It mostly depends on who are "we", how much time, skills and resource do you have. You would need a lot. We don't have resources to guide you in this apparently work which might require years to accomplish at best. I can easily guide you in using available speech recognition engines though. :-)
[EDIT: answering a follow-up question]
I don't think you can get a source code of Windows (or Nuance, or any other proprietary) speech recognition engine.
I would try to find some relevant code for Linux. Please see:
http://en.wikipedia.org/wiki/Speech_recognition_in_Linux[
^].
Also, look in VoxForge, a repository for open source recognition languages:
http://en.wikipedia.org/wiki/VoxForge[
^],
http://www.voxforge.org/[
^].
—SA