Continuous speech transcription: A reality?
Continuous speech transcription: A reality?
Even where used, someone has to edit it.’
Designers of electronic speech recognition technology are making determined strides toward eliminating the "human factor" in medical transcription, but medical transcriptionists needn’t worry about being forced to find a new career path any time soon, experts say, because the transcribed text still requires final editing.
Continuous voice recognition is the latest wave of electronic voice recognition technology that transcribes spoken words into computer text. Unlike earlier, discrete voice recognition technology that required the speaker to talk slowly and pause between words, continuous voice recognition allows the dictating clinician or physician to speak at normal speed.
Accuracy reaches 97% to 99%, says Joe Weber, CEO of Narratek Inc., a Brookline, MA, reseller of IBM MedSpeak continuous speech recognition systems for radiology. (For a list of continuous voice recognition software and manufacturers, see story, p. 139.) Statistically that translates into roughly one mistake every three to nine lines. And it only applies to speakers with standard North American accents; non-native English speakers typically achieve a slightly lower accuracy rate, Weber says.
The software "enables radiologists to continue dictating the way they’ve always dictated without having to pause between words," Weber says.
Continuous speech recognition technology is still essentially limited to a few fields of medicine that use small vocabularies. Radiology vocabulary, for example, consists of only about 2,000 to 5,000 medical terms, says Catherine Baxter, president of Voice Info Transitions, a Houston consulting firm that specializes in medical transcription outsourcing and medical records re-engineering and quality review. That limited vocabulary compares with roughly 70,000 medical terms in health care overall. Pathology will likely be the next area to use continuous voice recognition, Baxter says.
Continuous voice recognition software has also been developed for orthopedics, mental health, and emergency medicine, says Dimitri Kartsonas, director of research and development for TransQuick, a health data capture company in Atlanta. Kartsonas’ company uses software made by Philips Speech Processing Inc., which also sells software for radiology.
An adaptation process
IBM’s software uses an "enrollment" period during which the dictator repeats 50 sentences that are flashed on the computer screen for about 12 minutes, Weber says. A longer version of the enrollment period, recommended for those with accents, lasts 45 minutes and flashes 150 additional sentences on the screen.
Meanwhile, the Philips’ technology uses an adaptation process that compares a recorded digital voice file with the edited text to "learn" the speech and word patterns of a particular speaker, Kartsonas says.
The new technology is a timesaver for overburdened medical records departments because the clinician can dictate notes directly into a computer, and a money saver because fewer transcriptionists are needed, Weber says. "There is something beautiful when a radiologist dictates a report to have it available immediately and to have the [electronic] signature be a true authentication. It’s a very efficient and effective way to do [medical transcription] in real time.
But the technology has not been perfected to the point of complete reliability, some experts caution. "It is being used in a limited area, says Pat Forbis, CMT, associate executive director for the American Association for Medical Transcription in Modesto, CA. "Some medical transcriptionists themselves are using it, but it has a long way to go."
There will always be a need for medical transcriptionists because a computer program, no matter how sophisticated or "smart" will never be able to reason as well as the human mind and correct other human errors, Baxter says. "It’s not at the point where the computer can truly tell the difference between one word and another. Physicians make a lot of errors when they dictate. We can’t assume those errors are going to go away just because they’re using a different kind of microphone."
And Forbis notes, "If you have a cold or sinus infection, the [computer] doesn’t recognize you; it will create its own gobbledygook language. Then it all has to be edited. Even where it is being used, someone has to edit it."
"Only a medical transcriptionist can determine when a human error has been made," Baxter adds. "What a good transcriptionist does is say, The doctor said right side when we know the patient is having surgery on the left side.’"
Catching such mistakes, argues Weber, is the responsibility of the doctor not an editor or transcriptionist. "The doctor should catch that. The doctor should be responsible for reading over the notes. When the notes are read over in real time, the person who should be responsible for it is responsible."
But the new technology and traditionally trained transcriptionists are not mutually exclusive, Kartsonas says. "We try to utilize technology not to supplant the transcriptionist but to increase efficiency because it’s not going to be 100% accurate, and you always are going to need someone with medical knowledge.
"The keyboarding skill which is a required of a successful medical transcriptionist is not required [with voice recognition technology] so it has the potential to allow someone with the medical background but not the keyboarding skill to be a correctionist rather than as a transcriptionist."
But will doctors accept the technology?
Another factor to consider before you purchase continuous voice recognition software is the stumbling block physicians are likely to throw in your path. When you try to switch to a new technology, "It’s a matter of changing [doctors’] practice patterns. Any time you have that, it’s not going to be well accepted. It’s not going to be widely used," Baxter says.
But the march of technology is steady and determined, and the move to extensive use of continuous speech recognition "is going to happen faster than most people expect," Weber says.
For more information contact:
Pat Forbis, Associate Executive Director, American Association for Medical Transcription, Modesto, CA. Telephone: (209) 551-0883.
Catherine Baxter, President, Health Info Transitions, Houston. Telephone: (713) 728-4184.
Joe Weber, CEO, Narratek Inc., Brookline, MA. Telephone: (617) 566-1066.
Dimitri Kartsonas, Director of Research and Development, TransQuick, Atlanta. Telephone: (770) 909-1220.
Subscribe Now for Access
You have reached your article limit for the month. We hope you found our articles both enjoyable and insightful. For information on new subscriptions, product trials, alternative billing arrangements or group and site discounts please call 800-688-2421. We look forward to having you as a long-term member of the Relias Media community.