Voice Recognition

The Bottom Line: It Ain't Ready for Prime Time

By Bruce L. Dorner

Editor's note: This is the latest of Bruce Dorner's occasional missives on his adventures with voice recognition software, gossip and trends.

AS A PRACTICAL matter, I compare voice recognition software programs to a base line secretary. They are generally very good, but occasionally make a few typographical or syntax errors. Part of the problem is the flexibility you give the secretary. If they know "what you meant to say," and make the change, then you don't realize you even made a mistake. Computer voice recognition doesn't have a "you know what I meant" button.

With that out of the way, I'll come straight to the point. I've used voice tools for several years. I used the original Dragon system when it was hardware based in a 386 computer. It was so expensive I had to borrow a unit for a demonstration lecture. It was very challenging using discrete speech with - every - word - separated - by - a - pause. In those days, we talked about a $15,000 hardware and software system. Today, the software cost ranges from less than $100 to about $600, depending upon the feature set.

I've used a few different flavors of voice tools. I keep coming back to Dragon NaturallySpeaking. Maybe it's a bias from the early days where the first tool I used set the benchmark for all others. Maybe it's just that I'm too lazy to keep retraining myself in new voice commands. Maybe it's just that it works for me!

For now, I'm using Dragon Naturally Speaking Professional 4.0 on my portable computer and Dragon Naturally Speaking 5 Professional on my desktop. Why the different versions? Well, my portable is only a P-II 300 MHz, and my desktop machine is a P-III 733 MHz. The newer version of the software requires a little more horsepower and I'm just too frugal to buy a new portable computer (the joys of a solo practice.)

Dragon works from within my word processor, WordPerfect. (Yes, it also works from inside Microsoft Word.) I don't want to dictate into a scratch pad and then transfer text. I want it working from within my word processor and I want it to understand editing commands from inside WordPerfect. It's nice to have it running from within Internet Explorer and other software applications too. In fact, most programs which enter information by typing are at least baseline compatible with Dragon Naturally-Speaking. Again, the critical feature to look for is the ability to dictate directly into the running software application, not the speech software itself.

Next, the Professional version allows me to create my own voice commands to activate my already existing WordPerfect macros. For example, If I say, "take a letter," my screen displays my letterhead, enters today's date, and moves to the insertion point to enter the inside address by voice or keyboard. I have another macro, "sign that," which puts my closing and signature block on the screen.

I could go further with the software, but never seem to have the time to really tune it up all the way. In fact, that's the big problem with voice tools. They all require a major time commitment to become truly proficient. But, isn't that true for anything in which you wish to become proficient?

This raises the big question of voice tools. You only get what you put into it. You need to spend a great deal of time training, training, and training. I don't care that the manuals claim you can be up and running in 10 to 15 minutes. Yes, you can be up and running, but that's like claiming you can understand any software package in 15 minutes. With voice, you need to correct the errors that will inevitably appear. You can't just type over them. The Dragon software learns by voice training, or the software will repeat your errors believing that the wrong word or phrase is what you really intended.

For the next part, let's get real with the day-to-day stuff. My confession. I play with voice tools. I don't use them regularly. I generally type faster than 60 words per minute. Therefore, I don't have a great incentive -- except for intellectual curiosity -- to get voice running as my primary means of entering information into the computer. In fact, I'm typing this article rather than dictating it. I have good intentions to use Dragon as my primary means of text creation, but never seem to deliver on my own promise. Now that I have a fast machine on my desktop, I've made another promise to try again to use Dragon NaturallySpeaking ("DNS") as my primary text entry tool. In a few months, I'll let you know if I kept the promise. However, if you are a two-fingered typist, then computerized voice dictation may be your salvation.

Hardware

As to hardware, Dragon requires a current computer. Although the manufacturer has lower threshold requirements, I recommend more horsepower. I run DNS Pro 4 on a P-II, 300MHz machine with 128 MB RAM. It is passable, but not great. I suggest a P-III machine of at least 600MHz and at least 128 MB RAM. It is better if you have 256 MB RAM, and 512, although overkill for most, makes DNS really fly. A fast hard disk is also a major plus. Voice software sucks up power, but produces better results with faster chips and systems. If you want the real sizzle, you need to pay one of the "value added resellers" who can take a pile of your documents and electronically "read" them into a computer file that will customize your vocabulary to your particular form of speech or area of law. They also assist you with macro creation and numerous other support services. The accuracy and ease of use increases if you spend the money for these services. However, if you aren't serious about using voice as your primary tool, save your money. Costs for the after market assistance can range upward from $1,000.

Microphone

Another often overlooked component is the microphone. Few people realize that a USB microphone (one that plugs directly into the USB port on your computer) virtually bypasses the sound card. This increases the recognition factor and avoids a lot of set-up tweaking and adjustment.

Buy a USB connected microphone! I have tried hand-held microphones and used the desk-top array microphones. Although they work, they are not as good as a headset microphone. If accuracy is important, and in the law office I believe that to be paramount, use a headset microphone even if it is less convenient. Your personal results may vary, but from my experience, keeping the microphone a fixed distance from your mouth goes a long way toward consistent accuracy.

For the record, after I installed DNS Pro 4.0 on my portable, I spent about 45 minutes training the system. I then dictated a two-paragraph letter into WordPerfect with 100 percent accuracy. I was amazed and impressed. Maybe it was the USB microphone. Maybe it was the voice inflection. Maybe it was the cycle of the moon. The result was amazing. My accuracy has gone down from there to about 96 percent. But when it does not understand what I say, it makes some very strange substitutions. The critical part of avoiding the repetition of errors is to use the voice correction mode, not the keyboard, to change the errant words. Yes, I know I said it before, but it is so critical that it bears repeating. In my experience, it takes longer to make corrections by voice than by keyboard and mouse.

Is voice recognition ready for prime time? Yes, with certain caveats. First, you have to have the right hardware. Second, you have to be patient and continue to train the software much as you would with any new inexperienced employee. Third, you need to set reasonable expectations. This is not the bridge of the Star Ship Enterprise, and a true talking computer does not exist in the law office environment. Can you be productive using voice tools? Yes, if you take your time, learn your lessons and use the tools every day.

Gossip and Caveats

Dragon NaturallySpeaking is owned by Lernout & Hauspie. They also produce the voice recognition product, Voice Xpress. Microsoft invested about $60,000,000 in L&H and owned about 7 percent of the company in the summer of 2000. L&H also owns Dictaphone. Clearly with Dragon for software and Dictaphone for hardware it's not hard to imagine the power this combination can bring to the market. Now, couple that with the market dominance of Microsoft and just imagine where this all goes.

Unfortunately, near the end of 2000, L&H filed for bankruptcy protection. Most of the problems are reported to evolve from the European and Asian operations of L&H. As this article is being written in January of 2001, it is anticipated that L&H will survive and emerge a stronger company. John Duerden, the head of L&H, also just resigned.

Under any scenario, I believe that Dragon will remain the dominant player in the voice recognition market, whether it remains part of the L&H family or is otherwise spun off.

An update on this situation will be presented during the voice recognition session at the ABA's TECHSHOW 2001, March 15-17, at the Sheraton Chicago Hotel and Towers. (Info: 800 888 8300 x 9035).

Bruce L. Dorner is a Londonderry, N.H. solo practitioner.