No longer must you be a star-ship captain to talk to your computer productively. Voice recognition technology, which converts the spoken word into text and lets you control a computer's actions with verbal commands, is now a practical option for the general medical community, particularly as a low-cost way to speed up your charting. If you already dictate your notes, voice recognition programs can help you do so less expensively, more accurately and without the turnaround time involved with using a transcription service. (For a description of how one family physician is using this technology to make his charting more efficient, see “Voice Recognition Software: A Tool for Encounter Notes,” February 1999.)

But once you decide to make the leap into voice recognition, how do you know what program to buy? Reviewing three of the leading packages — Dragon Systems' Dragon NaturallySpeaking, Lernout & Hauspie's Voice Xpress and IBM's ViaVoice 98 — has shown me that although they're all powerful in their ability to convert speech into written text, each has its own strengths and weaknesses. For busy and cost-conscious family physicians, one program, ViaVoice 98, does stand out as the superior choice.

New features of voice recognition

Until recently, voice recognition products required users to speak very artificially, pausing after each word. Now these systems use continuous speech recognition, which lets them understand normal speech and produce text at rates of up to 160 words per minute. (Despite the improvement, the act of dictating does take a bit longer with a voice recognition system than with a tape recorder because you must choose your words more precisely. All your conversational stammering — including every “um” and “err” — appears on the screen.)

Another new technology, natural language recognition, lets users speak commands in a flexible style using everyday language. Because the software looks for the core elements in commands, phrases such as “bold the last sentence” and “make the last sentence bold” can be used interchangeably. In addition, instead of having to issue many separate short commands (such as “Cut, move, paste”), users can now use normal sentences (such as “Move this word back two words”).

The software

Briefly, here are the three voice recognition systems I reviewed:

• Dragon NaturallySpeaking is available in six different packages with prices ranging from $60 (Point & Speak) to $695 (the fully featured Professional Edition) to $995 (Medical Suite, which became available last April). Medical Suite is essentially Professional Edition with an added medical vocabulary, and it's available at a lower cost as an upgrade to Professional Edition.

• Voice Xpress for Medicine (General Medicine Edition) was introduced last March for $500. Specialty editions (including one for primary care) also are available and cost $1,500. The specialty editions aren't available as upgrades.

• ViaVoice 98 (Executive Edition) became available last June, and IBM released its Medical Vocabulary for ViaVoice 98 one month later. ViaVoice 98 Executive Edition and the Medical Vocabulary are each about $150 and can be purchased separately.

Hardware requirements

All three products recommend use with a computer that has at least a Pentium 133 processor and 40 to 60 megabytes of RAM. The programs will require 100 to 200 megabytes of disk space to run optimally. If more than one user will be working with the software, you'll need at least 30 to 80 more megabytes of disk space for each person. Using a slower processor or having less than 40 megabytes of RAM will cause any of these programs to run poorly. I evaluated them on both a Pentium 166 with 80 megabytes of RAM and a Pentium II 300 with 128 megabytes of RAM.

Comparing the software

To begin using any of these programs, you must go through a “training session” that teaches the program how you speak. ViaVoice 98 and Dragon NaturallySpeaking let you choose a brief (five-minute) or a long (45-minute) session; Voice Xpress provides only a 45-minute session, and impatient users may find this frustrating.

To test the systems, I read passages from medical journals and dictated a history and physical exam for a middle-aged patient with chest pain. I tested Dragon NaturallySpeaking and ViaVoice 98 twice, once after completing the brief session and again after completing the extended session. (With Dragon NaturallySpeaking, unless you complete both the brief and extended training, you must go through the brief session again each time you run the program.)

I found that after the brief training sessions, ViaVoice 98 was the most accurate, roughly equivalent to the accuracy of dictation using a transcription service. Surprisingly, ViaVoice 98 appeared to make fewer mistakes with only its brief session than did Voice Xpress after its 45-minute session.

The three programs differ most in their ability to execute commands by voice. Because it has a partnership with Microsoft, only Voice Xpress can perform natural-language commands not only in Microsoft Word (as ViaVoice 98 and Dragon NaturallySpeaking can) but also in Microsoft Excel, Microsoft PowerPoint, Microsoft Outlook and Windows Explorer. Dragon NaturallySpeaking is alone in its ability to provide control of the mouse pointer and to emulate pressing the mouse buttons completely by voice. Though clearly not as efficient as moving the mouse by hand, this may be a very helpful feature for some users.

Another difference is that Voice Xpress and Dragon NaturallySpeaking let you perform “mobile dictation.” Using certain hand-held recording devices (usually digital), you can play previously recorded speech directly into the application.

Finally, the programs differ in their abilities to handle numerals, dates and dollar amounts. ViaVoice 98 and Voice Xpress accommodate natural speech; for example, to display $500.38 on the screen, you would say “five hundred dollars and thirty-eight cents.” However, with Dragon NaturallySpeaking, you would say “dollar sign five hundred point thirty-eight.” So if you plan to dictate many dates, times or numbers, Dragon NaturallySpeaking may not be your best choice.

The verdict

Although all three programs are impressive, I would recommend ViaVoice 98 and Dragon NaturallySpeaking. Many consider Dragon NaturallySpeaking the “best of breed,” and its performance measures up to that reputation. However, I find ViaVoice 98 to be at least its equal. ViaVoice 98 is priced significantly lower than its competitors, it offers fast and accurate dictation with minimal training, and it includes useful features such as natural speech for numbers and dates. As a result, it may be the product of choice for family physicians.