How to extract the common part between two audio signals and remove it from the signal?

Question

If I have two audio signals Y1 and Y2 in Fourier domain that are the results of multiplication of S with H1 and H2 respectively (convolution in time domain):

Y1=H1*S

Y2=H2*S

And I don't have S and H1, H2, but I know that S is the same in both Y1 and Y2. Is there a good way to extract S out to only have H1 and H2 in the right side of the equation?

thanks

So you have many examples or just these two? – Jon Nordby Dec 02 '18 at 02:34 — Jon Nordby, Dec 02 '18 at 02:34
Can you stimulate the systems that produced Y1 and Y2? – Jon Nordby Dec 02 '18 at 02:34 — Jon Nordby, Dec 02 '18 at 02:34

score 1 · Accepted Answer · answered Aug 21 '18 at 06:05

1

Looks like two equations with three variables. I don't think that you can extract the original signal, S.

answered Aug 21 '18 at 06:05

Gideon Kogan

662
4
18

Sure. But I was thinking of more like a blind deconvolution that can get benefit from having two signals with a mutual signal in them. – Atra Es Aug 21 '18 at 17:28
As long as I know, blind separation helps separate the signal to components but does not alow reconstruction of the original input – Gideon Kogan Aug 21 '18 at 17:40
Mainly my purpose was to get rid of the S and to have just the H. But fortunately, I don’t need to do that anymore! Thanks for your help though. – Atra Es Aug 21 '18 at 17:45

score 1 · Answer 2 · answered Apr 25 '20 at 11:38

I was asking myself the same question. My first approach was to do a crosscorellation between the signals, to get the point where they are most similar. then I padded one of the two so they are exactly "time synchronous", meaning the main sound source was exactly centered. I then substracted the one from the other. I think this is a valid approach, although I didn't follow through on it and its not working properly. If youve got additional ideas, let me know ;)

How to extract the common part between two audio signals and remove it from the signal?

2 Answers2