Thanks for the fast reply. Right now I am only interested in localization. I believe that even before you use the MultiFFT node in a standard localization network there has to be some sort of low-pass filtering. Otherwise the real-time spectrum coming out of the MultiFFT node is corrupted by aliasing.
EDIT: I just realized my mistake. Of course the relatively small sampling frequency of 16 kHz will always introduce a small amount of aliasing into the signal. It makes absolutely no sense asking for a filter AFTER the audio has been recorded.