Training of coarticulation models using dominance functions and visual unit selection methods for audio-visual speech synthesis

Abstract

This paper presents results of training of coarticulation models for Czech audio-visual speech synthesis. Two approaches for solution of coarticulation in audio-visual speech synthesis were used, coarticulation based on dominance functions and visual unit selection. For both approaches, coarticulation models were trained. Models for unit selection approach were trained by visualy clustered data. These data were obtained using decision tree algorithm. Outputs of audio-visual speech synthesis for both approaches were assessed and compared objectively. Index Terms: audio-visual speech synthesis, talking head, coarticulation model