requent readers of this column know that I have a typing-related injury that flares up from time to time. It has been pretty painful lately, but I have managed to conquer my disability by using a new generation of voice dictation software from Dragon Systems, a 15-year-old pioneer in speech recognition technology based in Newton. Called Dragon NaturallySpeaking, this program is at the forefront of the natural speech recognition revolution.
I have used speech recognition programs in the past, but found them frustrating. That's because older programs, such as Dragon's DragonDictate, forced the user to pause between each spoken word. So I only used them when absolutely necessary.
The world of speech recognition changed this spring, when Dragon introduced NaturallySpeaking, a system that can recognize continuous speech. It allows you to simply dictate into your computer, and the words appear on the screen. The system is so good that it can even transcribe from a minicassette recorder.
NaturallySpeaking is superior in other ways. It has a 230,000-word dictionary, compared with just 30,000 or 60,000 for versions of DragonDictate. The bigger the vocabulary, the more words the system can automatically recognize. Many of those words are proper nouns. As a result, I rarely have to spell out a word.
Setting up the system takes about 30 minutes. After adjusting the microphone, the system asks you to read aloud as it flashes sentences on the screen. The computer then crunches numbers for another 30 minutes, adjusting to your speech, after which you are ready to dictate. You can improve dictation with additional training or by having it scan over your own documents.
For many users, the product's most important breakthrough is price. For years, Dragon's top-of-the-line software cost more than $1,000. NaturallySpeaking Personal Edition is $129. Driving this price is the dramatically larger market to which NaturallySpeaking appeals, combined with stiff competition from IBM.
Despite these kudos, users of NaturallySpeaking are sure to discover a variety of problems. The first is getting the program to start working. Few PC sound cards can handle the exquisite demands of NaturallySpeaking. Before you buy the program, check the compatibility list posted on the Dragon Systems Web site (www.dragonsys.com).
Users report varying levels of success with the product. For me, I get a ''speako'' - a misinterpretation of my words - in roughly every other sentence. If I take the time to correct these errors by repeating the words, rather than by typing the correct words, the system will learn from its mistakes.
Unlike DragonDictate, NaturallySpeaking cannot be used for completely hands-free operation. You need to be able to type, or at least to use the mouse. As a result, many people with severe disabilities use both NaturallySpeaking and DragonDictate at the same time. There are occasional conflicts with this combination, although DragonDictate 3.0 is supposed to resolve many of them.
NaturallySpeaking requires a powerful computer. Dragon suggests a PC running Windows 95 with it least a Pentium 133 processor and 32 megabytes of memory. In my experience, you need a Pentium 150 or faster (MMX helps a lot) and at least 64 MB of memory. If you want to have several applications running at once, get 96 MB of memory. Fortunately, RAM is cheap right now.
Another problem is compatibility with other applications, such as word processors. Although DragonDictate allows you to dictate into any application, NaturallySpeaking Personal Edition will dictate only into a mini-word processor. Based on Microsoft's WordPad, this application lacks key features such as word count and spelling correction (for things that you type yourself). You can select all of your text and copy it into another program.
Last week, Dragon announced two new products in its NaturallySpeaking line. The first is NaturallySpeaking Preferred ($169), which will allow dictation directly into Microsoft Word. The second is NaturallySpeaking Deluxe ($695), which can handle more than one user and will let you create macros that can control other applications. Unfortunately, the macro system is awkward to use - there's no way to record a macro by example, and there is no easy way to print all of the on-line documentation for the command language.
Dragon's main competition comes from IBM, which released its own general-purpose continuous speech recognition program, ViaVoice, in August. A big selling point is that it integrates with Microsoft Word. But this integration can backfire: if you aren't careful, ViaVoice will save the audio transcript of your entire dictation in your document file, increasing its size to 20 MB or more. In my experience, ViaVoice was not as accurate as NaturallySpeaking.
Although NaturallySpeaking comes with its own microphone, you can get better recognition by purchasing a Gold headset from the VXI Corp. in Rollinsford, N.H. (www.vxicorp.com, or 800-742-8588). I have been using the company's CTS 10 headset ($76). There is also a stereo model, the CTS-ST ($88), which is great if you want to play CDs in stereo through your computer between dictation sessions.
VXI also makes a box called the GSA 60V ($116) that lets you use your headset with both your computer and your telephone. If you ever need to type while talking on the phone, you should absolutely be using a headset. The telephone box only works with the monaural headsets.
NaturallySpeaking works under both Windows 95 and Windows NT. Dragon hopes to develop some sort of programs for Macintosh users, but realistically, if you need speech recognition you should probably switch to Windows. And if you are in the market for a new PC and want to run this software, be sure to get a machine with it least 200 MHz Pentium MMX and 128 MB of RAM.
Technology writer Simson L. Garfinkel can be reached at plugged-in@simson.net
Dow:
