Dragon NaturallySpeaking Computerized Voice Transcription

standingtopAI and Robotics

Nov 17, 2013 (4 years and 8 months ago)

74 views


Dragon NaturallySpeaking Computerized Voice Transcription


This article is “written” using direct voice-recognition into my computer. I did not have time to
“train” the software installation to more perfectly recognize my own voice and accent - I’m using
it straight out of the box. To give you an idea of the accuracy of NaturallySpeaking, neither I nor
my editor will correct any typos or odd grammar in this section of my column. Any errors in this
section are voice transcription problems

Speech recognition has been around since about 1997 when IBM shipped a version of voice-
recognition for its OS/2 operating system. Speech recognition never quite caught on because
earlier versions were not very accurate, required substantial training, or somewhat difficult to
correct mis-recognized text, and tended to be somewhat slow on older computer hardware, it
never quite caught on. For many people, myself included, typing tended to be faster.

Version 10 of Nuance’s Dragon NaturallySpeaking Professional corrects all those problems and
is a true pleasure to set up and use, even when you have a “distinctive” accent like myself.
NaturallySpeaking did not require any significant training to recognize my voice almost
perfectly. In fact, they’re probably fewer typos in this week’s column because I’ve been dictating
it rather than typing it. The newest version of NaturallySpeaking can actually be said for many
different English accents, including East Coast English, British English, Spanish-accented
English, Australian accents etc.

Be sure to carefully correct any voice-recognition text, however. Several years ago, I dictated
a lengthy brief for the court using an earlier version of NaturallySpeaking. I thought I carefully
corrected that brief, but Judge Harold Brown, then our senior Superior Court Judge at Kenai, was
sufficiently astute that he inquired on the record whether I had dictated that brief using voice
recognition software. Obviously, I fail to catch some necessary corrections. Luckily, editing,
particularly inserting text, is very easy. Just place your cursor where you wish to insert a word, a
phrase or a sentence and began talking.

One of the more productive features of NaturallySpeaking Professional is its capability of
recording a voice macro command. For example, I could easily create a command that would
bring up letterhead already dressed to an appropriate party or attorney or bring up standard form
contract clauses when drafting lengthy legal documents. To get to that point requires quite a bit
of substantive work, mostly identifying appropriate form clauses, such as arbitration
requirements, and then setting them up as a voice macro command.

Voice recognition software tends to work best with a very clean digital audio signal and with a
very fast computer. Although Nuance packages an analog headset/microphone with
NaturallySpeaking software, I found that a USB Digital Signal Processing (DSP) headset results
in a cleaner digital input that’s recognize more accurately. The DSP headset includes hardware
that is optimized for converting analog voice into a digital USB signal.

A fast computer also helps reduce the delay between the spoken word and its appearance on your
computer screen. Nuance appears to use an approach pioneered by IBM in the late 1990s - a
statistical model of how frequently specific words are within reasonable proximity to each other
in average or more sophisticated speech patterns. This linguistic model then corrects any
ambiguous voice-recognition by finding the best statistical match between clearly recognized
words in close proximity with each other. That definitely increases recognition accuracy, but
requires some additional computer performance. Although a slower computer will still get the
job done, you may find it frustrating.

The professional version of Dragon NaturallySpeaking costs $199 and works with both 32-bit
and 64-bit versions of Windows XP, Windows XP x64, Windows Vista and Windows 7. If you
do a great deal of typing and are comfortable dictating at a reasonably fast clip, then
NaturallySpeaking should be very productive technology in your office or business. In fact,
writing this week’s overdue column is moving along so quickly that it feels as though I’m
slacking off.