RULE AS MEANS FOR IDENTIFYING AND SIMULATING EACH SOUND ETC. OF JAPANESE FIFTY SYLLABARY TABLE (INCLUDING VOICED SOUND, P-SOUND, CONTRACTED SOUND AND SYLLABLE NASAL ETC.)

摘要

PROBLEM TO BE SOLVED: To solve the problem that in a conventional speech recognition method, a large amount of statistic data of speech is required, and it is not flexibly adapted to change of a speaker, and there is little flexibility of synthesis sound, as actual speech data is utilized for synthesizing speech.;SOLUTION: Air vibration characteristic of each sound etc. of Japanese fifty syllabary table (including voiced sound, p-sound, contracted sound and syllable nasal etc.) is investigated, and irrespective of a pitch or tune of a note, a characteristic for defining each sound is extracted and made as a rule for each sound. According to the rule, discrimination and synthesis of speech are performed. When the discrimination is performed, input speech is analyzed one by one, and its result is collated to the rule to determine which note of each sound etc. of the fifty syllabary table corresponds. When the synthesis is performed, a frequency characteristic and a transit characteristic etc. are set according to the rule, amplitude limit and transit variation etc, by these characteristics are added to a basic wave and harmonics etc. which are separately generated.;COPYRIGHT: (C)2009,JPO&INPIT