First International Conference on Spoken Language Processing (ICSLP 90)
This paper describes the methods for a Japanese text-to-speech system the authors have developed and its application to a voice supply mechanism for electronic mail. This system analyzes an arbitrary text morphologically with a dictionary which has 100,000 entry words, their grammatical attributes, and their phonetic and accent information. This analysis separates the text into individual terms, and the phonetic and accent data for the individual terms are obtained from the dictionary. In the proposed system, rules play an important role in improving the speech in order to produce a high quality speech. These rules decide whether the phonetic symbols for each term are separated or unified, and whether a part of them are modified or not : to modify a vowel sound to a long sound, to make a breath sound, and to make a nasal sound, whether the accent position of each term is shifted or not, and how long the duration between individual terms is. The authors tried out this system in the field of electronic mail. If the mail receiver wishes to confirm a mail text when out of his or her office, the person can call the mail box by telephone and obtain a synthesized speech of the mail produced by the system.
Bibliographic reference. Saito, Hiroyoshi / Kurihara, Motoshi / Kobayashi, Ken-ichiro / Hara, Yoshiyuki / Saito, Naritoshi (1990): "A Japanese text-to-speech system for electronic mail", In ICSLP-1990, 881-884.