Speech Recognition Software Do you like what you hear?

by David Orenstein freelance writer

Speech recognition technology is destined to forever seem “emerging,” rather than “real” or “useful.” But recently it has begun to perform well in real-world applications. It’s time to begin studying whether it is appropriate for your practice.

The answer probably will hinge on the size of your practice. The bigger you are, the more cost-effective it is. The technology seems to work, but the economics of the technology might not.

Some practices have begun to try speech recognition and have discovered that, when it works accurately, speech recognition has the potential to save time and money. Accuracy is the key word here. Speech recognition had earned a reputation for being laughable, but now many software packages (at least on the high end) get it right 95 percent of the time or more, which many organizations find acceptable. But proofreading is always wise.

At the Healthcare Information Management Systems Society’s 2000 conference in Dallas, speech recognition got a testimonial from the University of North Carolina Health System in Chapel Hill. UNC’s emergency department requires attending physicians to use speech recognition software for all their notes. Although the emergency room has a lot of the loud, random ambient noise that historically has hindered speech recognition systems, UNC reports a 96 percent accuracy rate. Physicians can dictate notes into templates set up to accommodate the hospital’s data needs and to meet HCFA regulations. UNC plans to expand its use of speech recognition as an alternative to traditional transcription in other areas.

UNC is not alone in its eagerness to make broader use of speech recognition software, according to a HIMSS survey of senior-level technology executives at 858 healthcare provider organizations. The June 2000 survey found that only 21 percent of organizations were using speech recognition, but another 35 percent expected to start doing so in the next two years. The organizations surveyed were large organizations like UNC, which probably can afford to spend what it takes to buy or develop a great speech recognition system. In smaller practices, the benefits will be a bit more modest.

Take the example of Irving psychiatrist Michael H. Brophy, MD. He has been considering speech recognition technology, but as of mid-December he hadn’t bought any, and with good reason. He rightly describes his 166 MHz IBM PC as an “antiquity,” given the heavy demands speech recognition places on PCs. The technology is only as good as the processor it is running on. Although software vendors claim you can get by with a 300 MHz Pentium II, don’t believe it. You should have at least a 500 MHz system and at least 128 MB of RAM. This means Dr Brophy would need a new computer, and that adds substantial cost.

So for Dr Brophy to improve on the current method of typing his notes into his computer, he has two low-cost alternatives: 1) Spend $1530 for a $1200 computer, a $190 copy of IBM’s ViaVoice Pro software, and a $140 upgrade that includes a general-purpose, 25,000-word medical vocabulary, or 2) Start using a dictation company that bombards him with flyers quoting a rate of 5 cents a line. The $1530 system would become cheaper than the dictation service after 30,800 lines of notes.

But the economics work out better for a physician who already has a relatively new computer. In that case the $340 of software would pay off after 6800 lines. If two physicians in a practice share that computer, the return gets better still. The benefits or rewards boil down to the paramaters of your practice, but it might finally be true that speech recognition can help you recognize real savings.

David Orenstein is a technology and business writer in Silicon Valley. If you are interested in learning more about a technology topic in Computing Care, e-mail him at davealli@earthlink.net.