I have probably 5,000 hours of experience with speech recognition applications (Dragon Preferred 7, 8, and 9).
I'm not sure, but I may have read that Microsoft and Dragon cooperated on the Win 7 speech recognition app??
A few clues based on my experience:
Don't get your hopes up too high.
Don't get caught up in the advice to tell you to upgrade your microphone, etc.
Live with the fact that you will have a certain irreducible percentage of errors.
Lower your expectations even further to the extent you are NOT using a powerful PC, such as a Core 2 Duo, Quad, i5, or i7. Speech programs need a lot of power to know whether you mean to, too, or two and whether you said dog, bog, or cog. They rely on context to do that and it takes a lot of analysis.
If you are going to write a document no longer than this post, don't bother with it. You won't save any time.
If you were to read a random 10,000 words out of a novel, it would help you get that onto the printed page much faster than simply typing it out. Overall, I'd say you could cut your total time by maybe 40%, after all hand corrections and proofreading.
You can expect probably 20 errors on a double-spaced 8 by 11 inch of random text. A little more or less, depending on your enunciation and equipment.
With a powerful machine, you can speak at fairly close to normal speed, but you have to enunciate like a newscaster--otherwise your error rate will skyrocket.
On this powerful machine, words will appear damn near as fast as you speak.
Don't obsess over "training" the application. It helps only to a point. If it isn't working tolerably well on your equipment after 10 hours of practice with a $10 headset, it probably won't every be worth your while.
