Computer Consult

Voice recognition programs speak your language

They can also save you tens of thousands of dollars in transcription costs.

Noah Gilson, MD

I'm glad I didn't give up on voice recognition software.

Six months after I installed a program in 1999 to automatically transcribe my dictation, I was tempted to junk it. I hated how it transmogrified sentences like "I evaluated the patient" into "I violated the patient." I hated even more how a message would pop up on the screen declaring that, due to an "illegal operation," the program would be shut down—and I'd lose all the dictation I hadn't saved.

I nonetheless remained convinced that this technology would streamline the practice of medicine. And eventually it did. My computer's now saving me the $20,000 a year I once paid flesh-and-blood transcriptionists.

Voice recognition software also benefits my practice in other ways. I can dictate complicated instructions to a patient while he's in my office and hand him a printout (another copy goes into his paper chart). And automated transcription improves my relationship with referring doctors. I sometimes dictate a letter to a colleague and have it faxed to him before his patient walks out the door. Given this fast, on- the-ball update, the referring physician's more likely to trust me with another patient.

I didn't reach this promised land without climbing up the learning curve, however. Here are some lessons I've gleaned that can help you exploit speech recognition software for all it's worth:

Hire a guru. Two of my partners and I bought our software, IBM's ViaVoice, through a reseller—a consultant who represents a vendor. We paid him roughly $3,000 to train us on the program, and we still call him when we run into problems.

The cost of training was almost quadruple the $800 that we paid for the basic software and some medical and neurology modules, but the extra expense was worth it. From what I've observed, most doctors who buy software directly from a vendor and teach themselves won't stick with it. Vendors like IBM and ScanSoft, which markets Dragon Naturally Speaking, can guide you to resellers in your area.

Install the software on a fast computer. A faster computer lets you correct mistakes more quickly. I recommend a machine with at least 1 gigahertz of processing speed, which is what you'll find in a high-end Pentium III computer. And go heavy on random access memory—512 RAM is a safe bet—because the vocabularies found in speech recognition software hog a lot of space.

Does your current computer lack this firepower? You can find a system with the right specs for under $1,500.

Buy a good sound card. A sound card allows your computer to produce sound on speakers and to record sound from a microphone. The better the sound card, the more accurately the software transcribes dictation.

Many computers come with sound cards built into the computer's so-called motherboard, but such cards usually aren't top-of-the-line. Installing a separate sound card—like a Sound Blaster from Creative Technology—can improve recognition accuracy. A Sound Blaster card costs as little as $30.

Buy a good microphone. My voice recognition software came with an okay headset microphone, but I was able to boost accuracy by buying a higher-quality headset from a company called VXI for about $75. Make sure your microphone is designed for voice recognition duty. It should bear the label "noise canceling."

Position the microphone correctly. Puffs of air following consonants like p, as well as nasal breath sounds, can make voice recognition software throw extra words and syllables into dictation. That's why experts recommend positioning a headset microphone to the side of your mouth.

I found that getting headset microphone placement correct was tricky since I kept taking it off and putting it back on throughout the day. Plus, I didn't like being tethered to the computer. I eventually paid under $100 for an "array" microphone that sits on a table. It picks up my voice from several feet away, eliminating the problem of puffy p's and nasal gusts. Now I can move about more freely and stop fretting about microphone position.

Edit yourself. Sure, a staffer can correct your raw dictation, but you should be your own editor, at least during the first six months when you "train" your software. With ViaVoice, that's done by making the correction, then saying the word. The program learns the word for the next time, and accuracy improves.

Compensate for late-day fatigue.Dictation recognition accuracy often drops off late in the afternoon because I'm not enunciating as clearly. Rushing to get dictation done before I go home also tends to introduce more errors. For that reason, I've installed my voice recognition software on my home computer, too, so I can finish that last batch of chart notes when I'm more relaxed.

Incidentally, overenunciation garbles dictation as much as underenunciation does. For best results, speak in a normal, relaxed manner, albeit slightly slower than usual.

Speed up dictation with macros. You can program your software to create whole paragraphs of text as soon as you utter a word or two. I've created macros for routine prescriptions, patient instructions, and the introductory and closing remarks of letters.

It doesn't cost a truckload of money to get started with voice recognition software—maybe $3,500 if you throw in a computer and training—but it does require a big investment of patience. I still average five transcription goofs per page. Some are minor, such as a plural noun mysteriously becoming singular. Other mistakes are ridiculous. I once said, "The patient was shunted and then made a nice recovery," only to see the computer write, "The patient was shot dead and then made a nice recovery." I'm a good neurologist, but not that good.

However, every annual update of my voice recognition software has improved accuracy. When I lump this improvement together with progressively more stable operating systems and faster computer speeds, I'm glad that I've stayed with the technology. The benefits—particularly what I'm saving in transcription costs—definitely outweigh the inconveniences.

The author is a neurologist in West Long Branch, NJ. Computer Consult is edited by Senior Editor Robert Lowes.

 

Noah Gilson. Computer Consult: Voice recognition programs speak your language. Medical Economics 2002;5:25.