Synthesis of English intonation using explicit models of reading and spontaneous speech

A model of English intonation is presented in which a variety of intonation contours can be generated from a quantitative prominence labelling of stressed syllables. In one style of speech production, spontaneous speech, a short lookahead model can generate a variety of contours from the same quantitative prominence labelling. For another style, reading aloud, a long lookahead model determines the types of intonation patterns associated with texts. These typically come out as sequences of downward stepping contours, given appropriate initial conditions (though there is no explicit down step constant in the model). In both styles, the intonation contours are generated on the basis of a quantitative model of contour pitch prominence, in which the pitch prominence of the contour segments which make up the accent contours (and thence the intonation contours) is computed as a nonlinear function of the duration of the contour segment, the ratio of the FO value at one end of the segment to that at the other, and a rhythm constant.