Issues in corpus based speech synthesis

This paper reviews the main components of large database, concatenative speech synthesis systems. The design of these components can have an important influence on the system performance (in terms of memory and CPU) and, more importantly, on the speech quality itself. The main design questions to be addressed concern the speech database, the waveform selector and the waveform concatenator. The paper describes the key issues involved, with reference to how we tackled these issues in our laboratory systems.