Modifying the S2 low-level speech synthesis engine for more intelligible output

This paper gives an overview of the latest improvements that were implemented into the S2 low-level speech synthesizer. The S2 is a Slovak CHATR style corpus based unit selection engine. The improvements were based on studies for non Slovak speech analyses and syntheses. These clarified criterions of importance of CHATR parameters for the human sound perception and their impact on the quality of synthesized speech respectively. The synthesizer was also extended of a new female voice and a diphone style synthesis add-in which both are expected to produce a more intelligible output.