Eliminating inter-speaker variability prior to discriminant transforms | IEEE Conference Publication | IEEE Xplore