[linux-audio-user] Re: Decent reverb

Steve Harris S.W.Harris at ecs.soton.ac.uk
Wed Feb 2 04:12:20 EST 2005

On Tue, Feb 01, 2005 at 12:45:53 -0800, davidrclark at earthlink.net wrote:
> The "correct" or perhaps "most correct" impulse is a Dirac delta so-called 
> "function."  Practically speaking, we cannot generate one.

Also, I have experienced problems with using one sample impulses (arguably
the nearest you can get to a Dirac), they dont always produce accurate
impulses. I found that a Chirp (rapid sinewave sweep) often got good room
impulses, but theres no ideal excitation impulse in my experience.

Ofcourse there are some rooms where you dont have any choice, due to lack
of power or dampness and you have to settle for something mechanical
(baloon, starter pistol, spark generator etc.). e.g. the are some old salt
water storage tanks in the basement of one of the buildings at work, a
pair of huge interconnected concrete tanks, sound fantastic, but theres no
way I'm taking anything mains or car battery powered down there :)
> Steve Harris said that the impulse does not matter too much as long as it's
> wide band.  Loosely speaking, and theoretically speaking, this is quite
> correct.  But because we cannot actually physically generate Dirac delta
> "functions," it turns out to matter.  You can see (hear) this for yourself by 
> creating square waves, ramps, truncated sin waves, etc. and generating the
> response to these.  One could argue that these don't qualify as impulses, 
> but that's what we actually generate in any empirical situation --- a 
> nonideal impulse.

Right, but you deconvolve with the impuilse after to reconstruct the
impulse - again its not ideal, but what you get is an impulse ass long as
you did the maths correctly and made a good recording.

OTOH I would agree that calculated impulses are also great :) and they
have the advantage that you can model that space you wish to simulate and
change its parameters to regenerate different impulses to allow control of
parameters that captured impulses dont.
> This is why I use calculated IR's rather then recorded ones (see below).
> The IR's calculated use as close to an ideal Dirac delta "function" as is
> possible on a particular machine.

True enough, though the downside of this is that it is still a synthetic
space, and I would expect it to be lacking some of the subtle aspects of
real spaces. I havent heard the outputs of your software though, I would
imagine you can get very good synthetic impulses if youre willing to madel
spaces in enough detail and burn enough compute time ray tracing them.

- Steve

More information about the Linux-audio-user mailing list