Utterance verification using fuzzy methods

In utterance verification, a claim identity is a linguistic unit such as a word or a phrase. An utterance of this claimed identity is verified by computing a similarity score to compare the content of spoken utterance with the word or phrase midel whose identity is claimed. Most of the current normalisation methods compute the score as the ratio of the claimed utterance's and the impostors' likelihood functions. Based on analysing false acceptance and false rejection errors occured by the current methods, we propose normalisation methods based on Fuzzy C-Means, Fuzzy Entropy and Noise Clustering methods to find better scores which can reduce those errors. Experiments performed on the TI46 speech corpora show better results for the proposed methods.