From: xhost91!lance@uunet.UU.NET (Lance Norskog)
Subject: Re: Virtual Sound
Date: Wed, 20 Feb 91 19:10:03 GMT



ap207nra@venus.ycc.yale.edu writes:


>If anyone knows where I can find more information on research being done
>on "virtual sound", please let me know.

The February 1991 issue of Computer Music Research contains an article
on placing N speakers around you, controlled by N/2 MIDI synths with
stereo reverbs.  They put mikes in the center and calibrated against
MIDI volume numbers.  The article includes the mathematical model.
The Midi comes out from a Mac sequencer, and the sequence player
was modified to control the synths and reverb units according to
math model and the calibrated sound inputs.  They show a Mac 
application where you draw the sound placement over time in a 2D
plane; the sequencer takes this into account during playback. Very slick.

Their model is 2D, i.e. placed in a 2D plane.  The last line says
something to the effect that extending it to 3D should be easy.
I'll go along with this just as soon as I understand the page
of formulae they gave.

If you stick to playing back pre-built sound samples, a separate
processor could handle 3D sound placement.  Antex in Gardena, CA
(213-something) is advertising a PC card with stereo digital sound
output, VGA with NTSC, a TI 32010, and 1Mb of RAM.  Two of these 
could handle sound and video for a head-mounted rig, with 3 
speakers around you and one overhead or 2 speakers in each
ear mounted diagonally.

Given that the functionality of $500 synths would take $5000 to
add to an all-digital VR system, I think MIDI has a place in VR,
especially in room-based systems.  (As opposed to body-mounted systems.)

On the subject:
I brought up MIT's Csound package under 386 UNIX, and it takes
1-30 times real-time to compute scorefiles.  That is, 1 minute
of 8K/s samples takes 1-30 minutes to generate.  Quadraphonic or
3D placement will take a little longer.  (Now, if only I could
find quadraphonic or octaphonic headphones.)   Csound takes 
instrument and scorefile specifications, and generates sound samples 
with controllable sample rates and sound quality.  The csound player
is built around partial interpretation; it builds up an internal
database of what you want done, then cycles through it.  The code
is pretty unreadable, but it looks like they're trying to be efficient.
They claim a Sun-4 can generate in real-time.  It has stereo and
quad sound by panning volume; this is cheesy.  Real placement needs
phase control.  

Csound has a lot of neat stuff.  You can take a sound sample, 
analyze it to death, and turn it into a frequency-based
instrument specification.  This apparently will do pitch-shifting
correctly, with no hamsterization.  I have long dreamed of an 
electronic "pan" or steeldrum; this might help.

Also, CMusic and the CARL library from UCSD are available to researchers.
F. Richard Moore, Mr. CARL, has a book out on CMusic.  I haven't
downloaded the package yet.

Lance Norskog