Separate reviews by
Eric P. Daniels,
MD; and
Robert Hogan, MD
In 1993, Dragon Systems of Newton, Massachusetts received a $47,500
Phase I Small Business award from the US Army Space and Missile
Defense Command to study automated speech transcriptiontechnology
that enables a computer to transcribe recorded speech into written
text. This funding helped Dragon Systems to secure two other contracts
from the Defense Advanced Research Projects Agency to continue its
R&D efforts. In 1997, the company produced Dragon Systems NaturallySpeaking
software, a commercial software package that accurately transcribes
dictation into printed words. Various versions of this product are
described on the Dragon Systems Web site (www.dragonsys.com).
Because of the unusual nature of this technology and its powerful
implications for medical practice, two reviewers from different
fields of practice assess the latest medical version of this product.
Dragon Systems NaturallySpeaking Software: Use
in a Radiology Department
Is there a benefit in being an early user of voice-recognition technology?
Clearly, voice recognition software has the potential to lower labor
costs and to remove a clerical aspect of the physician's job, transforming
transcriptionists into editors and quality-control professionals.
Voice recognition software also provides the benefit of immediacy:
reports are available at the moment the x-ray films are read. But
how well does the software currently do this?
The CEO of Applied Voice Recognition recently stated, "Voice
recognition technology is not perfect and is unlikely to be that
way for 10 to 15 years at best." Perhaps understandably, Claudier
Tessier, Executive Director of the American Association for Medical
Transcription, stated that he "would not hire a transcriptionist
whose first pass looked like what is being produced as final from
voice-recognition software." By contrast, my own findings are
of 95% accuracy under ideal conditions, and better than 80% accuracy
in actual practice, and are quite in line with findings of others.
As a result, at least six radiologists in my department are eager
to pilot the technology. After a more formal cost/benefit analysis,
the idea may be worth a trial. The basis for my conclusions is as
follows: two different versions of Dragon Systems NaturallySpeaking
software, Professional Suite, were tested on two computers with
different microprocessor speeds. Both variablesversion of software
and speed of microprocessorproduced important effects.
Methods
I first tested Dragon Systems NaturallySpeaking software, Professional
Suite, v2.02, in the hospital on a 166-MHz Pentium II system equipped
with 96 MB of RAM. After 45 minutes of being trained to the nuances
of my voice, the system was able to identify approximately 85% to
90% of the words dictated. However, the software performed too slowly
for clinical use and was very sensitive to pauses. In addition,
breath sounds were detected and shown on the screen. (I have been
told that this aspect of the software can be corrected by further
training.)
I next tested the same version (v2.02) of the software at home
on a 500-MHz Pentium III system equipped with 256 MB of RAM. After
a similar 45-minute training period, the system successfully identified
85% to 90% of the words dictated. Speed of response was no longer
unsatisfactory.
At home, I then tested the latest version of the software (v3.52)
using each of the two Pentium systems. Before this software upgrade
could be used, the speech files stored in the application had to
be retrained; however, v3.52 is faster and more accurate than the
earlier version and includes many new features. (For instance, it
readily accommodates completion of special forms and tables.) Using
the Pentium III systemwhich provides many of the software's needs
directly from the motherboardI had the exhilarating experience of
being unable to speak more rapidly than the system could transcribe
my words. After only 30 minutes of training, the dictation accuracy
was impressive. Between 90% and 95% of common as well as specialty-based
words were transcribed correctly.
Discussion
My in-hospital, 166-MHz Pentium II proved too slow and unresponsive
for radiologic dictation, so subsequent testing was done using a
much faster machine in a quiet, peaceful environment. Accuracy may
be decreased by factors unrelated to the software itself (eg, use
of headphones, which can be distracting; the sounds made by handling
film jackets; environmental disturbances; interruptions; and other
extraneous noise); however, on the basis of my successful preliminary
trial at home, an attempt will be made to upgrade the equipment
in the hospital (ie, by adding a faster processor) for further evaluation
of the software.
Conclusion
I suspect that people who are already comfortable with using computer
technology will be right at home with this software. Conversely,
those who do not currently enjoy working with computers now are
unlikely to find this program a satisfactory experience. The real
question remains, what about most people, who are in-between?
