Hidden Semi-Markov Model Based Speech Synthesis

In the present paper, a hidden-semi Markov model (HSMM) based speech synthesis system is proposed. In a hidden Markov model (HMM) based speech synthesis system which we have proposed, rhythm and tempo are controlled by state duration probability distributions modeled by single Gaussian distribu-tions. To synthesis speech, it constructs a sentence HMM cor-responding to an arbitralily given text and determine state du-rations maximizing their probabilities, then a speech parameter vector sequence is generated for the given state sequence. How-ever, there is an inconsistency: although the speech is synthe-sized from HMMs with explicit state duration probability dis-tributions, HMMs are trained without them. In the present pa-per, we introduce an HSMM, which is an HMM with explicit state duration probability distributions, into the HMM-based speech synthesis system. Experimental results show that the use of HSMM training improves the naturalness of the synthesized speech. 1