Feature selection for the prosody modelling of synthetic speech

Communication problems between human and computers are in the centre of interest of the experts from several domains. The quality of the synthetic speech is the main reason, why they must take into account not only technical equipment but also phonetic and phonological rules. Several years ago only the ability to understand was the main goal of speech synthesizers, whereas today synthesizers output must be as close to natural language as possible. It is a very difficult task to fulfil this requirement because the speech signal is the conjunction of several influences, physiological, technical and linguistic ones. The last two domains are the subjects of this paper.