Modelling tone and intonation in Japanese
Date
1989
Authors
Miyamoto, Tadao
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The purpose of this thesis is to investigate the acoustic characteristics of tones and the syntactic, semantic, and extra-linguistic nature of intonation in order to design a simple linear phrasing and pitch contour model for a text-to-speech program in Japanese.
In the model, sentences are, first, parsed semi-automatically into inter mediate (major) phrases. In doing so, the interactions of syntactic, semantic, and extra-linguistic factors are framed as the algorithmic linear parsing model which creates at least three types of phrasing variations for given sentences.
Once the intermediate phrase boundaries are located, the accentual phrasing algorithm parses intermediate phrases into accentual (minor) phrases by means of examining accentuations and accentual features of words in a two-word-sized-window-cursor . Whenever there is at least one item in the cursor which is underlyingly (originally) accented, an accentual phrase boundary is inserted in between the items.
Based on Pierrehumbert and Beckman's ( 1989) mechanism, relevant tones are assigned to the accentual phrases. These tones are then scaled in transform spaces to have vertical spatial values. The scaled tones are linearly interpolated to draw a schematic pitch contours, completing the operations which provides a speech synthesis program with natural pitch contours.