Comparison between two predicting methods of labial coarticulation

The construction of a highly intelligible talking head involving relevant lip gestures is especially important for hearing impaired people. This requires realistic rendering of lip and jaw movements and thus relevant modeling of lip coarticulation. This paper presents the comparison between the Cohen & Massaro prediction algorithm and our concatenation plus completion strategy guided by phonetic knowledge. Although results show that Cohen & Massaro perform slightly better, the concatenation and completion strategy approximates consonant clusters markedly better particularly for the protrusion parameter. These results also show the concatenation and completion strategy could be easily improved via the recording of better reference models for isolated