1、Designation: E 2364 04Standard Guide toSpeech Recognition Technology Products in Health Care1This standard is issued under the fixed designation E 2364; the number immediately following the designation indicates the year oforiginal adoption or, in the case of revision, the year of last revision. A n
2、umber in parentheses indicates the year of last reapproval. Asuperscript epsilon (e) indicates an editorial change since the last revision or reapproval.1. Scope1.1 This guide identifies system types and describes variousfeatures of speech recognition technology (SRT) products usedto create the heal
3、thcare record. This will assist users (healthinformation professionals, medical report originators, admin-istrators, medical transcriptionists, speech recognition medicaltranscription editors (SRMTEs), system integrators, supportpersonnel, trainers, and others) to make informed decisionsrelating to
4、the design and utilization of SRT systems.1.2 This guide does not address the following items:1.2.1 System and data (voice and text) security.1.2.2 Administrative processes such as authentication of thedocument, productivity measurements, etc.2. Referenced Documents2.1 ASTM Standards:2E 1902 Specifi
5、cation for Management of the Confidentialityand Security of Dictation, Transcription, and TranscribedHealth RecordsE 1985 Guide for User Authentication and AuthorizationE 2084 Specification for Authentication of Healthcare In-formation Using Digital SignaturesE 2184 Specification for Healthcare Docu
6、ment FormatsE 2185 Specification for Transferring Digital Voice DataBetween Independent Digital Dictation Systems and Work-stationsE 2344 Guide for Data Capture Through the DictationProcess2.2 Other Documents:Resource Interchange File Format (RIFF) Standard3. Terminology3.1 Definitions:3.1.1 acousti
7、c model, nphoneme map of user.3.1.2 authentication, nthe process of confirming author-ship of an entry or of a document, for example, by verifyingwith a written signature, identifiable initials, computer key, orother methods.3.1.3 author, nperson responsible for content of text file.3.1.4 back-end s
8、ystem, ndelayed processing for documentcompletion.3.1.5 compound file, na file containing recorded voicewith its transcribed text.3.1.6 context, na long list of vocabulary words andphrases used for the particular subject matter, with theirspellings and pronunciations, statistical information aboutus
9、age of each word alone and in combination. For example, thecontext may include the number of times that “right,”“Wright,” “turn right,” “right turn,” “right hand,” and “Mr.Wright” occur in a body of text. It also includes grammar andstyle information. Language model, lexicon, topic, and vocabu-lary
10、are terms that are all used synonymously with context.3.1.7 digital signature, ndata associated with, or a cryp-tographic transformation of, a data unit that allows a recipientto prove the source and integrity of the data unit and protectagainst forgery, for example, by the recipient.3.1.8 edit, vto
11、 review the document while listening to theoriginators recorded voice and reading the associated tran-scribed text (compound file), checking for recognition errorsand correcting document formatting and other inconsistencies.When the SRMTE is not the originator, the SRMTE may needto flag the document
12、 for originator/author clarification ofunclear content or intent.3.1.9 encryption, nthe process of transforming plain text(readable) into cipher text (unreadable) for the purpose ofsecurity and privacy.3.1.10 front-end system, na system incorporating real-time recognition and may include real-time s
13、elf-editing by theoriginator.3.1.11 language model, ncontext specific to medical spe-cialty, user, or practice setting.3.1.12 lossless compression, na lossless compression re-duces the amount of data required to represent the originalvoice file but has no impact on sound quality. The original fileca
14、n be replicated precisely at any time.1This guide is under the jurisdiction of ASTM Committee E31 on HealthcareInformatics and is the direct responsibility of Subcommittee E31.15 on HealthInformation Capture and Documentation.Current edition approved June 1, 2004. Published July 2004.2For referenced
15、 ASTM standards, visit the ASTM website, www.astm.org, orcontact ASTM Customer Service at serviceastm.org. For Annual Book of ASTMStandards volume information, refer to the standards Document Summary page onthe ASTM website.1Copyright ASTM International, 100 Barr Harbor Drive, PO Box C700, West Cons
16、hohocken, PA 19428-2959, United States.3.1.13 lossy compression, na lossy compression losessome information, resulting in degradation of the sound qualityinherent in the original voice file and an inability to preciselyregenerate that original file.3.1.14 microphone, nan instrument whereby sound wav
17、esare caused to generate or modulate an electric current usuallyfor the purpose of transmitting or recording sound (as speech ormusic).3.1.15 microphone element, nthe component within themicrophone that does the actual conversion from sound wavesto electrical signals.3.1.16 natural language processi
