On Sat, 29 Jan 2005 10:28:23 +0100 Fons Adriaensen <fons.adriaensen@xxxxxxxxx> wrote: > On Sat, Jan 29, 2005 at 12:14:49AM +0100, Florian Schmidt wrote: > > > So there's plenty room for optimization (and some return value checking > > will be added too ;)).. If you know some tricks, let me know.. The > > sourcecode is pasted below for easier reference. > > For a typcal stereo room simulation, what you need is actually > > | L' | | a b | | L | > | | = | | | | > | R' | | c d | | R | > > where a, b, c, d are the responses to be convolved with. hmm, are a and b real and imag part of the FFT of the response for one channel? otherwise i wouldn't see how a stereo response file would have 4 values.. care to elaborate some? I hacked this thing in 1 day [incl. learning how convolution and partitioned convolution works - i actually wondered: what takes longer: writing a convolution engine or grokking brutefir's config files ;)], so i wouldn't be surprised if i made fundamental mistakes.. OTOH it sounds quite good even now.. > > In its present form, you code will do this if you split inputs > and recombine outputs, but only with a lot of redundant work. > > It could be a good idea to have a more flexible I/O configuration > that would allow this. as i don't grok the first part, i don't grok this either.. -- Palimm Palimm! http://affenbande.org/~tapas/