This paper describes a system for the automatic labelling of prosodic events. A rule-besed approach was chosen in order to make explicit the generally implicit knowledge and classification strategy applied in prosodic labelling. The purpose of the system is to label accents and phrase boudaries. The language chosen was French. First, the psychoacoustic parameters, pitch, duriation and loudness are extracted from the signal and pre-processed in order to find out clues related to the prosodic ecents. Then a set of rules is used to examine the results of this analysis and to determine the placemnet of the labels.
|Tijdschrift||IPO Annual Progress Report|
|Status||Gepubliceerd - 1996|