The particularity of this study resides in the spectral representation space. Resonances are defined as the formant frequencies but their order is modified so that spectral continuities are respected: the affiliation of these resonances to certain cavities may produce formant crossings for some sound pairs. Analysis of natural vowel-vowel transitions uttered by two male French speakers was performed, using cepstral and pitch synchronous LPC at the closed phase of the glottis. Resonance trajectories were obtained by the interpolation of target resonances with emergence functions obtained by the Temporal Decomposition Technique. Vocalic resonance targets were extrapolated using a continuity criterion and a one-sound one-target rule. Based on this representation, rules will be given, for describing coarticulation of [b], [d] and [g] in VCV context. These rules have been implemented and tested by a rule-compiler: COMPOST. The synthesizer is an extended version of Klatt's parallel configuration: this includes a model of Fant's voice source.
Cite as: Guerti, M., Bailly, G. (1991) Synthesis-by-rule using compost: modelling resonance trajectories. Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991), 43-46, doi: 10.21437/Eurospeech.1991-8
@inproceedings{guerti91_eurospeech,
author={M. Guerti and G. Bailly},
title={{Synthesis-by-rule using compost: modelling resonance trajectories}},
year=1991,
booktitle={Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991)},
pages={43--46},
doi={10.21437/Eurospeech.1991-8}
}