It's not new, the Sphinx project from the Carnegie Melon University is a real amazing project.
I did a test drive of it a year ago as i was looking for something for voice recognition in my project.
This software is really something different from anything you saw before.
It's not only trying to recognize one sound envelope but it tries to do one semantical analysis.
That's really GREAT to have one machine that obeys and understand the mean of sentences.
So if you are running Linux, give it a try, it's really worth to work on it.
I don't use it because unfortunately the recognition is only for American/English language, so no luck to use it in other languages. Well it could be but requires a lab with really heavy serious material to make a model for each language.
It's a huge code and time is needed to compile but after that you touch something apart. Try it, it's a must