The code that goes with the LAC2012 conference paper does 3D and higher orders. In fact we used it to make a new 3rd-order Ambdec config for CCRMA's 22 speaker array. Its written in MATLAB/Gnu Octave, and it's not a lot of code. So plenty of opportunity for tinkering with the goal functions.
One comment that is not in the LAC paper is that with less than 50-70 parameters, the non-linear optimizer works quite well and converges quickly (less than a few minutes). 5 speaker first-order 2-D is 15 parameters. 22 speakers, 3rd-order is 352 parameters so some strategy is needed to guide it. The decoder for CCRMA took about 2 hours. Constraints and exploiting symmetries to reduce the number of parameters would help, but we haven't experimented with that yet. The bulk of the computation is matrix multiplies, so it would be amenable to a Cuda/GPU implementation. Aaron (hel...@ai.sri.com) Menlo Park, CA US _______________________________________________ Sursound mailing list Sursound@music.vt.edu https://mail.music.vt.edu/mailman/listinfo/sursound
