テキスト音声合成のための韻律制御の基礎実験

概要

論文の詳細を見る
Following items are necessary to process for text-to-speech in Japanese. 1) Replacing KANJI (ideographic) characters by KANA (phonetic) letters (articulation). 2) Processing the prosodic information (recognition of word, BUNSETSU (word unit), accent-phrase, and so on, and extraction of position of accent kernel). Each word has its own position of accent kernel. However, the position varies by combining some words. The details of the change of position are "movement", "generation" and "disappearance". Two factors decide the position of accent kernel. The first is the accent types of each word which constitutes the accent-phrase and the second is the information such as each part of speech of words being combined. In this paper, the change of position of accent kernel is made a rule by investigating these factors.
岡山理科大学の論文