Second International Conference on Spoken Language Processing (ICSLP'92)
Banff, Alberta, Canada
This paper describes the assignment of correct rhythm and intonation contours to speech generated by the SPRUCE (SPeech Response from Unconstrained English) speech synthesis system. The system employs around 10,000 normalised parametrically analysed syllables excised from recorded natural human speech as the basic units from which the synthetic output is built. Incoming plain text prompts a search and retrieval strategy in a 200,000 entry word dictionary. For each entry in the dictionary there is information concerning syntactic category, semantic relationships and syllable based phonological content. The paper discusses the philosophy and strategies involved in the SPRUCE method of prosodic assignment.
Bibliographic reference. Tatham, Marcel / Lewis, Eric (1992): "Prosodics in a syllable-based text-to-speech synthesis system", In ICSLP-1992, 1179-1182.