[Update 10/98: Ralf Stephan has informed me that he is no longer involved with the JOS project. KevinRaulerson]
Currently working on an open-source SpeechRecognition which, through the use of Ptolemy, will be abstract enough that a Java version for JOS will pop out automagically when Ptolemy II is released. You can find a very early C++ implementation of a primitive and (somewhat) working system on my homepage.
News: Apr-98 - there will be soon a closed-source continous speech recognition for Linux, by a German student, laughing all those lazy companies in the face. An open-source system will stay important as a goal.