Estimation of speaker position using audio information

Real-time conversational video telecommunications services, such as video-conferencing, are becoming ever more important as a substitute for face-to-face meetings. One of the perceived weaknesses of existing services is the picture quality achieved, especially around the face of a speaker. A possible solution would be to identify the location of face, which is then transmitted at a higher quality than the rest of the picture. In this paper, we present a new technique for identifying the face using an array of microphones. As opposed to other techniques proposed so far, which make assumptions about the content of the video material, the idea relies on the estimation of lip position based on the audio processing from the speaker's speech,. Once this estimation is performed, then a two or possibly three stage quantisation on video information will facilitate the compression of the subjectively more important parts, i.e. the face of a speaker with lower distortion. This new technique, which is compatible with all existing video compression standards, is much cheaper and easier to implement than previous techniques.