Wern Jun Wang, W.N.Campbell, Naoto Iwahashi & Yoshinori Sagisaka
Tree-based Unit Selection
for English Speech Synthesis
Abstract:Data scarcity for many contexts is a serious problem in English speech synthesis. In this paper, we proposed a new unit selection scheme using broad
categories for segment description and use regression trees to factorize the
data. The cross-validation method was used to evaluate the validity of this
algorithm. Experiment results show that regression trees offer a promising
solution for data scarcity problem in the unit selection procedures for concatenative English speech synthesis.