PURPOSE: To obtain continuous and smooth speech power when a continuous speech is synthesized by connecting speech units.

CONSTITUTION: A text as an object of speech synthesis is inputted from a terminal 10. A text analysis part 20 analyzes the input text to obtain phoneme series information and morpheme information. An information storage part 30 is stored with various information (Japanese dictionary, accent dictionary, power information, etc.) for synthesis. A rhythm information generation part 40 generates rhythm information by referring to the output information of the text analysis part 20 and the information in the information storage memory 30. A vowel power setting part 51 finds the speech power of the vowels on the basis of phoneme environment information and pitch information. A non- vowel power setting part 52 determines the power envelope of a phoneme put between vowels by utilizing the found vowel power. A speech synthesis part 60 connects the speech units and the synthesized speech is outputted from a terminal 70.