First European Conference on Speech Communication and Technology

Paris, France
September 27-29, 1989

A Female Voice for a Text-to-Speech System

Inger Karlsson

Department of Speech Communication and Music Acoustics, KTH, Stockholm, Sweden

There is a great need for better voice quality in text-to-speech systems. Today, only mechanically sounding male voices can be produced. The lack of success in producing a better sounding voice quality has been due mainly to a lack of knowledge of the voice source. We have also needed a good voice source model and an implementation of such a source in a text-to-speech system. The LF-model for the voice source has given us a tool for a description of the voice source dynamics in speech. The implementation of this source model in our text-to-speech system raises opportunities for synthesis with better voice quality and with different voices. In this paper the work on a female voice is described. The voice source dynamics in sentences and in different stress environments are studied. Acoustic parameters for a female reference speaker are compared to the male synthetic voice. These data are compiled into rules for synthesis and the results of these rules will be played at the conference.

Full Paper

Bibliographic reference.  Karlsson, Inger (1989): "A female voice for a text-to-speech system", In EUROSPEECH-1989, 1349-1352.