🚀 go-pugleaf

Ross Muckey <muckeyr@mail.oit.osshe.edu> wrote:

>56002 processor, a stereo audio input is processed and then output in real 
>time.  I'm trying to create the effect of what a listener would hear if 
>this audio signal was broadcast into an enclosed room.  Varying the room 
>size and the reflectiveness of the walls would change the sound reaching 
>the listener.
>	Is there a name for this effect or process?  Any basic 
>information, theories, or algorithms on this effect would be extremely 
>helpfull.

The process you are trying to achieve is called "Auralisation"
(?largely thanks to Mendel Kleiner's paper of a few years ago, the
details of which I don't have with me?) and there are two aspects:

a) modeling the room, the source and the listener in such a way that
you can calculate the total impulse response for the path from the
source to the listener

b) implementing the resulting filter (the convolution of the impulse
response with the "dry" source material).

The first part seems straight forward at first, but is complicated by
the fact that sound waves don't quite behave like light waves, so
simple ray-tracing techniques are rarely completely convincing.

The best acoustic models that I've heard are produced by
"CATT-Acoustic" by Bengt-Inge Dalenback <catt@netg.se>.  Another
modelling product, with more emphasis on speaker models is "EASE/EARS"
by Wolfgang Ahnert (sorry, I don't have his e-mail address with me.)

Bose Corp also do acoustic modelling for their Auditioner tool, but I
don't think that it's available to the public.

TimeWare Corp (Tokyo, Japan) use a technique that they call
"RealWave", but more than that I cannot say.

For part (b) you have few choices, because room impulse responses are
(compared to most other sorts of audio-application FIR filters) really
long.  If you just want to hear what the room sounds like, you can
perform the convolution in non-real time on a workstation or PC with
something like Matlab, convert the result back to a .wav file and
listen to it over headphones.  If you want to listen to continuous
source material, but can live with some computational latency, your
best bet is probably a Lake FDP-1+ box, which can perform up to 190K
taps of FIR filtering (convolution) in real time at 48kHz, but has a
couple of seconds latency.  If you want to listen to room impulse
responses convolved with live sources in real time (no latency) then
you need a Lake Huron Digital Audio Convolution Workstation.  That
will give you up to 256K taps of FIR filter in real time at 48kHz,
with no latency.

If you want to get fancy and do all that and also be able to move the
source around in the room, move the virtual listener around in the
room, rotate your head as you listen (with the simulation fixed in
space), then you also need the AniScape software package (which runs
on the Huron platform or the dual-card CP-4 system).

For product literature contact len@lake.com.au.
For more general info try http://www.lake.com.
If you want to point out any glaring errors or violations of physics
or causality implied by this post, please contact me.

Andrew.
Andrew Reilly: reilly@zeta.org.au                            +61 2 559 5029 (h)
36 Acton St.            Senior DSP Design Engineer           +61 2 233 8655 (w)
Hurlstone Park          Lake DSP Pty Ltd                Fax: +61 2 233 8656 (w)
NSW 2193 Australia      GPO Box 4067, Sydney 2001            andrew@lake.com.au