A relation model which relates the physical changes present in emotion-containing speech to the emotional content perceived in the speech is proposed. By using statistical bases extracted from the physical parameters of the speech and from the associated emotion words rather than the parameters and words themselves, the model makes it possible to relate physical changes and emotional content independently of the choice of variables for consideration. Also in this model, emotions which are shared among listeners, or emotion "stereotypes" are used as the standard of judgment instead of the emotion intended by the speaker, and accordingly the emotions can be assumed to be observable and reproducible. In this study, first, the physical parameters of several speech samples are calculated and the emotional content of the same speech samples is obtained in a psychological experiment, and these data sets are processed by statistical methods to obtain orthogonal bases. Next, these bases are related linearly by multiple regression analysis. As a result, relation information which allows conversion between the physical parameters and emotional content of the speech to be performed is obtain.
|ジャーナル||Systems and Computers in Japan|
|出版ステータス||Published - 2001 3月 1|
ASJC Scopus subject areas