Big Talk in Healthcare -- Voice Recognition

Voice recognition technology has sold in healthcare markets for years, usually as stand-alone systems for individual users. But two changes that occurred in the past year have made voice recognition far more attractive to customers and integrators alike. First and foremost is the introduction of continuous speech products by all of the major speech vendors, including Dragon Systems, IBM, and Lernout & Hauspie This innovation is directly related to improvements in recognition speed and accuracy. "Continuous speech systems can record 100 to 130 words a minute, with accuracy in the 98 percent range. I find that voice recognition systems provide results that are at least as good as most transcription services," says Steve Scheer, VP of marketing for Global Digital Information in Bellevue, WA. Global Digital publishes CaduSys Medical Record software for Windows NT. The other big change is the introduction of software development kits by the major vendors. These kits provide integrators and ISVs with shortcuts that make incorporating voice recognition into existing applications easier than before.
			"Voice recognition eliminates delays in creating patient records." Rachelle Reisenberg, President of AM Technologies a Newton, MA-based integrator that focuses on speech recognition.

"Doing so is well worth the effort," says Scheer. His company recently integrated Newton, MA-based Dragon Systems' NaturallySpeaking product with CaduSys. "We estimate that adding voice can save clinics $1,000 per month, per provider, in transcription costs," he says. "That savings by itself cost-justifies our entire medical records system. And doctors love it because they can treat the same number of patients each day and then go home at a decent hour."

Customer Education Still Required

Because it is new and unfamiliar to many customers, voice recognition is a technology area where integrators can make a big impact, according to Rachelle Reisenberg, president of AM Technologies, a Newton, MA-based integrator that focuses on speech recognition.

"If you saw a demo just a year ago, the capabilities will amaze you," she says. "One of the challenges of selling the technology is managing customer expectations about the technology's capabilities, which are equally likely to be too high or too low."

"Customers have to be trained to understand what kinds of mistakes voice recognition makes and how to make corrections," Reisberg says. The other side of the equation is that they may not be aware of of its capabilities.

Reisberg's firm has benefited from the recent surge in customer interest. Last year most of the firm's voice recognition sales involved only one or two product units Recently the integrator has been asked to bid on solutions scaled to midsize practices and emergency rooms that require support for a dozen users, as well as larger organizations that want to run pilots before scaling to a departmental buy.

The shrink-wrapped $50 voice recognition products sold by mass marketers are usually inappropriate for professional installations. It is for that reason that all of the major players encourage their channel partners to sell the high-end versions of their products, which can be customized for particular audiences. Some products are tweaked by manufacturers before they're shipped to provide the best possible performance for particular vertical markets.

These highly customized products are targeted at vertical markets where users have a large amount of dictation to do and use a limited vocabulary. They provide the biggest return on investment for customers and deliver the biggest profits to integrators.

For example, IBM's MedSpeak/Radiology which boasts a 98-percent accuracy rate, is building a loyal following. Integrators that want to develop their own vertical-market packages can also build on products like Dragon's NaturallySpeaking Medical Suite, which includes a copy of Dragon's high-end engine, a customizable medical dictionary, and a suite of customized medical templates and applications from Corel.

The payoff with voice recognition has been impressive for many healthcare providers. "Voice recognition improved my job dramatically," says Stacey Wanless, an administrative assistant for an Alma, MI, orthopedic surgeon's office, who uses Dragon's NaturallySpeaking. "I no longer transcribe. It used to take 8 to 10 hours to transcribe 20 charts a day. Now it takes 2.5 hours, and most of my time is spent getting money back into the practice."

Joe Devlin and Emily Berk are partners in Armadillo Associates Inc., a solutions integrator with offices in Half Moon Bay, CA.