Computers ‘learn’ to take dictation
Software can handle orthopaedic terms.
The ability to recognize continuous, fluent speech from a variety of individuals is no longer restricted to government supercomputers.
IBM and Dragon Systems compete head-to-head in the marketplace with programs priced at about $150 apiece, although many reviewers tout the Dragon product as being more accurate and faster. Both require at least one-hour of training using scripts provided with each program, as well as additional corrections of the first several dictation sessions in order to improve accuracy.
IBM was the first company to introduce continuous speech recognition, in September of 1996, with IBM Medspeak for radiologists. The University of Miami reduced the turn-around time on reports by an order of magnitude and currently more than 80 percent of the reports are dictated using the system. More than 10,000 reports have been dictated using voice recognition exclusively. However, this system costs from $9,000 to $14,000 depending upon support, setup and training, and it only handles radiology.
IBMís general continuous speech product, ViaVoice Gold (http://www.ibm.com/viavoice) allows the user to dictate into almost any PC application, including Word 97, Office 95, PowerPoint, Lotus SmartSuite, and WordPerfect. Minimum requirements for ViaVoice Gold are a 166 MHz Pentium, 32 MB of RAM, although a faster machine with more memory is preferred.
IBM is partnering with Olympus (http://www.olympus.com) to introduce a digital voice recorder, the D1000, that downloads dictation into you computer where it is automatically transcribed by Via Voice Gold.
I tried out Via Voice Gold for a few weeks and found the setup to be straightforward and the training to be somewhat tedious. After adding several computer files containing abstracts and articles that I had written in the past, the accuracy did pick up.The system even learned to recognize "subtrochanteric" and "scanning electron microscope."
Dragon NaturallySpeaking (http://www.naturallyspeaking.com) runs on Windows 95 and Windows NT and also offers continuous speech recognition. There is a companion product for the Macintosh, Dragon Power Secretary. Dragon comes with a 60,000 word vocabulary and allows the user to add specific topics such as radiology, orthopaedics and pathology. Minimum requirements are a 133 MHz Pentium processor with 32 MB RAM and at least 100MB of hard disk space. Dragon also is introducing a portable solution for speech recognition by partnering with Norcom Electronics (http://www.norcomelectronics.com/speechrec.htm)
Eric Fishman, MD, an orthopaedist in Palm Beach, Fla., has used Dragon NaturallySpeaking in his office for some time. He has developed an orthopaedic vocabulary and language model for use by two of his physician assistants who now dictate 100 percent of their reports with the program. He suggests that two hours per day for two to three weeks would be sufficient for a physician to successfully dictate more than 50 percent of his/her office reports using voice recognition.
Speech Machines (http://www.speechmachines.com) avoids the problem of individual training by relying upon software developed by the Speech Research Unit at Britainís Defense Research and Evaluation Agency. I registered for a free trial of the service by visiting their web site and obtaining an account number and a personal identification number (PIN). I then called a toll-free number and began speaking directly into the phone. I read several paragraphs out of an Instructional Course Lecture on total knee replacement and less than 12 hours later, three pages of double-spaced dictation arrived via e-mail to my office computer.
For a first-time attempt over phone lines, Speech Machines got more than 95 percent of the dictation correct, even picking up words like "thromboembolic" and "total hip arthroplasty." However, they did drop a few words without indication and misspelled others. You can also dictate directly into a hand-held recording device from Voice It Worldwide and then submit your recording to Speech Machines.
The cost is $29.95 to register plus $9.95 per month for four pages of transcribed text, plus $3.50.
Either the IBM or the Dragon product will get you started at minimal expense. Detailed information for several products is available on the web at (http://voicerecognition.com).
Jay D. Mabrey, MD, is chairman of the Academy's Committee on Electronic Media