Seems like the more you prepare the better you get. 

Oculus Rift would be perfect for this, just simulate the environment and practice by talking to scripted people... getting deeper into that, you (I use the term in general not you specifically) can generate an algorithm based on your voice patterns and time and movement to coach you and simulate reactions from the audience. That would take a lot of understanding and audience modeling but that has already been done it just needs to be turned into numbers.

Audience modeling sounds SciFi but I still love this idea, Nathaniel. :)