Sphinx seems to be only real option for Java speech recognition. Documentation is sparse and it requires a high-level of domain knowledge. I used their example of a starting p
