I, too, am interested in the prospect of having voice recognition and speech synthesis integrated into this project.  From what I understand, CMUsphinx is a backend, much like mythtv has a frontend and a backend.  So all that would need to be done is create a small program that is able to tell the backend to listen for voice, then be able to parse the voice it captured to a command, then be able to pass that command on to whatever program was called in the command.  So, say, "Call Fred" would tell Asterisk to start up, and it would pass the argument "Fred" to it to call Fred automatically.  I was thinking about toying with the notion of integrating this somehow.  Now that it seems that other people are  engaged in this, maybe I could lend a hand?