Multimodal Speech Synthesis for Polish Language

Abstract

The main aim of this study is to describe the process of creating a multimodal speech synthesis system for the Polish language. It consists of two modules: a unit-selection speech synthesizer and a 3D avatar. The naturalness of unit Selection Speech Synthesis is achieved by the careful joining together of suitable acoustic units, covering the whole of the utterance. The main prerequisite for a unit selection system is a speech database. A speech corpus was constructed in such a way that its phonetic representation serves as the basis for the design of the cost function. This cost function is often decomposed into two costs: a target cost (how closely candidate units in the inventory match the specification of the target phone sequence) and join cost (how well neighboring units can be joined). The implementation of the new Polish voice was prepared in metasystem Festival. To obtain higher quality of synthetic speech the optimization of the cost function was conducted by applying a genetic algorithm. Additionally a prototype 3D talking head was built, containing 9 visems corresponding to Polish groups of phonemes.