Abstract: In this paper, we describe a simple and fast way to build a 3D talking head which can be used in many applications requiring audiovisual speech animation system. The talking head is constructed from a synthetic 3D viseme dataset, which is realistic enough and can be generated with 3D modeling software. To build the talking head, at first the viseme dataset is analyzed statistically to obtain the optimal linear parameters to control the movements of the lips and jaw of the 3D head model. These parameters correspond to some of the low-level MPEG-4 FAPs, hence our method can be used to extract the speech-relevant MPEG-4 FAPs from a dataset of phonemes/visemes. The parameterized head model is eventually combined with a Text-to-Speech (TTS) system to synthesize audiovisual speech from a given text. To make the talking head looks more realistic, eye-blink and movements are also animated during the speech. We implemented this work in an interactive text-to-audio-visual speech system.