[linux-audio-dev] Realtime convolution - a threading problem?

Fons Adriaensen fons.adriaensen at skynet.be
Wed Apr 6 23:58:28 UTC 2005

On Thu, Apr 07, 2005 at 01:33:01AM +0200, Florian Schmidt wrote:

> - Using convolution only for the "early reflections" and using classic
> dsp based reverb tails (i don't know much about how these work though)..
> - Making use of the fact the tail is typically not having much variation
> except for exponentially dropping volume. Thus one could use a
> relatively short (and relatively early) chunk of the tail over and over
> just fitted to the right volume with a gain factor.. This would not
> reduce the number of multiplications, but it would help with memory
> access..

Dividing the reverb into the early reflections and the tail part is
a very good idea. You can use different short convolutions for the
ER (depending on source location), and since you can tolerate more
delay for the tail, just use a larger convolution size for that, and
absorb the latency into the delay.


More information about the Linux-audio-dev mailing list