The paper presents a description of F0 control for purposes of the synthesis of Polish speech. It was assumed that information on sentence structure should be sufficient for adequate description of basic melodic patterns. From the corpus including about 1 minute-long newspaper passage, read three times by 6 persons, most frequent F0 contours were selected. In order to establish intonation pattern similarities in various fragments of the text, statistical methods of distance estimation between individual replications were applied, along with perceptual evaluation. The Fujisaki model was adopted to approximate fundamental frequency courses. Initially, standard values of parameters controlling the phrase component (defining declination) and accent component (determined for individual accent groups) were applied. Approximation results obtained from mathematical analysis indicated that modifications of functions controlling the phrase and accent components were necessary, both with respect to parameter values and the manner of their control. Information on sentence structure (the number of clauses, their structure, length and type) was applied in the module generating F0 contours. Control parameter values were optimized on the basis of results of perceptual experiments.

Keywords: prosody, F0 modelling