To secure the quality of voice combination while reducing throughput for generating a fundamental frequency pattern for attaching intonation, etc., in voice combination that reads text out.

Approximation is made by a plurality of linear declinations, on the basis of the matter that the pattern of a fundamental frequency F0 can be approximated, by superimposing a component (called 'phrase component'), that gradually descends from the starting part (start edge) of voice to the ending part (end edge) within a section unit called a breath group on an undulating component (accent component) corresponding to an accent type of a word, and the inclination of the linear declinations is set to be gentler, the closer a straight line comes to the lowest fundamental frequency.