It has been my experience that Dragon products are generally more likely to
interpret noises as words than Kurzweil.
With DragonDictate this was particularly so. With DragonDictate and the
Shure SM10A we saw many UK Dragon dealers cancel exhibition presentations,
or stick to well-tried macros.
Even at CeBit, the demo using a VXi was plagued by mis-recognitions. We
watched for over an hour. I don't think we saw more than five lines.
I don't want to criticise Dragon technology. We were once one of their prime
dealers and UK beta testers. This is a 'noise-as-word' issue only. And is
plainly a combination of mics, cards, software AND USER.
Kurzweil VoicePlus is quite easy to adjust for background noise and we
regularly demonstrate in the noisiest of exhibitions. The Talk Mic's high
gain means you can lower the sound card gain - usually taking card noise
down with it. The ESS notebook card was always troublesome as the internal
noise was quite high.
Using L+H/Kurzweil Voice Xpress continuous speech (in pre-alpha) the Talk
Mic was simple to set up and worked exceptionally well. Good
out-of-the-box-no-training recognition. This may be due to the software as
we didn't get any of the 'below average' messages common to so many
NaturallySpeaking users with the usual mics.
Pressure gradient technology is better than noise cancelling. With Voice
Xpress we left the mic on the table and spoke individually and collectively
from about a metre away. And played the radio. Nothing on screen.
The Talk Mic worked well with NaturallySpeaking although we don't sell it.
And SS and Via. But didn't demo them at shows so can't comment on high noise
rejection by the software.
There is a Talk Mic earphone model and we do have 2 metre low loss extensions.
Larry Allen at Softnet in Phoenix currently has the Talk Mics (without
earphones). We have no control over retail prices. Nor would wish to have
any. If you wan't a better mic, buy one. If not, have a good meal out.
With regard to Voice Xpress it does have continuous natural language command
and control. This means that there is little to learn in the way of fixed
grammar commands - you just talk to it more like a person to edit and
format. It's fascinating to use.
It has also been our experience with Talk Mics that we have never had to use
a battery box. The only add-on is a small mechanical adapter for the
ThinkPad's 'unique' mic socket.
And finally, the discussions about 96% and 97% accuracy seem quite bizarre.
Recognition is affected by many different factors including the user. I have
seen people use ALL of the voice systems for the last seven years. I've seen
demonstrators get in a muddle. I've seen out-of-the-box fabuluous
recognition. And out-of-the-box abyssmal. I've watched professional users
having all sorts of problems. The IBM presentations have sometimes gone way
off-line. Users at home and at work. Kids at school. Disabled. Dyslexic.
Erratic recognition for no reason. Stray words. Context errors. I've watched
and helped thousands of users.
We deal with a lot of corporates and many say that voice demos have been bad
or their users eventually found it easier to type. Many users say they got
on better with discrete speech.
One company with 18 700 employees has three voice users. Some have trialled
voice recognition and got ten users. We've got a 160 in one division.
However you look at it, we've got a long way to go.
Talking Technologies Limited
34A Glazbury Road
London
W14 9AS
talk@@easynet.co.uk
http://www.talk-systems.com/
vox 0171 602 4107 fax 0171 603 2109
![]() |