There are known good and bad stances when you are giving a speech. The idea would be to follow the speaker and recognize his position using for example a kinect camera.

You could then get feedback either at the end of a dry run, or in real-time during the presentation.