Investigation of usual and unusual articulation based on simulations and observations (Speech) -- (国際ワークショップ"Asian workshop on speech science and technology")

概要

論文の詳細を見る
Inverse estimation of articulatory movement from speech signal faces the one-to-many problem. To reduce the one-to-many problem, we classify possible articulation with normal acoustic properties into usual and unusual articulation, where we suppose only the usual one is used in normal speech. In the inverse estimation for normal speech, the unusual articulation must be excluded. For that reason, we generated a variety of articulations using a physiological articulatory model, and synthesized the speech sound from the articulation. This data set was categorized into five Japanese vowels based on the acoustic features, and was analyzed and clustered using the principal components analysis. Then, we compared the simulation with real articulation data and derived a criterion for distinguishing the usual and unusual articulation. The proposed criterion was verified by consideration of distinctive features. As a result, the proposed criterion was suggested the possibility of excluding unusual articulation.
社団法人電子情報通信学会の論文
2008-03-13