Speech Prosody 2008 Campinas, Brazil May 6-9, 2008

Modeling Intonation Variability with HMM for Speech Synthesis

Cédric Boidin (1), Olivier Boeffard (2)

(1) Orange Labs, Lannion, France
(2) IRISA / University of Rennes 1 - ENSSAT, Lannion, France

This paper proposes a statistical intonation model designed to deal with intrinsic variability in speech. In combining the advantages of two well-known statistical algorithms, CART and HMM, the proposed model takes advantage of available linguistic information and successfully tackles the issue of missing para-linguistic information. Promising results of the training process are shown and analyzed. 1. Introduction

Full Paper

Bibliographic reference.  Boidin, Cédric / Boeffard, Olivier (2008): "Modeling intonation variability with HMM for speech synthesis", In SP-2008, 115-118.