Re: voice to text software

["Ned Bedinger" <doc -at- edwordsmith -dot- com>]

Can't speak for any of the current crop of speech
recognition engines, but the speech recognition
engine in the original MS SDK (10-ish years ago)
required a lot of hand-holding (training of the
engine, proof-reading of the output).  These
engines have a training phase, which is supposed
to tune the speech recognizer to more accurately
render
text from your speech, microphone, etc.  To train
it, you speak, then the recognizer then tells you
what it thinks you said, then you correct it and
try again.  I haven't ever dwelt on this phase at
any length because results never seemed to be
forthcoming at a level of reliability that I
thought worthwhile in an environment where
sentence structure and the vocabulary used was
extensive and variable.

There are instances where current speech
recognition software shines.  My wireless phone
provider's customer service line's voice
recognition seems to be very stable.  It never
misinterprets my voice response when it has
prompted me to press or say "one", or asks me to
say "Billing", "Customer Service", or "Other".
The vocabulary it is prepared for is, of course,
limited.

I found the novelty factor of text-to-speech much
more compelling than the productivity factor of
speech to text.  With a little bit of VBA
scripting, the Text-to-Speech control in the MS
SDK could be made to read documents or email.  I
believe that the monotonous-ness of the voice
precludes any serious uses beyond brief mails or
documents.  It is just too hard to listen
carefully to a synthesized voice where all of the
modulation and timing, such as you would expect to
hear in a vivid voice, is absent in the
synthesized voice's reading.

The really interesting voices heard by the user of
automated systems are usually recorded human
voices--it is far easier to record a message for
every conceivable scenario in, say a telephone
voice menu, than to program a synthesizer to
competently read scripts to the user. NOAA weather
radio uses a synthesized voice that is mildly
amusing ("Partly clo-oo-dee") and reasonably
intelligible, but the sentences and vocabulary it
uses are predictable, making it possible to add
polish through programming.

Ned Bedinger
doc at edwordsmith dot com


----- Original Message ----- 
From: "Downing, David" <DavidDowning -at- Users -dot- com>
To: "TECHWR-L" <techwr-l -at- lists -dot- raycomm -dot- com>
Sent: Monday, March 29, 2004 6:39 AM
Subject: RE: voice to text software



Just how much success -- or lack thereof -- have
people on this list had
with speech-to-text software?



^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

ROBOHELP X5 - ALL NEW VERSION. Now with Word 2003 support, Content 
Management, Multi-Author support, PDF and XML support and much more! 

Now is the best time to buy - special end of month promos, including:
$100 mail-in rebate; Free online orientation on content management
functionality; Huge savings on support and future product releases;
PLUS Great discounts on RoboHelp training. OFFER EXPIRES March 31! 
Call 1-800-358-9370 or visit: http://www.ehelp.com/techwr-l

---
You are currently subscribed to techwr-l as:
archiver -at- techwr-l -dot- com
To unsubscribe send a blank email to leave-techwr-l-obscured -at- lists -dot- raycomm -dot- com
Send administrative questions to ejray -at- raycomm -dot- com -dot-  Visit
http://www.raycomm.com/techwhirl/ for more resources and info.