Also, if anyone has any information about any lag between time-domain convolution (such as buffir~) and partitioned convolution, that information would be invaluable!

Am I right to assume that the lag is equal to the window length (which, in my case = 512samps = 1/86th of a second = roughly 11ms? am I right here?)

Cross platform - no (Mac only). Stable - yes. The external I was mentioning back in that post is now finished, with several gigs to attest to it's stability. It's currently not public, but I have it out to some people to test until such time in the next year as I can properly deal with releasing it (and some other bits and pieces).

Porting would be fairly feasible for anyone willing to deal with the code (I won't be doing it myself), but I've used the apple FFT routine, and vectorised the code so it's fast under mac, which makes the porting a little less easy. If you email me offlist at ajharker [at] g OR google mail [dot] com I'd happily send you the stuff to look at, but no PC stuff right now....

For delay - partitioned convolution should be the window length, which should be half the FFT size, as the data needs to be zero padded, so for FFT size 1024, the delay is 512 samples or 512 / 44100 = c. 11.6ms.

Regards

Alex

and if I can add to this, the demo we had at the GEMdays 4th MSP Poweruser Symposium was the most impressive thing I've seen for a long time!

pa