By the way, in case you are wondering, this evaluation was "typed"
using Dragon Dictate.
Eric P. Daniels, MD
| Dr Daniels is Chief of Radiology in San Diego
and the Regional Coordinating Chief for Radiology for the Southern
California Permanente Medical Group (SCPMG). |
If you have never used a voice recognition program, just imagine
entering text by merely speaking to the computer! Voice recognition
by computers has been around since about 1985, when Kurzweil began
marketing "Voice Rad," a product designed to be used by
radiologists. Much has changed in the intervening 15 years. Better
voice recognition programs are now at hand.
Dragon Systems NaturallySpeaking software provides continuous speech
recognition: this package transcribes your words whether you speak
in full or incomplete sentences, and you can speak rapidly. As you
speak, you can watch the computer screen and witness the nearly
instantaneous appearance of each word on the screenas I am doing
right now, writing this review.
Why would anyone wish to use continuous speech recognition instead
of dictating into a recording device or typing directly into a word-processing
program? Speed, of course, is one reason: Few of us can type 100
words per minute, and the opportunity to record our thoughts very
rapidly is attractive in its own right. In addition, use of voice
recognition software can eliminate several steps between initial
recording of thoughts and final production of finished, "polished"
text.
To use a conventional transcription service, thoughts must first
be recorded and then transcribed; later, the transcribed results
of the dictation must be inspected to identify errors in the dictation.
Imagine if errors could instead be spotted and corrected immediately
as the words were appearing on the screen, thereby allowing production
of a perfectly wordprocessed document the first time. Indeed, continuous
speech recognition promises error-free text produced almost instantaneously
the first timemovement to paper at nearly the speed of thought.
This review is being prepared using Dragon Systems NaturallySpeaking
software installed on an IBM-compatible computer equipped with a
166-MHz processor and about 100 MB of RAM. Text appears on the screen
as words are spoken. The occasional errors are quickly corrected
using spoken correction features. Alternatively, the keyboard can
be used for correction; however, as you become more practiced with
this software, less and less use of the keyboard is required.
Essentially, the software learns the user's voice and pronunciation
of much of the English language so that accuracy increases substantially
with repeated use of the software. Ultimately, you can produce 100
words per minute with very few stops for correction. At the point
at which recognition is flawless, the keyboard could even be discarded.
This review so far has consisted of nontechnical language. However,
Dragon Systems NaturallySpeaking software can also recognize terms
such as "flex," "hysterectomy," and "trigeminal
neuralgia." Dragon Systems NaturallySpeaking software might
entirely replace medical transcription. Thus, any one of us can
equip our examination room or office with a fully functional, computerized
transcription system capable of rapidly producing medical notes
in finished form. The professional benefits of this capability are
obvious and come readily to mind. Perhaps not so obvious is the
possible and potentially life-enhancing use of the simpler, nonmedical
version of the Dragon Systems software for written communication
by quadriplegic patients.
For physicians who wish to replace their handwritten notes with
typed entries, for physicians who wish to avoid the expense of hiring
transcriptionists, and for anyone who yearns for immediate completion
of keyboard-free word-processed documents, the future is sufficiently
at hand to carefully consider using this product.
Robert Hogan, MD
| Dr Hogan is a senior family practitioner with
the Southern California Permanente Medical Group (SCPMG) in
San Diego, is medical software review editor for JAMA, and has
been an elected General Director of SCPMG. E-mail: RWHogan@Scal.kp.org |
Dragon NaturallySpeaking, Medical Suite, v3.52 Dragon
Systems, Newton, MA. Requires >133 Pentium processor, 64 MB RAM,
180 MB hard disk space, 16-bit sound card. Includes headset microphone.
$995.
If you have a medically related book coming out, please
ask your publisher to send a copy to The Permanente
Journal for review. Permanente authors will receive preference
in having their works selected for review. Reviewers are needed:
if you enjoy reading, can express yourself clearly in writing,
and would like to review medical books received, please let
me know by e-mail. If you have areas of special competence
and interest, please specify.
– Vincent J. Felitti, MD, Associate Editor (Vincent.J.Felitti@kp.org)
|
