Phrasing strategies in prosodic parsing and speech synthesis

This paper reports on some experiments and results from ci research project called Prosodic Phrasing in Swedish in which the overall aim has been to investigate and model prosodic aspects of phrasing. A series of prosodic parsing experiments are reported where a prosody expert was given the task of identifying prosodic phrases solely on the basis of a visual representcltion of unknown spoken text passages showing the waveform, intensity and fundamental frequency. The results were then compared to two independent, auditively based transcriptions of the readings. The comparison demonstrated a close relationship between the two types of judgements, indicating that a rather reduced acoustic representation can serve as the basis for prosodic parsing. Discrepancies encountered were often the result of an interrelationship between phrase-boundary gestures and accentual gestures. Preliminary guidelines for the prosodic parsing of phrasing are proposed. The dependence between phrasing and accentuation is further explored in a speech synthesis framework, and the influence of focal accentuation on the phrasing impression is discussed.