I’ve been using voice recognition software for about two years now. First IBM ViaVoice, then Dragon NaturallySpeaking. Now, I wouldn’t say my recognition is perfect (especially when writing fiction), but it has improved to the point where I can get a page’s worth of text down faster with Dragon than by typing (and I type 80 wpm). For me, there were a couple of things in particular that made noticeable leaps in recognition accuracy.
Better Microphone — A few months back, I purchased a USB Plantronics Headset, and it was quite an improvement over the crummy little microphone that came with Dragon NaturallySpeaking.
Enunciate Carefully — In everyday conversation, we have a tendency to slur words together or mumble. While people may be able to figure out what you’re saying, chances are your computer won’t.
Use Complete Phrases or Sentences — This is sometimes hard for me, especially when I’m composing at the computer, but, much like the human ear, voice recognition software works better when it has a string of words to work with.
Quiet in the Room — Recognition improves noticeably when outside noise is eliminated. Televisions, family members, stereos, even if they’re not directly in your room, can present difficulties. I find that with my new headset I can “cheat” by playing MP3s through the ear phones.
Here are some good links with more tips for improving voice recognition: