Abstract

In this paper we present several algorithms that increase the robustness of
SPHINXth, e CMU continuous-spccch speaker-independent recognition
system, by normalizing the acoustic spacc via minimization of the overall
VQ distortion. We propose an affme transformation of the cepstrum in
which a matrix multiplication performs frequency normalization and a
vector addition attempts environment normalization. The algorithms for
environment normalization are very efficient and they improve dramatically
the recognition accuracy when the system is tested on a microphone
othcr from the one on which it was trained. The frequency normalization
algorithm applies a different warping of the kquency axis to different
speakers and it achieves a 10% decrease in error rate.