ISCA Archive ASRIV 1994
ISCA Archive ASRIV 1994

Radial basis functions for speaker identification

Steven E. Fredrickson, Lionel Tarassenko

Artificial neural networks are applied to the problem of recognising people from their voices. With input vectors consisting of time-normalised cepstral coefficients, radial basis function networks are shown to exceed the speaker classification performance of multilayer percep-trons. With an optimised input vocabulary and the use of multiple observations, error-free results are obtained on a 50 speaker identification task.