18、ng, nmethod used inartificial intelligence to process and derive interpretation ofhuman language.3.1.17 networked system, nsystem connected to a net-work.3.1.18 “normal” dictation, nroutine phrases or para-graphs.3.1.19 originator, nperson who provides oral input ordictation, not necessarily the per
19、son responsible for the con-tent.3.1.20 phoneme, nsmallest unit of sound in a spokenlanguage.3.1.21 prompts, nreminders provided in order to completea task.3.1.22 real-time recognition, nsimultaneous speech-to-text transcription.3.1.23 RecOspeech recognition error3.1.24 RIFF file, nResource Intercha
20、nge File Format(RIFF) is self-descriptive; that is, the voice file format isdefined within the file.3.1.25 speech recognition, ncomputerized transcriptionof speech to text.3.1.26 speech recognition medical transcription editor,nmedical transcriptionist who edits compound files and/orthe SRT language
21、 model.3.1.27 SRT engine, nspeech recognition processor.3.1.28 standalone system, nsystem not connected to anetwork.3.1.29 synchronization, vhaving voice and text matchedsuch as in a point-and-play manner.3.1.30 text file, na file that contains text message.3.1.31 voice enrollment, nthe process wher
22、eby a userreads aloud selected text so the SRT software can map orrecord the users speech sound pattern (phonemes).3.1.32 voice file, ndigitalized audio message representingvoice input.3.1.33 voice macros, nstored keystrokes that are activatedby a voice command.3.1.34 WAV, nvoice file format.3.2 Acr
23、onyms:3.2.1 MTmedical transcriptionist3.2.2 SRMTEspeech recognition medical transcriptioneditor3.2.3 RIFFresource interchange file format3.2.4 SRTspeech recognition technology4. Significance and Use4.1 This guide is intended to provide general guidelinestoward the design and utilization of SRT produ
24、cts used forhealthcare documentation. It is intended to recommend theessential elements required of SRT systems in healthcare.4.2 This guide will not identify specific products or makerecommendations regarding specific vendors or their productsor services.4.3 A well-edited SRT document may result in
25、 improvedquality over current methods of documentation, that is, hand-written notes and improved productivity over traditional dic-tation and transcription.4.3.1 Faster turnaround times.4.3.2 Legible documentation over handwriting has manyadvantages:4.3.2.1 Improved patient care communication.4.3.2.
26、2 Enhanced patient safety.4.3.2.3 Reduced malpractice risks.4.3.2.4 Facilitation of appropriate reimbursement.4.3.3 For the medical transcriptionist and/or SRMTE, de-creased repetitive stress injuries, such as neck, arm, wrist, andheel pain.4.3.4 Facilitation of cost controls related to documentcomp
27、letion.4.3.5 Better utilization of medical language skills of MTs asproductivity is not limited by keyboarding skills.5. Speech Recognition Technology Systems5.1 Speech recognition technology (SRT) is designed tocapture voice and transcribe that speech into text. This can bedone by a single user wor
28、king at a standalone computer or bya large group of users working on a network. Another methodis processing a pre-recorded digital voice file through an SRTsystem, with the resulting text and/or SRT engine being editedby the MTE.5.2 Speech recognition technology system workflow.5.2.1 Front-end speec
29、h recognition process involves:5.2.1.1 Recording the voice.5.2.1.2 SRT transcription of the voice file to text.5.2.1.3 Editing may be done by the originator and/orSRMTE.5.2.1.4 Compound file may be saved as an option.5.2.1.5 Text file can be printed, archived, transmitted, orintegrated into an elect
30、ronic health record.5.2.1.6 Update the SRT context for RecOs and new termi-nology.5.2.2 Back-end speech recognition process involves:5.2.2.1 Recording the voice.5.2.2.2 Transmitting the voice file to the speech recognitionengine.5.2.2.3 SRT transcription of the voice file to text.5.2.2.4 Saving the
31、voice and text as a compound file.5.2.2.5 Routing the compound file to the SRMTE.5.2.2.6 Editing done by the SRMTE.5.2.2.7 Saving the text file.5.2.2.8 Returning the edited text file to the originator forauthentication.E23640425.2.2.9 SRMTE updates the SRT context for RecOs andnew terminology.5.2.3
32、Standalone SRT System:5.2.3.1 Only one person at a time can use a standalonesystem.5.2.3.2 Context is limited by the hard drive space.5.2.3.3 Editing is done locally, at the point of input, eitherby the originator or by the SRMTE.5.2.3.4 Input devices.(1) Noise-canceling SRT microphones.(2) Handheld
33、 digital recorders.(3) Digital dictation systems.(4) Telephones.5.2.3.5 The following scenarios are offered to give thereader examples of how these systems work. They are notintended to represent every possible scenario for these systems.(1) A radiologist (originator) dictates into a microphoneconne
