PURPOSE: To approximate a phrase shape to the phrase shape of natural voicing and to generate natural pitch patterns adaptively to phrases having various lengths by independently controlling both a rise response speed and a speed reduction speed.

CONSTITUTION: One combination of three kind of constants is selected from a constant setting part 14 according to the number of moras of a phrase obtained from a character string analysis part 11 and a phrase mechanism 12 calculates a phrase component. An accent control mechanism 13, on the other hand, calculates an accent component with a step response function shown by an equation for the accent command supplied from the character string analysis part 11. Here, α and β are constants determining the response speeds of the phrase control mechanism 12 and accent control mechanism 13, and use values almost satisfying α=3.0 and β=20.0. Further, θ is the upper-limit value of the accent component and normally set to 0.9, etc. Thus, the phrase component and accent component are calculated and then added and the result is outputted to a vocal cord vibrating mechanism 15, which converts it into a fundamental frequency.