Tuned out Dibrom's tone/noise masking bug.  The problem was the
long-suspect 8kHz data from Fielder's paper; now we have proof it's
wrong.

Monty

svn path=/trunk/vorbis/; revision=2896