Speaker Verification with Fuzzy Fusion and Genetic Optimization

Most speaker verification systems are based on similarity or likelihood normalization techniques as they help to better cope with speaker variability. In the conventional normalization, the it a priori probabilities of the cohort speakers are assumed to be equal. From this standpoint, we apply the fuzzy integral and genetic algorithms to combine the likelihood values of the cohort speakers in which the assumption of equal <I>a priori</I> probabilities is relaxed. This approach replaces the conventional normalization term by the fuzzy integral which acts as a non-linear fusion of the similarity measures of an utterance assigned to the cohort speakers. Furthermore, genetic algorithms are applied to find optimal fuzzy densities which are very important for the fuzzy fusion. We illustrate the performance of the proposed approach by testing the speaker verification system with both the conventional and the proposed algorithms using the commercial speech corpus TI46. The results in terms of the equal error rates show that the speaker verification system using the fuzzy integral is more favorable than the conventional normalization method.