Prev: Steve Jobs/Speech Recognition/Myself - - - good coalition
Next: Why do Apple Mail accounts go off-line spontaneously?
From: Mark Conrad on 24 Nov 2009 23:36

Okay people, this is the last time I will subject
you to that complex medical speech recognition (SR)
example I keep using lately, the one with nasty phrases
like "perioperative transesophageal echocardiography"
and names of Japanese surgeons like:
"Fusahiko Itoh, MD, Takeshi Someya, MD".

This post will address the fallacy which some
people believe, namely that SR apps _have_ to
be used in a quiet environment.

Such is not the case, as you yourself can determine.


Preparation for Noisy Test
************************

Now I did the usual preparation for this noisy
test of our humble bug-ridden $200 MacSpeech app.

I installed it from scratch from its DVD, the
same as a SR newbie would do. (version 1.5.0)

After installation, I stumbled along for 15 minutes,
pretending I was a newbie, training the infuriating
app to recognize my voice, a very frustrating
experience for any newbie. Not fun at all.

Went online to download the most recent update,
version 1.5.7


Then I added 4 small files from a previous "trained"
MacSpeech: (trained to handle medical words)

1) User Profile (30 MB)
2) Special Medical Words, 21 of them (16 KB)
3) Six of my macro commands (37 KB)
4) Sample text file (3 KB)

The "Sample text file" was dragged into the
"Tools -> Vocabulary Training" window and the
default settings there were accepted.

Again, the purpose of the "Sample text file" is to
get the out-of-the-box MacSpeech to recognize the
few medical words that it would not normally
recognize, medical words such as "transesophageal".

One last preparation, I configured the microphone to
ignore loud outside noises, very easy for a newbie
to do this.


The Noise Test Itself
********************

I cranked up the volume on my TV so loud that it
was painful to my ears. Some guy yakking away
on an infomercial

Dictated the 1,479 syllables of medical speech into
the MacSpeech app, took me 7 minutes to dictate.

If I had been dictating one-syllable words, that
would be a dictation speed of 211 words per minute.

Perfect 100% raw accuracy, no text mistakes,
no correction whatever needed, despite the very loud
TV noise.

Keep in mind that if I had spoken any wrong word
during my 7 minute dictation session, that wrong word
would have registered as a mistake.


This pretty well establishes that SR apps can be used
in very noisy environments, contrary to common belief
by those who do not know how to use SR

Results darn good for our bug-ridden MacSpeech, right?

For one last time, the text results of this noise test
will be in the post immediately following this post.

Mark-
From: Mark Conrad on 24 Nov 2009 23:40
In article <241120092036546884%aeiou(a)mostly.invalid>, Mark Conrad
<aeiou(a)mostly.invalid> wrote:

> For one last time, the text results of this noise test...


New Surgical Procedure for Ischemic/Functional Mitral
Regurgitation: Mitral Complex Remodeling

Hirokuni Arai, MD, PhD*, Fusahiko Itoh, MD, Takeshi
Someya, MD, Keiji Oi, MD, PhD, Kiyoshi Tamura, MD,
PhD, Hiroyuki Tanaka, MD, PhD


Department of Cardiothoracic Surgery, Tokyo Medical and
Dental University Graduate School of Medicine, Tokyo,
Japan



* Address correspondence to Dr Arai, Department of
Cardiothoracic Surgery, Tokyo Medical and Dental
University Graduate School of Medicine, 1-5-45 Yushima,
Bunkyo-ku, Tokyo, 113-8519, Japan Email:
hiro.tsrg(a)tmd.ac.jp



On-pump beating heart mitral complex remodeling was
performed without aortic clamping. The mitral valve was
exposed through a left atriotomy posterior to the
interatrial groove. Interrupted 2-0 braided horizontal
mattress sutures without pledgets were placed around
the annulus to optimize exposure of the subvalvular
apparatus. Secondary chords to the anterior leaflet
from both papillary muscles were carefully separated
from primary chords with a nerve hook and were divided.

Two pairs of 5-0 and 4-0 Gore-Tex sutures (W. L. Gore &
Associates, Flagstaff, AZ) were each placed to both
fibrous portions of the anterior and posterior
papillary muscle tips, buttressed with pledgets of
autologous pericardium. Two pairs of the free arms of
the 5-0 Gore-Tex sutures were twice passed through the
free edge of the middle portion of the anterior leaflet
about 5 mm from the margin, from ventricular to atrial
side. Suture length was adjusted to be the same length
as the corresponding marginal chords, and the sutures
were tied.

Each pair of the free arms of the 4-0 Gore-Tex sutures
was passed through the posterior annulus at sites
around the border of the lateral and middle portions
and middle and medial portions of the annulus,
respectively (annulopapillary suture), and was also
passed through corresponding sites in the annuloplasty
ring (Carpentier-Edwards Physio; Edwards Lifesciences,
Irvine, CA). The 26-mm semi-rigid annuloplasty ring was
then seated. The annulopapillary sutures were pulled to
retract the papillary muscle tips closer to the
annulus, to the point at which leaflet coaptation
occurred in the plane of the mitral annulus during
systole, to visually confirm no residual MR. Suture
lengths were determined, and the sutures were tied.

To avoid air embolism, a vent cannula with a
pressure-monitoring catheter (TOYOBO Co Ltd, Osaka,
Japan) was inserted into the left ventricular apex and
was connected to the suction circuit equipped with a
small reservoir chamber (Senko Medical Instrument Mfg
Ltd, Saitama, Japan). During the final adjustment of
the annulopapillary suture length, this chamber was
filled with blood, and the height of the fluid level of
this chamber was adjusted to load the left ventricle.
The left ventricular systolic pressure was monitored to
keep it slightly lower than the systemic perfusion
pressure to avoid ejection through the aortic valve.

This new technique has been performed on 3 patients
with ischemic/functional MR. The patients were aged 61,
64, and 69 years; their ejection fractions were 0.34,
0.25, 0.32; their left ventricular diastolic diameters
were 62, 74, and 79 mm; and tenting heights were 11,
12, and 14 mm, respectively. Preoperatively, all
patients showed severe MR; perioperative
transesophageal echocardiography showed disappearance
of MR. Mitral valvular function has remained stable
during a mean short-term follow-up of 6 months (range,
1 to 12 months), with no or trivial MR noted.
 | 
Pages: 1
Prev: Steve Jobs/Speech Recognition/Myself - - - good coalition
Next: Why do Apple Mail accounts go off-line spontaneously?