Been playing around with speech recognition on cloudgarden. At the moment I have an eval version and not to sure if it's worth the upgrade.
The primary draw back with cloud garden is that is not a pure java implementation of the java speech API. Conceding the fact that it would be pretty hard to build a pure java implementation you might concede that it's ok to just make a wrapper for a native implementation. That is in fact what cloud garden is. Unfortunately what they have not done is provide their own native implemenation, you have to buy an implementation from a third party.
Once again you might concede that a simple old wrapper for a third party native implementation is acceptable if it was reasonably priced. Unfortunately it is not.