Speech Recognition For Family Medicine

 

 

Louis Spikol M.D.

1111 6th St.

Whitehall, Pa.18052

 

Educational Objectives:

 

At the end of this presentation you should be able to:

1. Understand the capabilities of medical speech recognition.

2. Evaluate the use and utility of voice macros.

3. Understand the ability of medical speech recognition software to complete various regions of medical forms.

4. Identify how medical speech recognition software can interact with other commercial form creation software.

5. Evaluate the use of speech recognition within the context of an electronic medical record.

 

Intended audience:

 

1.  Providers of medical care including physicians, physician assistants, nurse practitioners and others.

2.  Medical secretaries and transcriptionists.

3.  Anyone with an interest in speech recognition technology.

 

Introduction:

 

            My name is Lou Spikol and I practice full-time in Allentown, Pennsylvania.  I have been using speech recognition to construct office notes since continuous speech recognition came on the scene in May of 1997.  I use speech recognition almost everyday to write office notes, e-mail, correspondence and presentations like this one.  Since 1997 I've used speech recognition in almost every way possible-alone, with macros, with templates, and with electronic medical records.  I've also taught a number of family practitioners to use medical speech recognition I’ve taken a lot of time and effort and made many mistakes along the way.  My goal is to give you a head start implementing medical speech recognition into your workflow.  Although I've use Dragon NaturallySpeaking almost exclusively, the tips and advice that I will give you will be applicable for all speech recognition programs.

 

The Medical Speech Recognition Scene 2003

 

            I never thought I would say this but speech recognition and specifically medical speech recognition is now "ready for prime time".  What this means, is that with proper instruction and motivation and proper equipment any physician canned learn and implement medical speech recognition.  In 2003 a number of factors have converged to make this possible:

1.  Incredibly fast and relatively cheap desktop and laptop computers with plenty of memory and processor power to run speech recognition along with any other medical software that is necessary.  Every morning when I turn on my computers I load up speech recognition and it remains ready to use when needed.

2.  Increasing recognition by almost every electronic medical records software vendor that speech recognition represents a tool that most physicians will want to have available.  Accommodations are being made to include the easy use of speech recognition by their software.

3.  A significant core group of physicians who now use speech recognition effectively and are willing to demonstrate and teach others.

 

Getting Started with Speech Recognition

 

 

The Hardware

 

            Desktop computers: the processor should be at least a Pentium 3 (or Athlon) 800 MHz or greater with ideally 512 MB of RAM.  The amount of RAM is especially important for speed and accuracy.  If you have a choice between some extra processing power and extra RAM, choose extra RAM. Many of the newer speech recognition versions are optimized for the Pentium 4 processor.

Notebook computers: again, the processor should be at least a Pentium 3-800 MHz or greater with ideally 512 MB of RAM.  I believe notebook computers are becoming the platform of choice for physicians and speech recognition.  They have greatly matured in terms of their capabilities.  In addition, laptop computers can be used quite easily in the examination room.

            Sound Cards: in desktop computers the sound card should be a separate sound card.  Sound cards integrated into the motherboard have too much noise and are not acceptable for speech recognition.  Internal sound cards of notebook computers tend to vary-some are acceptable and some are not.  Recently, external USB sound cards have become available.  They are approximately $50 and are worth trying if you have difficulty with recognition either on a desktop or laptop using the internal sound card.

            Microphones: the microphones that ship with most speech recognition software are good but not excellent.  I recommend that you invest in a better microphone.  Excellent microphones cost approximately $60-$100.  A few excellent, high-quality microphones cost from $100-$400.

 

Hot hardware tips: adequate RAM, consider a USB sound card, consider getting a good microphone.

 

Installation and Training

 

·         Follow instructions to train your voice for general English.

·         After you have finished training for general English, check the program to make sure that your accuracy is adequate.  Your accuracy should be at least 90 percent or above after this initial training (try to speak continuously pronouncing each word.).  If your recognition is not adequate, consider your setup, microphone, computer etc. as noted above.  If your recognition is adequate, you must now teach the program family practice language.

·         The method that you use to teach the program the language of family practice depends upon whether you have text files of your dictation!  The ideal situation is where you have text files, perhaps produced by your Secretary, of your medical dictation.  Every speech recognition program has a module that can analyze your previous text dictations, present you with new words and learn your specific speech patterns and words.  I really cannot emphasize enough this aspect.  By using your own text files it's possible to get very high accuracy within a few minutes of using the program.  If you have written your notes and do not have text files, family practice language programs are available commercially.

·         Be patient initially and try to correct most miss-recognized words. The program will learn rapidly from corrections.

