A two-stage procedure for text prompted speaker verification is presen
ted. In this procedure, speech recognition (segmentation) and speaker
verification are carried out separately. In the first stage, Hidden Ma
rkov Models are used for identifying phone segments; in the second sta
ge, phoneme dependent Radial Basis Function networks are used for veri
fying the claimed speaker identity. Phone modelling is important, beca
use different phonemes characterise different aspects of a speaker. It
is found here that phone modelling makes it easier to reject impostor
s, because successful impostors are usually only successful for specif
ic phonemes. (C) 1997 Elsevier Science B.V.