15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003

Polish Synthesis and Representation Levels of Intonation

P. Durand (1), A. Durand-Deska (1), R. Gubrynowicz (2)

(1) LPL-CNRS, France
(2) Polska Akademia Nauk, Poland

In this paper, we describe the basic levels of prosodic Representation applied to "Text-to-Speech" synthesis of Polish sentences by concatenation of diphones. The goal is to supply the synthesis device used in this project prosodic information in a way to fit natural sentences one. This work is devoted to sentences with "Czy" interrogative clause which show specific F0 contours. In MBROLA synthesis system, its necessary to supply F0 and duration information for each segment. On the upper level, INTSINT labeling is used because of its simple formal melody coding. At the intermediate level, it is necessary to find a scale that focuses on perceptually relevant melodic variations for different voices. For this purpose the semitone scale is used for it takes into account melodic variation and is independent of pitch absolute value. Given the melodic span of a given voice, it's simple to get contour in Hz to apply in MBROLA system.

Full Paper

Bibliographic reference.  Durand, P. / Durand-Deska, A. / Gubrynowicz, R. (2003): "Polish synthesis and representation levels of intonation", In ICPhS-15, 1133-1136.