The trouble with these things to my mind is simply that the internet is a time unlocked system - the system's ability to change routes seamlessly plays havoc with systems that need to use can collation techniques, and maybe his issue is simply his voice, being low doesn't respond well to time alignment slides in real time. Maybe cutting the lowe frequencies would help, or maybe it wouldn't - you need to feed the system with audio least likely to be impacted. The reality is you cannot expect the system to be able to cope with a voice that repeatedly fails to get through. The people using his stream could be really skilled, with somebody really capable running them, or not - in which case the best they have are the zoom tools and the built in noise reduction systems they have. I really thought we had a proper application running and that was what you were talking about.
Realistically, if he becomes know for wanting special treatment, he just won't get it, and many users simply won't care if his audio is bad. especially if it makes him look foolish - they might not mind that at all. Can you at a distance replicate the problem if you use the system too talk to him? That's what I'd be trying.
