Java Speech has two parts: speech synthesis (i.e., generation) and speech recognition (which is what you want). Sun only provides the interfaces, but no actual implementation. There are a number of implementations for synthesis, but precious few for recognition. And those that do recognition are either commercial, or simply attempt voice-control of applications (which means recognizing very short phrases, not full sentences). So I'm afraid what you ask does not exist or is very expensive. Speech-to-Text is a very hard problem, and it's unlikely that anything will come out as open source. A listing of some existing projects and products can be
found here.