What NaturallySpeaking Can’t Do
Even with NaturallySpeaking, your computer’s capability to understand English is more limited than what you can reasonably expect from a human. People use a very wide sense of context to figure out what other people are saying.
You know that the teen behind the counter at Burger King means when he asks, Wonfryzat? (That's fast-food-employed teenspeak for Do you want fries with that?) You'd be completely confused if that same teenager walked up to you at a public library and asked, Wonfryzat?
NaturallySpeaking figures things out from context, too, but only from the verbal context (and a fairly small verbal context at that). It knows that two apples and too far make more sense than too apples and two far.
But two- to three-word context seems to be about the extent of the software's powers. It doesn’t understand the content of your document, so it can’t know that words like Labradoodles and Morkies are going to show up just because you’re talking about dogs.
Consequently, you can’t expect NaturallySpeaking to understand every form of speech that humans understand. In order to work well, it needs advantages like these:
Familiarity: Each person who dictates to NaturallySpeaking has to train it individually, so that NaturallySpeaking can build an individualized user model. So NaturallySpeaking can’t transcribe the voice mail that other people leave for you.
Identification: Each time before you start dictating, you need to identify yourself so that NaturallySpeaking can load the right user model.
One user at a time: NaturallySpeaking loads only one user model at a time, so it can’t transcribe a meeting during which several people talk, even if it has user models for all of them.
Constant volume: You can’t plunk a microphone down in the middle of the room and then pace around while you dictate.
Wear a good microphone (like the one that comes with NaturallySpeaking) and position it the same way every time you use it.
Don’t mumble or let your voice trail off.
It does a pretty good job with accents, though, as long as you’re consistent.
Reasonable background noise: Humans may be able to understand you when your favorite drummer is blasting away on your speakers or the blow dryer is on. They may be reading your lips at least part of the time, and they can guess that you're probably saying, Turn that thing down! NaturallySpeaking lacks in the lip-reading department, as well as in the capability to make obvious situational deductions.
Reasonable enunciation: You don’t have to start practicing Moses supposes his toeses are roses, but you do need to realize that NaturallySpeaking can’t transcribe sounds that you don’t make.
Standard turn-of-the-millennium English prose: If you want to be the next James Joyce, stick to typing. You can have some fun by trying to transcribe Shakespeare or things written in other languages, but it isn't going to work very well (unless you do some really extensive training). On the other hand, NaturallySpeaking is just the thing for writing books, blog posts, reports, short stories, and letters to Mom.