Speech recognition-aug03

When some hospitals talk productivity, speech is the word. alk, talk, talk. That’s all caregivers at core engine with a customized front end that inte- grates with applications that weren’t built for speech, explains Michael Cipriani, Voicebrook’s Morreale has anything to say about it.
Voicebrook developed an interface for the hospital’s EMR (Patient1, recently purchased from ogy for medical transcription. But at Jacobi Per-Se Technologies, Atlanta, by Misys Healthcare Medical, speech-based documentation is being Systems, Raleigh, N.C.) and created treatment tied to everything from electronic medical records templates for each discipline that were designed to mimic a clinician’s workflow. Morreale says users Jacobi Medical started with desktop speech love the customization, but he admits that a sig- systems in radiology. But about 2,000 medical nificant change in future Dragon software might charts per day still had to be circulated around the require overhauling the templates, too.
medical campus, Morreale says. “My motivation Voicebrook also linked speech recognition to was, I wanted to stop delivering paper charts.” So medical coding through templates, sealing a big the organization implemented VoiceOver from dollar-leak, Morreale says. “By getting the doctors speech integrator Voicebrook, Lake Success, N.Y.
to go through a more structured progress note, I’m VoiceOver is a server-based version of the Dragon capturing all that they’re doing. That will allow me Copyright The McGraw-Hill Companies, Inc.
Reprinted with permission from Healthcare Informatics, August 2003 to bill at a higher rate of service and improve the reim- into its Somers, N.Y.-based Pervasive Computing Division.
bursement by several million dollars a year.” According to translation and voice offerings manager Brian Morreale has added voice-input capabilities to more than Garr, IBM’s ViaVoice Pro version 10 comes with an 400 enterprise workstations so far. He explains that he told improved headset from Santa Cruz, Calif.-based Plantron- employees he’d give them all new PCs and connectivity, but ics Inc., support for more digital recorders and better voice he wanted them to do all their documentation online.
“I gave them three choices: You can type it, you can use For those who use a PDA, the ScanSoft and IBM sys- voice recognition, or you can go practice medicine some- tems are compatible with the most recent versions of iPAQ where else.” He calls his strategy the “Catherine de Medici” Pocket PC from Compaq (now owned by Hewlett-Packard, approach to compliance—based on “bribery, flattery, compulsion and persuasion.” But, he adds, employees were Jonathan Snider, M.D., a physician at Hopkinton offered a lot of hand-holding, and their case loads were nego- Family Practice, Hopkinton, Mass., uses his desktop ver- tiated to allow the extra time needed.
sion of Dragon NaturallySpeaking daily to speed up Yes, speech recognition can cut transcription costs, referrals and to “talk” his email. Since he communicates especially if caregivers edit their own documentation. But regularly with a core set of referral physicians, he trained some organizations view it as a tool to boost business the system to recognize their names and spell them correctly.
productivity across the enterprise, especially in combination “Because I can type so fast, I thought it wouldn’t save time,” with an EMR, Cipriani notes. “As we see it, what’s driving Snider says. “But you can talk and have your hands free interest in speech is on the electronic medical record side.” for other things. It helps you multitask a bit.” Other voice recognition and EMR vendors besides For now, Snider uses handwritten charts, but he says he those mentioned are reaching out a hand to each other. Scan- hopes to migrate to an electronic chart eventually. “Then I Soft, Peabody, Mass., which owns Dragon Naturally- can dictate the subjective and dump it right into the EMR.” Speaking, already offers interfaces for EpicCare EMR fromEpic Systems, Madison, Wis., and Praxis EMR from InforMed, Woodland Hills, Calif. More partnerships are to Many hospitals first experienced speech recognition in the come this summer, says Matt Revis, product manager for radiology department. Talk Technology’s TalkStation and Dragon NaturallySpeaking. The impetus is workflow, he Dictaphone’s PowerScribe lead the pack in this niche, says: “It’s what the integration with the EMR and HIS is according to a January study by KLAS Enterprises, a Provo, all about. It’s thinking about how speech recognition Utah-based healthcare IT researcher. TalkStation includes changes the way healthcare professionals interact with editing tools, quick links to ICD-9 codes and a report- checking feature. PowerScribe Radiology version 4.5, Talk Technology, Bensalem, Pa., a subsidiary of Belgium- scheduled to launch this summer, adds more Web-based based Agfa-Gevaert Group, lists several healthcare IT ven- features for remote users and simplifies training techniques, dors as partners, including Cerner Corp., Kansas City, Mo.; IDX Systems Corp., Burlington, Vt.; GE Medical Systems, And Dictaphone is starting to reach out across the Milwaukee; and Siemens Medical Solutions, Malvern, Pa.
healthcare enterprise. Its EXSpeech, introduced in 2001, This summer, Dictaphone Corp., Stratford, Conn., is allows for telephone-based dictation, transcribed by speech planning to introduce its natural language patient record, recognition software at the server level. The system pro- touted as an EMR-like enhancement to Dictaphone’s new vides a speech-recognized draft document and an audio Enterprise Workstation. It comes complete with templates file to help transcriptionists become editors instead of typ- and self-editing tools, explains Don Fallati, senior vice president of marketing and strategic planning.
