SYNTHESIZED SPEECH PAUSE SETTING SYSTEM

摘要

PURPOSE:To properly set the insertion intervals and number of pauses and output a more natural synthesized speech by deciding whether or not a pause is inserted according to a linear discrimination function in which the pause is placed. CONSTITUTION:A part of speech classified no-pause section estimation system inputs a KANJI (Chinese character)-KANA (Japanese syllabary) mixed sentence 100, takes a morpheme analysis, and finds the parts of speech of respective words. Then statistical distributions of mora numbers are investigated for various parts of speech and it is decided whether or not a pause is set at an accent phrase border. It is decided whether or not the pause is inserted according to the linear discrimination function based upon the probability that the pause is placed corresponding to the part of speech of a word right before the accent phrase border and a prepositioned part of speech section that the mora number of a no-pause section belongs to and the probability that the pause is placed corresponding to the part of speech of a word right behind the accent phrase border and the postpositioned part of speech section that the no-pause section mora number belongs. Consequently, the distributions of no-pause sections can be found individually for various parts of speech.