 

Hot installation and training tips: follow directions, test your recognition for general English, and let the program analyze your previous medical dictation if available, purchase a medical language model if necessary.

 

Speaking to the computer and general tips:

 

·         Speak in a clear, continuous manner, inserting all punctuation, leaving out extraneous sounds (uhs and ehs).  Don't "baby the program" by speaking in single words-it will learn much better if you speak clearly and fairly rapidly.

·         If you use a headset microphone, adjust the microphone properly to the side of your mouth with proper spacing.

·         Consider running the microphone setup each time you begin a long dictation session.

·         Consider buying the professional version of your chosen speech recognition program.  The extra capabilities, especially the macro capabilities make this worthwhile.

·         Consider learning and using the macro capabilities of your program to expedite your work and minimize the amount of speech recognition that you need to perform.  Let's face it, much of medical transcription is redundant and can be automated.

·         If you use speech recognition for medical and general English consider setting up two separate language models.  This serves the purpose of just keeping medical phraseology within the medical language model.

·         Consider participating in a speech recognition user group.  Believe me; if you have a question, someone "out there" has an answer.  Many of these people are unbelievably knowledgeable and helpful.

·         Consider using medical speech recognition as part of an overall strategy to improve documentation, compliance and billing.  If you have been under-coding and under billing but feel you have been doing the work and not able to produce documentation this is just your ticket!  If you feel you have been over coding with the amount of documentation that you have, this will help to increase compliance.

·         Finally, consider minimizing the need for speech recognition in the first place.  I understand that this sounds like heresy in a document about speech recognition but I'm quite serious.  Consider using speech recognition with other programs such as template programs, electronic medical record programs, database programs etc. that can build up most of your notes without speech recognition.  An additional bonus is that by using some of these programs, information and chart construction can be performed in the examination room to a large extent.  Speech recognition can then be used to fill in the free text and finish the note.

 

Voice-activated Data Entry

 

Dragon NaturallySpeaking:

      The speech recognition program and especially the professional program have significant capabilities in the following areas: 1. The ability to create "boilerplate text" and activate the text using a speech macro.  2.  The ability to create placeholders within boilerplate text whereby the physician can use speech macros to jump from place to place and fill in variable text.  3.  The ability to create macros that perform an action such as cutting and pasting text and activating other programs.

 

In addition to the basic speech recognition program there are other very useful add-on programs that function quite well with speech recognition.  They are all quite useful, relatively inexpensive and many provide you with the capability of trying the program before purchasing.  I will list them with highlights and to their web site.

 

Adobe Acrobat  www.adobe.com 

The full version of Adobe Acrobat makes it possible to create boilerplate text as well as interactive forms which can then be filled out using speech recognition.

 

OmniForm 5  www.scansoft.com  

This program can also be used to create interactive forms and will also convert paper forms into electronic forms.  Again, speech recognition can be used to fill out the forms.

 

Multiblock Storage  www.mbs4you.com 

This program is a very nice macro utility that runs in a taskbar at the top of your screen.  It's function is to create boilerplate text.  It is especially useful in that it gives you visual cues as to what boilerplate text is available.

 

Microsoft Word  www.microsoft.com  

Microsoft Word has fairly sophisticated macro capabilities that can be used quite well with medical speech recognition.

 

Shorthand  www.pcshorthand.com  

in addition to a macro utility this program allows the completion of check boxes which allow for fairly good flexibility and variable input during the patients encounter.

 

Soapware   www.docs.com  

A reasonably priced, well supported electronic medical record that works well with speech recognition.

 

Amazing Charts   www.amazingcharts.com  

A reasonably priced, well supported electronic medical record that works well with speech recognition.

 

Medical Speech Recognition Resources

 

Software and microphones:

 

www.scansoft.com  -dragon systems home page. You can also find the support forum here.

 

www.dragonsys.ca  an excellent source of software and microphones.

 

www.emicrophones.com - another excellent source of software, microphones and general information.

 

www.voiceautomated.com -source of software as well as medical language models.

 

www.zydoc.com -software and medical language models.

 

www.fastplates.com  - an add-on database program for Dragon NaturallySpeaking

 

 

 

Miscellaneous Sites

 

http://voicerecognition.com/voice-users/ -Voice recognition users group.

 

www.onelist.com the medical speech recognition users group and IBM users group can be found here

 

www.knowbrainer.com  commercial site specializing in customized Dragon macros

 

www.out-loud.com general helpful speech recognition site.

 

 

Louis Spikol M.D.   Allentown,Pa.  June, 2003