PURPOSE: To offer the voice rule synthesis device which prevents voice quality deterioration and outputs a natural synthesized voice close of a human voice.

CONSTITUTION: Information on a synthesized voice consisting of information on a phoneme, a rhythm, and a language which are inputted from an input part 1 is used to set rhythm parameters according to control rules of phoneme continuance, a fundamental frequency, and power. A voice parameter generation part 4 selects proper voice units required for synthesis in a phoneme parameter data base 5 and preferentially select voice units with specially, mean acoustic characteristics, and a voice synthesis part 6 synthesizes the voice, which is outputted from an output part 7.