Intonational phrase prediction based on a CART-guided TBL

Intonational phrase prediction is important for both the naturalness and intelligibility of synthetic speech.This paper presents a hybrid tree-guided transformation-based learning(TGTBL) algorithm which combines a classification and regression tree(CART) with transformation-based error-driven learning(TBL) to predict intonational phrase breaks from unrestricted text.CART automatically generates the TBL rule templates,thereby minimizing the need for human supervision during the TBL training.Results of comparative tests show that the F-score of TGTBL based intonational phrase prediction is 70.0% and that the TBL templates automatically generated by CART provide good alternatives or supplements to manual templates.