Pandora's current voice uses the Java Speech API using FreeTTS. It basically synthesizes anything text the Pandora outputs. The voice plugin works in the GUI application, but it does not work from the applet. I'm not sure how to get voice working as a web page, so this may be something that has to be limited to the full application.
Eventually a voice that stores sounds in Pandora's memory and vocalizes similar to a human, is desired, but for now FreeTTS works well.