PURPOSE: To obtain high-quality output voices in short processing time without executing intricate processing like syntax analysis of an input sentence by previously registering the phonetic symbols of every word of a dictionary section and the next word appearance reliability.

CONSTITUTION: The character strings inputted from a terminal 1 are collated with the character strings of the heading previously registered in the dictionary section 2 in a form element analysis section 3 and are divided to respective words; thereafter, the word strings with the phonetic symbols are inputted to a synthesis parameter forming section 6 and the accent type and the word strings with the next word appearance reliability are inputted to a rhythm forming section 4, respectively. The position of the silent parts existing in the output voices and the duration time thereof are calculated from the above- mentioned reliability and are outputted together with the accent type as rhythm information to the forming section 6. The synthesis parameter of every rhythm of a rhythm parameter storage section 5 are continuously interpolated by the information of the forming section 4 in accordance with the phonetic symbol strings from the analysis section 3 and are outputted from a terminal 8 via a voice synthesizing section 7.