Second ESCA/IEEE Workshop on Speech Synthesis September 12-15, 1994 Mohonk Mountain House, New Paltz, NY, USA

A Quantitative Model of German Intonation and its Application to Speech Synthesis

Bernd Möbius

Institut für Kommunikationsforschung und Phonetik (IKP), Universität Bonn, Germany, and Linguistics Research Department, AT&T Bell Laboratories, Murray Hill, NJ, USA

This paper presents the adaptation of FujisakTs quantitative model to the analysis of German intonation and its application to the synthesis of fundamental frequency (FO) contours by rule. We investigated the sources of variation of the moders parameter values as obtained by analyzing utterances in a database. On the basis of this analysis we have constructed a set of rules that capture linguistic as well as speaker-dependent features and generate artificial FO contours for target utterances. Acceptability of the rule-generated intonation patterns was tested in a series of perceptual experiments. The rules have been implemented in the HADIFIX speech synthesis system for German.

Full Paper

Bibliographic reference.  Möbius, Bernd (1994): "A quantitative model of German intonation and its application to speech synthesis", In SSW2-1994, 139-142.