HMM과 음소별 GMM을 결합한 고속 화자식별 방법

Abstract

This paper proposes a fast text-independent speaker identification method using HMMs and phonetic GMMs. The individual Gaussian component of a Gaussian Mixture Model (GMM) can accurately represent acoustic characteristics of a speaker, so the GMM is effective to make a speaker model in text-independent condition. On the oher hand, in the text-dependent speaker identification, is determined a priori. In the case, using the hidden Markov model (HMM) as a speaker model shows better accuracy, since the HMM can model the temporal structure of the input speech as well as the peaker identity. In order to combine the advatages of the GMMs and the HMMs in te text-independent speaker identification, we propose a system architecture using HMMs and phonetic GMMs. From the experiments, as the number of mixtures of the baseline GMM is increased to 320, we obtained an identification accuracy similar to that of the hponetic GMM with 14 mixtures for 45 phones, but the time elapsed to identify the speaker was five times longer thatn that of the phonetic GMM. And using the phonetic class GMMs, we can furher reduce the number of parameters to the half with similar identification accuracy of the phonetic GMMs except the parameters of HMMs.