In text-to-speech synthesis, linguistic and extra-linguistic knowledge is necessary to produce natural sounding multistyle and multivoice speech. Similarly, since a multispeaker speech recognition system is always confronted by the great variability in the speech signal, we need to know what factors contribute to this variability and how we can represent them formally. Where phonological variations are concerned, it is impossible to treat multispeaker phonological variability as a global and indivisible phenomenon. While linguistic constraints such as syntactic rules can reduce the number of allophonic possibilities, extra-linguistic knowledge is also necessary to better understand and formalize phonological variations which depend on the speaker characteristics. Actually, factors such as a speaker's geographic origin, his age, etc, determine to a large extent the allophonic choices and the deep phonological representation of each speaker.
The goal of this communication is to present an analysis of different strategies of allophonic choices in French according to the speech rate. From the analysis of a read corpus pronounced by four speakers, we underline that phonological reductions are linked to the speech rate and also to a speaker's phonetic style. We will then see if it is possible to differentiate classes of speakers according to their segmental variations at a given speech rate.
Cite as: Lacheret-Dujour, A. (1991) Phonological variations in read speech, reduction phenomena and speaker classes: do allophonic choices represent speaking style? Proc. ESCA Workshop on Phonetics and Phonology of Speaking Styles, paper 038
@inproceedings{lacheretdujour91_ppst,
author={Anne Lacheret-Dujour},
title={{Phonological variations in read speech, reduction phenomena and speaker classes: do allophonic choices represent speaking style?}},
year=1991,
booktitle={Proc. ESCA Workshop on Phonetics and Phonology of Speaking Styles},
pages={paper 038}
}