Abstract
The progress in multimedia, wireless communication and internet, has curtailed distance between people. These progresses have brought up new topical subjects like video conferencing, virtual agent and email reader. Movement of large quantity of audio and video information is usually involved in such applications. However, limited and finite bandwidth poses some restrictions on their usage. Measures must be taken to overcome these restrictions. In this thesis, we focus on establishing an audio-to-visual conversion which can generate visual lip movement parameter with audio signal alone to greatly reduce the transmission bandwidth required for transmitting the video signal. First, a system is developed to extract audio and visual data from recorded AVI files in a controlled environment to build an audio-visual database. GRBF and GMM are then examined for performing the audio-visual conversion. Moreover, different facial expressions are considered. A new method incorporating DTW to generate modified audio feature vector accounting for facial expression for use in the audio-visual conversion is proposed. Extensive computer simulations are conducted and the results demonstrate the effectiveness of the proposed method.