Prosodic aids to structural analysis of conversational speech

An algorithm for inferring the sentence structure of conversational speech using prosodic information is presented. The fundamental frequency (F0) contour shape, which is the major prosodic information parameter that best represents the structure of a spoken sentence, is analyzed in detail by means of F0 shape approximation with a sequence of linear lines. From such analyses, boundaries between grammatical units can be detected. In addition, the relation between these grammatical units can be inferred by reffering to natural conversational speech styles. Structural hypothesis inference capabilities are also investigated to determine the syntactic and semantic structure of sentences through computer simulation experiments.