Bayesian speech synthesis framework integrating training and synthesis processes

This paper proposes a speech synthesis technique integrating training and synthesis processes based on the Bayesian framework. In the Bayesian speech synthesis, all processes are derived from one single predictive distribution which represents the problem of speech synthesis directly. However, it typically assumes that the posterior distribution of model parameters is independent of synthesis data, and this separates the system into training and synthesis parts. This paper removes the approximation and derives an algorithm that the posterior distributions, decision trees and synthesis data are iteratively updated. Experimental results show that the proposed method improves the quality of synthesized speech.