I would probably go with a python voice recognition package, and have the python program listen for voice commands and send corresponding 'reaction' commands to Oculus via the telnet interface.
Haven't tried it, but this voice package looks to be a good one, as its based on the Dragon NaturallySpeaking libraries:
Once you get that going, you can add the simple python code needed to talk to and command Oculus.
Example python programs that connect to the Oculus Java server and are able to completely control the robot are
There are probably different good ways to do it using other languages as well, but python is a good one, since it tends to be quick and easy, there are so many good libraries available, and it's somewhat cross-platform.