[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Ki-Young 2 mic sound field question



Dear Ki-Young
It would be a good idea to add room reverberation of some form. The easy
way to do this is to use the image method. For details see:

@article{Allen79b,
   author = {Allen, J. B. and Berkley, D. A.},
   title = {Image method for efficiently simulating small-room acoustics},
   journal = JASA,
   volume = {65},
   pages = {943-950},
   year = {1979}
}

and the paper is posted at
http://auditorymodels.org/jba/PAPERS/Allen/Allen79b.djvu
You will need djvu to view this, which may be downloaded from
www.lizardtech.com (or http://www.djvuzone.org/download/index.html ).

In the next week I will place some matlab code on my website that
computes the room impulse response. In the mean-time, there is fortran
code in the above paper.

Another tweek is to add head defraction. There is code to do this as
well, but I dont have the reference in my head. The work was done by
Dick Duda. Do a search on google and you will find it.

Jont


Automatic digest processor wrote:

There is one message totalling 101 lines in this issue.

Topics of the day:

  1. Simulating a sound field and two mics

----------------------------------------------------------------------

Date:    Thu, 17 Apr 2003 09:51:00 -0500
From:    Pradyumna S Upadrashta <prad@MAIL.AHC.UMN.EDU>
Subject: Simulating a sound field and two mics

Dear Ki-Young,

Simulating your situation is straightforward. It is essentially the
"forward problem" -- that is, you define a linear mixing matrix and
proceed to write out the linear model that describes the activity at
each microphone, given a specific source distribution.

I'm more familiar with "forward" (you provide the geometry) and
"inverse" (no solutions without making assumptions) problems in EEG /
MEG analysis.

Say Vout(i) is the voltage measured at your microphone, i, then you can
write the forward model as:

Vout(i) = G1(i)*Y1 + G2(i)*Y2 + ... + Gn(i)*Yn + error

Where G1, G2, G3,...,Gn are the individual "mixing matrices" for each
source Yk, with n sources total.

Vout(1) is the output measured at mic 1
Vout(2) is the output measured at mic 2

All you need now is a reasonable approximation of the geometry of your
scenario, which will determine the elements of the Gk's. It will have to
account for loss as a function of distance and medium. Presumably this
is constant for a given geometry, which is what allows us to write the
linear equation above; i.e., the loss w.r.t. distance is nonlinear since
sound amplitude should fall off as r^2; i.e., sound generically
propagates with a spherical wavefront and is described by
compression/rarefaction or longitudinal propagation. Of course the
situation is complex for different types of sound sources; for a good
overview, look at the following:

http://www.kettering.edu/~drussell/Demos/rad2/mdq.html
http://www.kettering.edu/~drussell/Demos/radiation/radiation.html

The Yk are the activity (amplitudes) of the sources over time, i.e., a
time-series of the amplitudes generated by the sources.

The "error" term accounts for spurious background noises in the
environment (e.g., air flow around the mic head, spurious noises ala
60Hz electrical frequencies, etc.). Measurement of the output of a
single/double microphone in a "silent" room might be a good
approximation of this term.

Hope that helps. Or at least, hope it doesn't confuse ;-)
I might have missed something since I did this rather quickly.

pradyumna

__________________________________
Pradyumna S. Upadrashta, PhD Student
prad@med.umn.edu
612-725-2000 x 1464




----- Original Message -----
From: "Ki-Young Park" <pkyoung@EEINFO.KAIST.AC.KR>
To: <AUDITORY@LISTS.MCGILL.CA>
Sent: 13 April 2003 17:23
Subject: correlation btw 2 singals incoming two ears from
distributed souces



Dear all,

I am working on the speech stuff, recognition and enhancement. While
using two signals with two microphones moderate distance

apart, say a

few tens of centimeters. I assume there is one speech source,
and distributed noise sources all around a room
instead of a point-source.
( and also assume additive noise. )

Is there publication on the correlation of two signals incoming into
two mics, when there are distributed noise sources around.

and is there any way to simulate this situation?

Any comments and references will be appreciated.

Thank you.



------------------------------

End of AUDITORY Digest - 15 Apr 2003 to 16 Apr 2003 (#2003-84)
**************************************************************


------------------------------

End of AUDITORY Digest - 16 Apr 2003 to 17 Apr 2003 (#2003-85)
**************************************************************



--
Jont Allen, 217/244-9567w; jontalle@uiuc.edu
Rm 4061, MC 251; Beckman Inst., 405 N. Mathews, Urbana, IL 61801
http://auditorymodels.org/jba; jba@auditorymodels.org; JontAllen@ieee.org