34、cted to a personal computer running an SRT program.The voice is translated to text in real time. The originator editsthe text and/or the SRT context.(2) A family practitioner dictates into a personal computerthroughout the day. Each compound file is saved and then,using the same computer, the SRMTE
35、edits the text, listeningto the recorded voice as necessary for clarification. TheSRMTE may also be responsible for editing the SRT context.(3) A group of cardiologists dictate into handheld digitalrecording devices throughout the day. The voice files aretransmitted from the recorders to a computer
36、and recognized bythe SRT engine, using the cardiology context and each physi-cians acoustic model. Once recognized, each text file is editedby the SRMTE. The SRMTE may also be responsible forediting the SRT context.5.2.4 Networked SRT System:5.2.4.1 On a networked system, all files containing record
37、eddictation (voice files) are transmitted to a server, where the filesare queued up for recognition. The compound files are thenrouted to the SRMTE for editing.5.2.4.2 A networked system is designed to allow multipleoriginators and SRMTEs to work simultaneously. The voicefiles are recognized on a se
38、rver or at the workstation(s) and theresulting compound files are routed to the SRMTE for editing.5.2.4.3 Contexts.(1) The networked system may be programmed for a singlemedical specialty or subspecialty, such as radiology, pathology,family practice, physical therapy, or emergency medicine.(2) A net
39、worked system may also be programmed withmany contexts or language models so originators from manydifferent medical specialties can use it to improve speechrecognition accuracy.5.2.4.4 Editing may be done in the same facility, or thecompound files may be sent to a remote SRMTE.5.2.4.5 Input devices.
40、(1) Noise-canceling SRT microphones.(2) Handheld digital recorders.(3) Digital dictation systems.(4) Telephone.5.2.4.6 The following scenarios are offered to give thereader examples of how these systems work. They are notintended to represent every possible scenario for these systems.(1) Six radiolo
41、gists simultaneously dictate at individualworkstations. Each voice file is routed to a recognition server,or the processing may take place on each workstation, withinformation regarding the originators specialty and identifica-tion, allowing the recognition server to load the correspondingacoustic m
42、odel and context. The voice file is processed by theSRT engine and the resulting compound file (voice and textfiles) is routed to the SRMTE for editing. The SRMTE mayalso be responsible for editing the SRT context.(2) A hospital has 300 healthcare providers dictating intoportable handheld digital re
43、cording devices from the hospitaland several remote satellite clinics, or dictation may take placeon individual workstations. The voice files are encrypted andsecurely transmitted to the digital dictation system of acontracted transcription company. Each voice file is routed to arecognition server,
44、or the processing may take place onworkstations, with information regarding the originators spe-cialty and identification, allowing the recognition server to loadthe corresponding acoustic model and context. The voice file isprocessed by the SRT engine and the resulting compound file(voice and text
45、files) is routed to the SRMTE for editing.SRMTEs working both in the office and remotely receiverecognized compound files via encrypted Internet transmis-sions. The editing is performed on standalone computers andthe encrypted text files are returned. The SRMTE may also beresponsible for editing the
46、 SRT context.6. Training6.1 Originators:6.1.1 Voice enrollment and proper position of microphoneand proper placement of microphone element.6.1.2 Build customized language model.6.1.3 Build “normal” dictations per user.6.1.4 Develop skill sets.6.1.4.1 Proper correction technique for a RecO.6.1.4.2 Na
47、vigation/mobility skills for moving around in thedocument.6.1.4.3 Editing skills specific to SRT products.6.1.4.4 Editing language model.6.2 Speech Recognition Medical Transcription Editor:6.2.1 Voice enrollment and proper position of microphoneand proper placement of microphone element.6.2.2 Build
48、customized language model.6.2.3 Build “normal” dictations per user.6.2.4 Develop skill sets.6.2.4.1 Proper correction technique for a RecO.6.2.4.2 Navigation/mobility skills for moving around in thedocument.6.2.4.3 Editing skills specific to SRT products.6.2.4.4 Editing language model.6.2.4.5 Start
49、and stop audio file.6.2.4.6 Identify a RecO.7. Realities of Speech Recognition Technology7.1 Originators with good dictation habits will more likelybe successful using SRT. See Guide E 2344.E23640437.2 Originators with exceptionally heavy guttural accentsmay have more challenges. However, speakers of English as asecond language are not necessarily precluded from using SRT.7.2.1 Depending upon the SRT context it may not resolvethe differentiation of homonyms.7.3 SRT will not overcome dictation errors, improper gram-mar, incomplete or disorganiz
