ISCA Archive - Automatic prosody assignment for interactive synthesized dialogue systems (original) (raw)

Automatic prosody assignment for interactive synthesized dialogue systems

Julie P. Vonwiller, R. W. King, R. W. T. Lloyd

Correct assignment of prosodic data in synthesized speech has considerable importance to its intelligibility and acceptance. Accurate prosody is particularly significant for interactive speech response systems, where the synthesized speech responses form part of the inquirer-system dialogue, and it is known that specific prosodic patterns carry interactional meanings. These patterns have been described for natural speech in the intonation component of Halliday's systemic functional grammar model of language. This paper describes an initial investigation into the application of Halliday's model to compute the tone group pitch contour component within a text-to-speech synthesis system from a minimal specification. Pitch contours for fifteen interactional meanings are described, and examples of their automated computation are given. The paper discusses how the system might be incorporated within an automated speech response system. Keywords: speech synthesis, prosody, intonation, speech response systems