Rockford Health System, Rockford, Ill., installed EXSpeech earlier this year and has conducted months of parallel tests to see how many physician users dictate via As the speech market melds desktop, server-based and telephone well enough to achieve good accuracy rates.
mobile input modalities, much of healthcare still needs So far, nearly 75 percent of its 500+ physicians fare well a choice of platforms. For desktop users, Scansoft released with automatic transcription, reports Rockford’s CIO DragonNaturallySpeaking 7 Medical (DNS 7) in April.
and vice president Dennis L’Heureux. The best part, he Along with streamlined text rendering and a more resource- says, is the system’s transparency to users. “When physi- ful engine, DNS 7 includes a new Microsoft installer and cians in our organization pick up the phone to dictate, they administration tools for pushing out updates and vocabu- wouldn’t know whether their dictation is being voice- laries to desktops over the network, Revis says.
About 18 months ago, in a nod to the importance of The health system transcribes about 16 million virtual speech in pervasive computing, IBM moved its speech group lines per year, so getting physicians to use automatic Healthcare Informatics
transcription will streamline productivity and allow Project at IBM’s Yorktown, N.Y., research facility is devel- restructuring of costs, L’Heureux explains. “We were at a oping ways to combine speech recognition with visual cost of about 21 cents per virtual line. We think that with data, such as lip and mouth formations, and maybe even all the [process] changes we’re making, we can get it to about with emotional emphasis, notes IBM’s Garr. The compa- ny also is working on ways to combine multiple meth-ods of sending and receiving data, including a combination of VoiceXML and XHTML, he explains. This might mean One of the most annoying chores for speech recognition requesting information verbally but viewing the results users is having to say punctuation cues aloud. Scansoft’s in text, or vice versa. “Speech is evolving to be in many DNS 7 contains an optional “natural-punctuation” fea- places. It’s not just going to be on a desktop or on a ture, which uses probability and length of pauses to pro- server. It’ll be in everything you do,” Garr says.
vide punctuation automatically. For example, if a sentencebegins with “why,” chances are it’ll end in a question mark. Revis admits that the new feature is still rough Jacobi Medical has been busy trying to see that everyone around the edges, but it can help users who tend to repeat can be heard. As of May, speech was an input option for histories and admission notes in the adult primary care Today’s systems include sophisticated context engines clinic, the women’s health facility, the pediatric outpatient capable of discerning between similar-sounding words and center, sports medicine and the urology clinic. Morreale’s phrases based on the other words in the sentence. Part of next targets are cardiology and perhaps OB/GYN. His goal this is built-in and part is learned—the system keeps track is to provide speech as an input option anywhere docu- of which words are most often found together and makes ments are created in hopes that users will choose to talk increasingly intelligent guesses at the correct word.
But this is tricky stuff. For accuracy in a patient note, “But I have to mitigate the cost, because it’s not for example, the system must be able to discern the cheap,” Morreale says. “I have to provide a robust work- difference between a physician saying, “take a leave” station, special mikes and software licenses, and the train- (a rest period) and “take Aleve” (the Bayer drug). ing, maintenance and support. I have to move forward Voicebrook’s Cipriani says speech recognition systems somewhat cautiously, but it’s a good position to be in.” perform best when users train themselves to speak He adds: “I know that my bandwidth needs are going to logically and consistently—and to provide the crucial grow exponentially, especially with more voice recogni- keyword context. If physicians get used to saying “take tion and imaging going back and forth.” a leave of absence” and “take Aleve tablets,” the system Since speech recognition can make documentation will quickly learn to associate the proper choice with the available to the enterprise so rapidly, some wonder if it will eventually become linked to best practices and physi- For anything that falls through the cracks, most cian expectations. “We have small examples with that here desktop systems allow users to create a macro, forcing already,” Morreale says. “We process most of our lab the system to choose a specific word. In that case, results within an hour. People are used to that, and they’ll saying “macro Aleve” will limit the system to a single complain if their results are taking too long.” Someday speech recognition systems might be inter- Pamela Tabar is a freelance technology writer based in preting more than our voices. The Superhuman Speech Healthcare Informatics

Source: http://www.voicebrook.com/PressReleases/voicebrook-jmc1.pdf