This is really cutting edge stuff you are requesting here.
I did a lot of research into the state of AI when writing my master thesis, and found that voice recognition is a long way off yet. That is not to say it isn't possible to do some basic stuff. You can test it out for yourself with windows speech recognition. The speech recognition API for the Kinect (which is really the same) is available for use by developers, and might be worth looking into. http://www.renauddumont.be/en/2012/kinect-sdk-1-0-5-reconnaissance-vocale
This api is probably not going to allow you to have natural conversations however..
If this is what you are looking for, you are in for an even larger challenge. Google is doing a lot of work in the field: http://research.google.com/pubs/NaturalLanguageProcessing.html
There is a lot of research going on in this field, and there are also open source solutions you might want to look into. Keep in mind however, that the main challenge here is mapping recognized language to in-game results. You need a database of viable responses, voice-acting for all these responses, etc.
To summarize: this is a large to massive challenge, depending on your goals. It is cutting edge AI research, which would require a highly technical team to work with the project. It is possible to make something like this, but expensive and/or difficult.