A mixed inventory structure for German concatenative synthesis

Abstract

In speech synthesis by unit concatenation a major point is the definition of the unit inventory. Diphone or demisyliable inventories are widely used but both unit types have their drawbacks. This chapter describes a mixed inventory structure that is syllable-oriented but does not demand a definite decision about the position of a syllable boundary. In the definition process of the inventory the results of a comprehensive investigation of coarticulatory phenomena at syllable boundaries were used as well as a machine-readable pronunciation dictionary. An evaluation comparing the mixed inventory with a demisyllable and a diphone inventory confirms that speech generated with the mixed inventory is superior regarding general acceptance. A segmental intelligibility test shows the high intelligibility of the synthetic speech.