Expressive speech synthesis for Czech limited domain dialogue system — Basic experiments

This paper describes a development of limited domain expressive speech synthesis for the Czech language. Our current speech synthesis system is based on unit selection methods and produces high quality speech in a neutral speaking style. This work focuses on modifications made in the synthesis algorithm to integrate expressivity into generated speech. There is also introduced a listening test, which should prove or disprove that expressivity in limited domain speech synthesis could be modelled by so-called communicative functions. A comparison between neutral and expressive speech synthesis is presented. This paper also briefly discusses the process of an expressive speech corpus recording and its annotating using the communicative functions by means of another listening test.