The windowed sinc filter generation code seems to be working fine. The FIR filtering code is also now working reasonably well. Occasionally the FIR filter will pop, but setting the normalization factor lower seems to help this. I suspect the problem is in the fixed point multiply/add. I also have a hunch the zero stuffing/sample discarding code is not correct, and I'll look at that soon to get it sorted out.

(binary:audio/x-wav)