I think the best way to solve this is by doing AGC first. This should make sure audio level is similar for all participants.