"Most physicians think they can take it out of the box and just use it," says Julianne Weight, CEO of Alphabest, a workflow management consultancy serving the medical profession based in Los Angeles. "It ends up on the shelf, and they swear it doesn't work because they don't have the patience to get it to the point where it actually makes them productive."

One major mistake that many physicians make is that they don't correct the system while they are working, says Weight. For example, they may say "lab results," but the system hears "ladder faults." "If you don't correct it, the system will base all forward work on that sound meaning that word, and it will only get worse as time goes on. Accuracy will dramatically decline. In effect, you are training the system to mis-hear you."

"It can be close to 100 percent accurate if you take the time to train it well," says Weight. "Or it can be as awful as 30 percent if someone lacks the patience to do a good job."

Use the Extended Features
Another thing that most physicians don't do is to fully leverage the "macros," which are ways to program the system to take shortcuts by saying a word. For example, when performing a physical exam, a doctor could program a macro to be triggered by the word "normal" that would automatically drop in normal values into the chart without having to specifically say what the metrics were each time.

Still, there have been enough technological advances to make investing in speech recognition well worth physicians' time. "The biggest milestone was the Intel Pentium 4 chip, which finally had the built-in capability to run speech recognition engines," says Anthony Benach, CEO of Crossings, a Los Angeles-based speech recognition consultant specializing in installations for physicians' offices, who says that the Dragon speech recognition system, which he recommends, is now 99.5 percent accurate if implemented correctly.

"Transcriptions generally run $1,600 to $2,200 a month per physician. And that's only the transcribing cost. The physician still has to wait – sometimes for weeks – to get it back, check it over for accuracy, edit it, and, frequently, send it back to be redone," says Benach. "Moreover, most transcription services will only guarantee 98 percent accuracy."

One thing that Benach is adamant about: "when they list the minimum requirements on the speech recognition software box, these are really meant to mean 'minimum,'" he says. "You really need a top-of-the-line system, with at least three gigabytes of RAM, especially if you are running the Vista operating system.

And there's a lot of apprehension out there, says Benach. "Many offices are trying to transition to electronic medical records (EMR) systems in attempts to make their office operations more efficient. They're already a bit traumatized by the cost and pain of moving to a new system, and hesitate to add more complexity on top of that."

Yet Benach's experience with clients is that by installing speech recognition, physicians can pay for their EMR systems within two years. "It's really a worthwhile investment," he says.