This directory contains results for scaled-down MixAR speaker recognition with WSJ.

Setup:
15 speakers training (about 65 s each), testing (aout 30 s each).

With different number of mixtures, and duration of window size.