A System for Rule Synthesis of Prosodic Features of Speech of Multiple Language Based on a Generative Model of Fundamental Frequency Contours

点击翻译按钮获取中文摘要

(1)Definition of the goal and choice of target languages It was decided that the generative model for the fundamental frequency contours (henceforth F_0contours) of the common Japanese, already established by the principal investigator and his colleagues, shall be used as the basis and shall be extended to apply to other languages, that the extraction of linguistic information on lexical accents, syntactic structures and locations of foci is to be modeling a separate research project, and that British English, German and Greek are to be selected as examples of target languages.(2)Elucidation of the generative models for F_0 contours of British English, German, and GreekSpeech samples of text reading and dialogue by a number of native male and female speakers of the three languages were collected and analyzed. The results indicated that F_0 contours of speech of these languages can be generated using the same mechanism as for those of the common Japanese, but the patterns of control commands can be different among various languages. British English differs from Japanese in the timing and magnitude of the phrase commands as well as in the fact that the polarity of accent commands can be both positive and negative. German differs from Japanese in that it has two types of phrase commands ; the impulse type and the step function types. On the other hand, the constraints on the control commands of Greek are found to be identical to those of Japanese.(3)Construction of a system for rule synthesis of prosodic features of multiple languagesOn the basis of the above-mentioned results, a system of prosodic symbols and a set of rules for determining the polarity, magnitude and timing of control commands were devised for each target languages, and a system for rule synthesis of prosodic features was constructed, using the system for the common Japanese as its basis.