Tracking an acoustic source in conference environment has attracted tremendous attentions for its applications in video conferencing, hands-free talking and many automated systems for surveillance.

The primary focus of this project is to track a single moving speaker in a reverberant room via time-difference-of-arrival (TDOA) measurements. The group is fascinated by the idea due to the large amount of real applications that would be depending on this algorithm. They construct whole system based on MATLAB and real-time simulation could demonstrate the result in simple yet professional manner.

To achieve single source tracking, it is essential to imitate physical environment at very first step. External impacts on human voice must be recognized and considered here for further processes. By using Time-Difference-of-Arrival (TDOA) measurements and a non-linear filter, it is possible to compute the position of the source at predefined time intervals, hence accomplishing tracking in continuous time. This algorithm can tolerate small to moderate noise in measuring and estimation processes, which in general is a good approximation of the real environment. In addition, the group would also give the public freedom to define most of the variables used in program. This will enable them to experiment with the closest possible descriptions of a simulated environment in the real sense. Performance of this algorithm will be displayed via MATLAB simulation and user-friendly interface ensures this algorithm can be easily adapted by people from other fields as well.

As video conferencing and other related applications are becoming increasingly popular, source localization is emerging as one of the most critical areas in achieving full-automatic control. The group hopes that this project could reduce the complexity of existing systems and realize real application in near future.

Samuel Leong, Pei-Ting Duan, Dr Ba-Ngu Vo