

I wasn't able to find a tool that will allow me to compare two sound file and separate the ambient sound from the wanted sound. I do have access to Matlab, so I figured maybe I can load the two sound files and use some existing algorithms. The problem is that I am not familiar with any algorithms.

What I really want to do is use 1 sound source as the ambient source and use that data to "strip" out the ambient sound from another sound sample. Where should I start?

Thanks!

 

[imported_inspire]

I'm not sure where you should start, but you'll probably be looking at residuals. That probably won't help you, though...

 

[Qacer]

I was reading this article on Active Noise Control and thought that I could probably implement the same thing in Matlab. I tried using the Noise Removal tool from Audacity, but it didn't do too well. I also had another idea of converting the sound samples and maybe using some Photshop filters to process the image difference. I'm not sure if this is feasible, though.

 

[BrownTown]

is it the EXACT same background noise, casue then you can just subtract the one from the other in 5 seconds...

 

[MAW1082]

filter design . . . . 1/s . . . 1/s^2 . . . no . . . die laplace die

 

[Loki726]

If the noise you are dealing with is truly random, then you are screwed. Once it is mixed with your signal, there is NO way to remove it without having some knowledge of the characteristics of the noise or your signal. For example, if you know that there is a fan oscillating at 50 hz, you can filter out everything at 50hz and get more of the noise then your signal. Or your noise is distributed randomly across all frequencies, but your signal is only distributed from 0 to 10khz, removing everything over 10khz will result in having less noise compared to your signal. But if you don't have any information beforehand about your signal, or your noise, then there is nothing you can do.

Just having a sample of noise produced by your environment does no more to help you characterize your true distribution of noise than pulling a red ball out of a box tells you whether there were 50 red balls and 50 blue balls, or 1 red ball and 5 million blue balls in the box...

 

[imported_Tenesu]

An adaptive filter should be able to filter out noise even if it's random (not "white" or "AWGN" noise however). It requires a good deal of math/statistics to design one though.