+/************************************************************************
+ * The basic idea here is that a VQ codebook is like an m-dimensional
+ * foam with n bubbles. The bubbles compete for space/volume and are
+ * 'pressurized' [biased] according to some metric. The basic alg
+ * iterates through allowing the bubbles to compete for space until
+ * they converge (if the damping is dome properly) on a steady-state
+ * solution.
+ *
+ * We use the ratio of local to average error as the metric to bias a
+ * variable-length word codebook, and probability of occurrence within
+ * that bubble as the metric to bias fixed length word
+ * codebooks. Individual input points, collected from libvorbis, are
+ * used to train the algorithm monte-carlo style. */