PURPOSE: To obtain a synthetic speed close to an actual speech by calculating a fundamental frequency pattern corresponding to a voicing form from respective calculated values of the size values of an accent component and a phrase component, their time variations in size, and lower-limit critical values.

CONSTITUTION: A KANJI(Chinese character)-KANA(Japanese syllabary) mixed sentence is inputted from an input part 101 to a word division processing part 102 and a code specifying the voicing form of a synthetic speech to be generated is inputted at the same time. A fundamental frequency pattern calculating means calculates the fundamental frequency pattern corresponding to the voicing form from the size of the accent component and its time variation that an accent component calculating means calculate, the size of the phrase component and its time variation that a phrase component calculating means, and the lower- limit critical value that a lower-limit critical value calculating means calculates. Consequently, a natural synthetic speech waveform is generated.