Rudimentary voice-recognition software is fairly easy to write if you're working with a limited vocabulary in a non-critical environment. This is especially true if you can have the end user record the command vocabulary at setup. Of course, you wouldn't use simple methods if you need a broad voabulary (like a voice-controlled typewriter) or in safety-critical applications.
Give a man a fish, he'll eat for one day. Teach a man to fish, he'll drink all your beer.
Cheers, Jeff (SCJP 1.4, SCJD in progress, if you can call that progress...)
(There is a list of implementations of the Java voice stuff on the Sun website.)
http://www.cloudgarden.com (30 day free trial and only $16 for a personal license or $420(I think) for developer license for 1 developer and up to 1000 distributions.) They have some example code which can help you.
I couldn't manage to download the IBM ViaVoice free trial. (A License there is over $4000 and you have to wait 4-6 weeks just to get one).
I am experimenting with the cloud garden implementation now. Using the Microsoft SAPI5 voice engine which you can get (for free) at http://www.microsoft.com/speech/download/sdk51/. You need to train it to understand words, which can be a little annoying.
The accuracy with this hasn't been so good, but I am using the low-end microphone that came with my Dell.
The coding is a little complicated, but cloud garden's example code together with the java API should get you started fairly quickly.