Usable Voice Recognition Technology Arrives

This column will be written completely without the use of a computer keyboard or a dictating machine. I am dictating these words directly into the computer using a new software program called NaturallySpeaking from Dragon systems. Several years ago at the State Bar Convention we demonstrated a voice recognition system that had been used by one of my former partners for about a year. He demonstrated the way that voice commands could be used to control a computer system and do simple dictation. The dictation module left a lot to be desired but, all in all, it was an impressive demonstration that indicated a future for broader voice recognition technology. In the last year the extensive research done by several major companies has resulted in the bringing to market of a number of products which incorporate various kinds of voice recognition technology. Until Dragon Systems brought NaturallySpeaking to market, however, actual functional voice recognition technology was an elusive dream. Now, however, it is fair to say that voice recognition technology is ready for prime time.

In the last year you may have noticed that when you call Telephone Information in various major cities you are first plugged into a voice recognition system. If the computer can recognize the name of the person you are trying to contact the system works reasonably well. If the system does not recognize the name, however, it becomes difficult to deal with. That is caused in part by the difficulty of understanding words in different accents and tones of voice. The vocabulary necessary to run a telephone directory system, however, is relatively small and the computer is able to guess at the name and get it right a good part of the time. That is not so when you are dictating sentences using a vocabulary of 30,000 or more words. Here the complexity is enormously multiplied and the possibility for error is extremely high. Two of the most popular ways of dealing with that problem made voice recognition difficult and unpleasant in the context of dictation. The first was the necessity to use what is called "discrete" speech and the second is the necessity to train the system to recognize your voice. Discrete speech is the careful enunciation of each word in a staccato style. It is easy to do for a short period of time but once you get involved in dictation and you are starting to think about what you are saying it becomes harder and harder to do. And, as a result, the error rate increases dramatically. One way of dealing with that error rate poses the second major problem that the older voice recognition systems had. If you talk long enough to a computer and you constantly tell it when it makes a mistake it will eventually build a pretty reasonable capability for recognizing your voice. The problem is that this training may take many hours or days and is incredibly boring. Even then the number of errors in the older systems was so high that there was no real time saving. Of course, people with disabilities or severely challenged typing were willing to put up with that. The vast majority of people, however, required a much more comfortable and natural system and that is what Dragon systems has produced. NaturallySpeaking is very easy to use, does not require discrete speech and does not require extensive "training" in order to recognize your voice. I installed the program on my computer in just a few minutes. It took about 20 minutes to do the initial training so that within a half an hour, I was dictating comfortably and with high accuracy to my computer screen. The Dragon dictate NaturallySpeaking system, comes with a high-quality microphone and the software program installed on a CD-ROM. It will run reasonably well on a slow Windows 95 computer such as my Toshiba laptop with a 120 MHZ chip and 48 megs of RAM. It is not fair to say, however, that it is comfortable to use it on that machine. On my desktop machine, however, it works incredibly well. I have a 200 MHZ Dell Dimension with 64 megs of RAM and a good sound card.

I have noticed some interesting aspects of use of the voice recognition system. I have always been a pretty good user of the dictating machine -- able to produce sensible prose in rough draft form pretty quickly. But my dictating sufferers from my deteriorating memory and I often cannot remember what I said several paragraphs before. I am often not at all sure that I included a particular point that I wanted to make and so sometimes my dictating is fairly scrambled and repetitive. Being able to dictate to the computer and see the text come up on the screen really overcomes that problem. It is easy to read what you have written and to move comfortably into the next topic. If you are one of those people, as am I, who laborers mightily over word choice and style, dictating directly to the computer is a gift from heaven.

Unlike many of the other voice recognition products, NaturallySpeaking does not allow you to dictate directly into your usual word processor. When the Wall Street Journal reviewed this product that seemed to the reviewer to be a big problem. I have not found it to be a problem at all. It is incredibly easy to cut and paste between the DragonDictate word processor and both WordPerfect and Word for Windows 95. In fact, the DragonDictate word processor is very easy to use because it is designed specifically to be used with the voice recognition software. The icons and menus relate only to that and as a result it is very uncluttered and simple.

The DragonDictate NaturallySpeaking system that I am using is version 1.0. It has only been out for a few months but it has received rave reviews from almost everyone. I just got a press release indicating that they have improved the system and it is soon to be released in a version 1.2 and of course I will report to you on that if it is significantly different from this system. For now, however, I am completely comfortable in recommending that you try the DragonDictate NaturallySpeaking voice recognition system. It is relatively inexpensive and the price appears to be dropping in the national mail order marketplace. If this is something that appears to be a solution for you this particular product is worth buying. I have used some of its competitors and I have not seen anything, including the previous version of DragonDictate, that is really usable in the workplace. IBM has announced a new version of its VoiceType software that it claims competes with NaturallySpeaking. I haven't tried the new version of the IBM product but the old version is little more than a playful diversion and a glimpse into a future that is now here. NaturallySpeaking works and it works well. The only caveat that I have is to remind you of my experience with my little 120 MHZ Toshiba. I have a feeling that anything less than a 166 MMX Pentium machine with 32 megs are RAM is likely to have a high enough error rate that your experimentation with voice recognition technology may be frustrating and cause you to give up. That would be too bad because this is truly important stuff!!! Voice recognition technology will fundamentally change many aspects of our work life and courtroom experience. Imagine, if you will, a system that translates into a different language. That dream seemed beyond reach only a few years ago. It is now certain to be reality in the not far distant future. The IT revolution is just begining, folks!!