First European Conference on Speech Communication and Technology

Paris, France
September 27-29, 1989

Voice Characteristics of Female Speech and Their Representation in Computer Speech Synthesis and Recognition

Dieter Huber

Department of Information Theory, Chalmers University of Technology, Goteborg, Sweden

This paper presents a comparative study of female versus male voice characteristics in read speech, both in a global (intonation) and local (laryngealization) perspective. Swedish newspaper texts comprising a total of 2610 running words were read by two female and two male speakers of Standard Swedish. The two women produced consistently more intonation units per text, with on the average higher F0 onsets and offsets, shorter durations, steeper falls, a larger proportion of rising versus falling declination lines, and a markedly stronger tendency to time-align their intonation units with features of syntactic structure in the subsentence domain. Also, both female speakers made significantly more extensive and more varied use of laryngealization as a boundary marker than their male counterparts. The various voice parameters are described in quantitative terms with respect to their acoustical characteristics. Correlations and distributional properties are established in probabilistic terms for use in computer speech applications.

Full Paper

Bibliographic reference.  Huber, Dieter (1989): "Voice characteristics of female speech and their representation in computer speech synthesis and recognition", In EUROSPEECH-1989, 2477-2480.