Visual speech synthesis for Polish using keyframe based animation

This paper is about designing visual speech synthesis system for Polish. Xface toolkit with keyframe interpolation based animation was chosen as animation method. The paper describes designing the “Karol” face model and Polish visemes. The idea of using half-visemes was proposed for synthesizing fast visual speech, and it was verified during testing. Finally this idea was combined with omitting selected keyframes. Subjective tests showed that the visual speech generated by the proposed system was found quite natural (3.9 in MOS scale) and with good audio-video synchronization.