And if players start turning to auto-matching voice services (Curse Voice and Razer Connect did this at one stage), they'll go and threaten bans for all.
Look into SteamRE for a good primer on interfacing with Valve chat services, although if you want gory details of server architecture I don't think they've ever written about that in detail.
EDIT:
Are there any articles like this on the architecture of Dota 2 or Steam?