MacSpeech Dictate Medical frustrations
Mark Conrad
Hi people, just a post here for my own benefit,
to relieve my frustrations with that tiny
New Hampshire company MacSpeech.
As you know, they have a new speech recognition
program named "MacSpeech Medical" for sale at
the mere [sic] cost of $595 in U.S. dollars.
Announced for sale on Thursday June 18, 2009
via various press releases, and emails to naive
customers like myself.
Tried my best to get a copy initially, but no
response from the sales dept of MacSpeech, so
to heck with them, being they want to make it
so darn difficult to contact them and ask a few
questions before I shell out $600.
I really believe that MacSpeech has made a big
mistake in trying to sell this new software,
when they should be concentrating on removing the
bugs from the $200 base "MacSpeech Dictate 1.5.1"
app which is the basis of the new medical program.
Slapping a big complex medical vocabulary onto a
very buggy base program just does not make sense.
FWIW, below I created a new "profile" in my existing
$200 version of MacSpeech.
The aim of this exercise is to demonstrate to myself
and everyone else that the present $200 MacSpeech app
is capable of handling complex medical jargon, at least
in my hands.
Please do not try this yourself, as it will just lead
to frustration.
Through trial and error and past experience, I know
most of the quirks of this buggy software, and how to
work around those bugs.
Below is a partial extract of a brain surgery
operation, as described at this website:
<http://journals.lww.com/neurosurgery/Abstract/1998/02000/Endoscope_assi
sted_Brain_Surgery__Part_2_Analysis.8.aspx>
I dictated the text from that website into MacSpeech.
There was only one lonely mistake that MacSpeech
made, mistake is "89 of these lesions"
which should be "eighty-nine of these lesions".
Big Deal, right?
I could have set up my medical "profile" so that
error never would have occurred, but I lost
patience with the author of that article,
who in my mind should have used "89" so that
it would be consistent with the rest of his
article, where numbers were used instead of words.
Below is the raw dictated text, with its one error
- - - - - - - - -
RESULTS: The lesions treated with
endoscope-assisted microsurgery comprised
205 tumors, 53 aneurysms, 85 cysts, and 36
neurovascular compression syndromes.
89 of these lesions where localized
in the ventricular system, 242 in the
subarachnoid space or intracerebral, and 49
in the sella. Endoscope-assisted
microsurgery was advantageous to reduce the
size and the operation-related tissue
trauma of approaches to lesions within the
ventricular system, in the brain tissue as
well as in the subarachnoid space at the
base of the brain. Using less retraction
during tumor removal, the visual control of
retrosellar, endosellar, retroclival, and
infratentorial structures was improved.
Video-endoscope instrumentation was
especially helpful during procedures in the
posterior cranial fossa and at the
craniocervical junction. It allowed for
inspection of channels and hidden
structures (e.g., the internal auditory
meatus, the ventral surface of the brain
stem, the ventral aspect of root entry
zones of cranial nerves, the content of the
foramen magnum, and the upper cervical
canal), most without retraction and without
resection of dura and bone edges. Endoscope
instrumentation during surgery or large or
giant aneurysms was useful to dissect
perforators on the back side of the
aneurysms and to control the completeness
of clipping.
- - - - - - - -
End of dictated text, with its one
so-called "error".
If I really wanted to, I could have set up a
_complete_ medical "profile" for brain surgery
terminology, such that any complex terminology
likely to be used in brain surgery would yield
the 99% raw accuracy that todays speech apps
are capable of producing.
Getting good output like the above example is
_very_ _difficult_ considering the present
buggy state of MacSpeech.
I will add my voice to many others, in that the
one feature that would make our life a lot easier
would be to allow any "built-in word" to be
trained to the users actual voice.
The present situation is that any built in word
can NOT be tied to a users actual voice for
improved accuracy, plus improved ability to
differentiate between similar sounding words.
There should also be the capability to temporarily
deactivate built-in words that will never be used
by a particular user of MacSpeech
Mark-