With the latest versions of speech recognition software, you can dictate directly into a PC, laptop, or into a digital recorder, then transfer the digital recording onto your PC and let the software dictate from the recording. Typically, you'll need a computer with a fast processor such as a Pentium III or equivalent, a good sound card, and 512 megabytes of RAM. Most new computers can meet these requirements, but make sure you know exactly what hardware you need before you buy.

All speech recognition products require you to spend some time training it to understand your vocal inflections before you can use it. With one common product, IBM's ViaVoice, for instance, the user spends more than an hour after initial installation reading prepared text into a headset microphone as the software analyzes his voice. Each physician who plans to use the software will have to spend time training it, and if your physicians like to use more than one computer in the office, it is probably easiest to just install the software to your server rather than loading it onto each of your practice's hard drives.

And because physicians' language is so distinct, it's imperative that the software has a medical vocabulary. Physicians are often tempted to buy generic, out-of-the-box speech recognition software because it's cheaper than more sophisticated versions and thus seems like a good product to experiment with. But consultants such as Nelson advise strongly against this approach; they argue that these basic versions are poorly suited for healthcare. Doctors quickly lose patience and pitch their new purchases in a drawer, Nelson says.

"The software folks have realized that there are different levels of need" for different users, she says. "So if you look at a couple of leading products out there, such as Dragon and ViaVoice — they both sell different versions, in that there's a small-office or home-office version, there's the professional version, and then there are the medical solutions, which have a more extensive and medical-specific vocabularies."

Nelson calls these medical solutions the "right" products for physicians, though they cost more, "because I think physicians need to recognize that there's a cost of getting in the game. If they want to try speech recognition, it may work very well, but too often, because it's only a trial, physicians invest in the most economical software they can find, and that's almost a pre-determined failure. Because it's not the right tool."

What it costs

The "right" speech recognition tool is not particularly expensive — at least, not compared to many other healthcare information technology solutions. For example, in November, ScanSoft, the owner of Dragon Naturally Speaking, released its latest medical version of the product, Dragon Naturally Speaking Medical 8, which retails for $1,095. Its standard version of the same product can be had for $99.

That may seem like a big difference, but a $1,000 price tag is not exorbitant when you consider what you're already paying for transcription service, which might run $10,000-$15,000 a year. Even if you don't have the right hardware, it shouldn't cost more than $2,000 to equip yourself with a brand-new computer.

"I tell people, just pretend you're going to school — you're going to speech recognition school and tuition is $2,000," says Nelson. "It'll cost $1,000 for the software, and maybe another $1,000 on hardware. If it doesn't work, you still have a good new PC, and you never know. It's really a smart thing to try."

Practices should be aware of the pros and cons of speech recognition software, and the efforts required to implement it. The typical 2 percent error rate can be significant, says Rhodes, especially for specialists who do a lot of dictation. (ScanSoft says its latest version of Dragon is down to a 1 percent error rate.)

Physicians who want to get the most of the technology are advised to correct errors manually, using a process that allows the software to learn how to avoid that mistake in the future. That process saves time in the long run but takes more time on the front end.
Another issue to consider is the set-up time, as well as the additional responsibilities of the physician, who must now do his own dictating and editing, which could lead to changes in the physician's workflow habits, says Rhodes.

But those changes need not be drastic. A physician who currently dictates into a tape recorder and then has the recordings transcribed could use a digital recorder and download the recording onto her PC, then run the speech recognition program. But she must still edit the transcription herself.

"Physicians should ask, what's their flow now?" advises Nelson. "When do they dictate now? Where do they do it and how do they do it? Do they step out of the room? Are they using a recording device? A phone? Do they do it from a couple or three workstations? If that's the case, they'll need that many PCs with the software loaded where they do that now. It'll be awfully hard to change those habits, and you might fall behind if you're not using your normal method."

Before you buy

Here are some other tips for getting the most out of speech recognition: