Speech Recognition

Like Version 3.x, a wizard guides you through the setup of your audio system and microphone, and then gives you a choice of diverse articles to train the software to your particular voice. The program starts with a default association between each word in its massive dictionary, and the sound it expects to hear when you read that word. If it hears something else during training, it will make a new association. It can therefore transcribe your dictation whether you hail from Spur, Texas, or Fargo, ND. Does it work? Ya sure, you betcha! But, as yet, the software trains itself to one voice only. Don’t expect to be able to plop down a microphone in the middle of a conference room and get a transcript of the conversation. No commercially available software in this category is capable of that yet. Like any software, expectations grow as the software matures.

You can create a new user profile for each person you want to use the software. Each person will have to undertake the training, and the software will create a unique file for each user, which can be selected during program start-up.

In addition to Texan, NaturallySpeaking can speak legalese, recognize citations and work with the Internet. So dictate res ipsa loquitur, nunc pro tunc, and so on to your heart’s content. A word of caution here: if the name of the case you are dictating is “Grapl v. Fahrquart,” it’s not going to get it the first time. But you can train the program to recognize these names so that it will recognize the case the next time you cite it. You also can control your Web browser (at least Internet Explorer and Netscape Navigator) with voice commands, including scrolling and navigating links and graphics. And you can control the Windows interface with such commands as “click” and “press,” which causes Windows to respond as if you had clicked or pressed the applicable control.

After training, click the microphone icon and just speak articulately as fast as you like. You can navigate the document with your voice. You can therefore format the document with voice commands while you are dictating, or later. You can dictate names, addresses, e-mail addresses, telephone numbers, regular numbers and URL’s (addresses to Web sites). You also can dictate outlines and numbered outlines. In both Word and WordPerfect, say “press enter” instead of “new paragraph” to advance to the next numbered paragraph. Correcting mistakes is as easy as saying “correct that,” or “correct (here insert the phrase the software mis-transcibed).” As in past versions, you should try to speak, and then correct in natural phrases, not unnatural phrases or one word-at-a-time.

Like its predecessor, you can dictate into the program’s own program window, and then copy and paste into another application using the voice command, “copy all to clipboard,” or you can dictate directly into the newer Windows versions of Corel WordPerfect or Microsoft Word. Much greater stability and support for a wide range of voice commands has made dictation directly into your word processing program the preferred method. And now, you can dictate into many more programs, and unlike version 3.x, retain the ability to “scratch that” and correct mistakes orally.

If you have had problems with voice recognition in the past because of a poor sound card in your computer (which is necessary for voice recognition), Dragon is now selling a Universal Serial Bus (USB) microphone, which bypasses your sound card. All the hardware necessary for voice recognition is built into the microphone. This is particularly useful for notebook computers that may skimp on the sound card. If the setup program tells you that your sound card is insufficient for voice recognition and you are unable to resolve the issue, the microphone will likely solve the problem.

The accuracy of Naturally-Speaking depends upon the quality of the training you give it in its initial setup. When you read the training article accurately, you are training the program to associate the sound you make in enunciating the words with the text it expects you to read. If you read the wrong words, or enunciate poorly, you are untraining the software. If, for example, you pronounce the word “casual” as “causal,” you will train the program to type “casual” each time you say “causal.” If you are a personal injury lawyer, this is not a good transposition. Hence, Rule 1: If a temp. came to transcribe a tape made from a dictation machine, you would presumably proofread her transcription. As good as it is, no voice-recognition software including Version 4 is as good as a good secretary that transcribes what you mean, rather than what you say. So far, Version 4 is only good at recognizing what you say. Therefore, proofread your dictation. Rule 2: Remeber Rule 1, and check your numbers carefully.

Win 95/98/2000

$59

Reviewed by Steve Schmidt, attorney, Singer Smith & Williams, Albuquerque, N.M.

Aug/Sep '00 Issue

PROS
More stable, more accurate and works better with more programs.

CONS
Memory hog, can’t recognize
multiple voices in a conference room setting.

VERDICT
A marked improvement over
Version 3.x, and a good, stable program to assist your transcription services.

  | Home  | 

|  Issue Archive  |  Resources  |  About Us  |  Contact Us  |  Subscribe  |

|  Subscribers  |  Advertisers  |

Updated 09/18/01
© Law Office Computing Magazine
www.lawofficecomputing.com
(800) 394-2626