Expressive text-to-speech approaches. Productive homologies and irreducible discrepancies between natural and singing voice synthesis modelling.
Résumé
The core concern of this paper is the modelling and the tractability of expressiveness in natural voice syn-thesis. In the first part we quickly discuss the imponderable gap between natural and singing voice synthesis approaches. In the second part we outline a four level model and a corpus-based methodology in modelling expressive forms?an essential step towards expressive voice synthesis. We then try to contrast them with recurrent concerns in singing voice synthesis. We finally undertake a first reflection about a possible transposition of the approach to singing voice. We conclude with some program considerations in Research and Development for the singing voice synthesis, inspired from natural voice synthes is techniques