Development of Expressive Speech Synthesis for Czech Limited Domain Dialogue System

The paper describes several stages of the development of Czech expressive speech synthesis for limited domain dialogue system. The first phase includes the creation of a corpus containing natural human-machine dialogues, gathered using a WoZ technique. The recording setup and the process itself are described in the first part of the paper. In the second phase, an expressive corpus for speech synthesis is created. The theoretical background of the expressive TTS corpus design, the technical measures taken to ensure the corpus appropriateness and the corpus recording process are presented. All this research is being done within COMPANIONS project scenario, where one of the aims is to develop a Czech dialogue system allowing elderly people to reminiscence about their photographs.