Abstract
In this paper, we present normalization methods based on fuzzy set theory to enhance the performance of voice-based person authentication systems. For an input utterance and a claimed identity, a similarity score is calculated and compared with a given threshold to accept or reject the claimant. Using fuzzy set theory, the similarity score can be viewed as a fuzzy membership function, which denotes the degree of belonging of the input utterance to the claimant model. We present fuzzy similarity scores based on fuzzy entropy and fuzzy c-means membership functions. Furthermore, the noise clustering method supplies a very effective modification to all methods, which can overcome some of the problems of ratio-type scores and reduce the false acceptance rate. Experiments were performed to evaluate proposed normalization methods for speaker verification using the ANDOSL database and utterance verification using the TI46 database. Experiments showed better results for fuzzy similarity scores.
