Tormenting the Dragon

If it isn't evident by the somewhat changed writing style so far this year, I'm doing very little typing anymore. I'm attempting to use my new—and still experimental—dictation method as much as possible, which relies on Dragon NaturallySpeaking 8.0 for transcription to the form you see here. I'm making good progress, but it isn't perfect yet. Really, it may never be perfect, but I'm not stopping now.

The recognition system relies a great deal on context, and I've noticed an impressive ability to draw distinctions between words on that basis alone. A simple example is the possessive form of the word it. The Dragon has yet to insert an apostrophe in that word where one isn't required; phrases such as it's pages were yellowed with age never creep into the rendered text. This is gratifying, especially since this error is so common.

As a point of interest, I noted during initial setup that the Dragon was poking its snout into Windows' My Documents folder to get an idea of my personal vocabulary. Among other things, it found my previous blog posts so the Dragon now has little problem recognizing my particular choice of words. It also attempted to search my e-mail client for more of the same, so if you install this—or similar—software you may want to ensure you have arranged things beforehand so it finds files that reflect your vocabulary and style. It's also possible to do this after installation, but sometimes it's easier to get certain things out of the way right off the bat.

But vocabulary is only part of the speech-recognition formula. My bad habit of slurring certain words and inventing contractions combined with the ever-changing surroundings when I'm dictating creates a more difficult job for the Dragon. Although a relatively quiet environment is important to speech recognition, the ability of the microphone to reproduce as clearly as possible the many frequencies of the human voice can't be overlooked. This, I think, has been a problem for some who have attempted to use a voice recorder, or an inexpensive microphone; the fidelity may not be adequate to detect the difference between syllables—and particularly consonants—that act as cues for determining the proper word. This is why so many reviewers repeatedly mention the need for a high-quality microphone. The other weak link, apparently, is the sound card, which may induce noise that confounds the recognition system. In any event, high-fidelity voice input is a necessary component of the recognition formula.

I must now return to the task of correcting certain mush-mouthed pronunciations so the Dragon can better understand what I mean, and not just what I say. Some of my slurred inventions are really quite comical when played back; it's a wonder they can be translated at all.


No comments:

Post a Comment