Any pointers to ambience extraction/ synthesis algorithms? I'd assume it starts with (L-R)*filter, but I'd guess that the filter is tricky - it'd need both amplitude and phase characteristics carefully tailored.
L-R is not an ambience part of the recording. This is a false assumption of all the ambience extraction methods.
It should start with the measurement of the room you want to emulate.
A very good site:
http://pcfarina.eng.unipr.it/He has a cooledit module which can derive the impulse response of the hall to any direction from the 4 channel ambisonic measurement. You can generate the proper responses according to your speaker setup, but in reality it is not that critical.
Older, pre 5.1 ambience synthetisers ( JVC, Yamaha, Lexicon) are doing actually a very good job.
