[linux-audio-user] Re: Decent reverb

davidrclark at earthlink.net davidrclark at earthlink.net
Wed Feb 2 15:33:17 EST 2005

Hi Steve,

Thanks for your comments on impulses, etc.

My main concern was that, and I'm sure you know this perfectly well, that
mathematical physicists are fond of saying about impulses, in the context
of integration, that the shape of the impulse doesn't matter too much.
It's a handy way to sweep away a difficult derivation!

This is true for many physical phenomena.  But in audio --- and again
I'm sure you know this but others may not --- the ear is an extraordinarily
sensitive detector, and audio includes many different physical situations,
sometimes on very short time scales.  So I think it's important to ensure
that everyone understands that "doesn't matter too much" doesn't necessarily
mean "isn't audible."  


(The following is more specific, hopefully of interest to at least  some

On my web page is a song "latest" that uses calculated IR's.  Some software
is there to produce them, also.  But I know you're a busy guy!  The song
"latest" is for headphones, as usual.  Long ago you and some others 
encouraged me to post the IR generators.  They were so user unfriendly
that I took some time to "put some lipstick" on them.  Hopefully they're
not too bad now.

Regarding synthetic space --- ironically a problem with the Green's function
technique that I'm using, the calculated IR's are a little too accurate.
They have extremely sharp spikes, and the timing is very accurate.  You can
hear echos where you might actually hear more of a reverb (with actual
diffuse reflections).  Amplitudes can also become very large, yet the
sound have very little power.  Here again, I think that diffusion and diffuse
reflections need to be incorporated, or something mimicing them.  However,
this accuracy produces crystal clear sound, and I find more and more that
I like it compared to the mushy sound that a lot of audio software produces.
Some reverb models sound hilarious to me because they're so physically
erroneous.  I can't hear the music over the sound of my own laughter.

I'm not a big fan of ray-tracing at this time.  It just seems to me that
it's equivalent to a sparse grid without some sort of optimization or
pruning method.  But I have no experience other than reading technical

I've worked in the semiconductor industry with finite difference codes, so
naturally it occurs to me to try and build a IR generator using that 
technique.  Do you have any experience along those lines?  Please feel
free to contact me via email if you prefer, because that's probably not
of interest to LAUer's.  My crude estimates lead to a monstrous mesh and
no chance of reducing the time step.  This may be OK for IR generation...

Thanks for any further comments.

Best regards,

More information about the Linux-audio-user mailing list