Performance evaluation of GMM and KD-KNN algorithms implemented in speaker identification web-application based on Java EE

This paper presents a possible improvement of the performance of a speaker identification web-application by introducing GMM and KD-KNN algorithms. The purpose of the web-application is to allow authentication of the user directly from the web-browser using his voice. The captured speech is streamed to the server where the MFCCs are extracted. The classification phase implements four different algorithms: KNN, KD-KNN, non-adapted GMM and adapted GMM. In this paper, error rate and time execution of each of the implement classification method is presented and discussed. The experimental results are then evaluated and the algorithm with the best performance result is given.