From what I can tell, the speech API seems more focused on higher-level computer synthesized voices (kind of like you get on some GPS's) and voice recognition. Since you just want a voice chat...