Well, RakNet has a VoIP part. I planed to implement it, but you have to wait for the update after the final release.
p.s RakVoice is a new feature of RakNet that allows real time voice communication at a cost of only 2200 bytes per second at 8000 8 bit samples per second. It uses Speex (http://www.speex.org/) to do the encoding -Rak'Kar
I donīt really think directX can beat that
But I canīt promise anything.