Microphone Array Processing Techniques for Automatic Lecture Monitoring

The gain in popularity of massive open online courses and other online educational lectures prompts the investigation of methods for automatically recording such lectures. While most previous systems in this area have utilized computer vision techniques for tracking, we take an approach utilizing microphone arrays for both recording audio and tracking lecturers. Different source localization and source tracking methods are tested, including cross correlation and beamforming methods combined with various state space model approaches. We investigate how certain constraints granted by a lecture setting may be used to influence our tracking models, and evaluate the relative strengths and weaknesses of several possible techniques. In addition, we explore characterizations of the lecture space that allow for the microphone array to work along with a separate camera to properly record the lecturer's movement. By using the audio to track lecturers we add flexibility to the system, but also introduce difficulties in consolidating information between the microphone array and the camera. Possible methods for communication between the two are addressed, and we again find that constraints imposed by the lecture setting may be used to resolve such problems.Ope