The problem has a number of domains; but possibly the biggest will be "fairness".

To minimise favouring any particular audience member the audio source (microphone) or sources must be located a good distance from the entire audience (maybe a rooftop) in an attempt to make the average distance to each audience member proportionally similar.

Then, peak audio response, could potentially be done by sampling input wave forms and measuring average audio response over a given time period. Because it's total audience energy you're looking for, not just a peak sample that happened to occur at some point in time.