Newsgroups: comp.speech
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!gatech!newsfeed.pitt.edu!uunet!psinntp!isc-newsserver!ultb!adw3345
From: adw3345@ultb.isc.rit.edu (A.D. Williams)
Subject: SPEECH CUES 
Message-ID: <1995Jan24.035829.26956@ultb.isc.rit.edu>
Sender: news@ultb.isc.rit.edu (USENET News System)
Nntp-Posting-Host: ultb-gw.isc.rit.edu
Reply-To: adw3345@ultb.rit.edu ()
Organization: Rochester Institute of Technology
Date: Tue, 24 Jan 1995 03:58:29 GMT
Lines: 68


 I am currently considering a project for myself to help me use the
telephone. I am hard of hearing, and while I can hear environmental
sounds well, I have a difficulty in hearing the differences in sounds
in speech, especially over the telephone. For example, "Hello, how are
you?" might sound something like "Orro, ow or oo?"  You might understand
my apprehension when I call, say, Mistress Olga's Leather and Domination
chat line and not be sure if I'm really getting the most out of my
money.

 But that's not really the point. The point is that I'm wondering if
today's voice technology will help me speak to various people over the
telephone. 

 My fantasty device is a black box that will print out what the caller
is saying, such as "Will that be Visa again this time, Mr. Williams?"
But alas, I understand that that day is not here today, but, not, of
course, to any lack of effort to any of the subscribers to comp.speech.
Rather, researchers are rushing headlong with all due haste toward the
ultimate achievement of voice recognition, but alas, I have little
patience and wish to reap the benefits of science today.

 However, I can be willing to compromise, nobly lowering my expectations
to accomodate the early stages of voice recognition techology. Perhaps,
if my black box printed out the phenomes or even the syllables being
spoken, it would be a great advantage toward helping me understand 
speech over the phone. For example, phonetics on the screen, such as
"oe n ur nees or f eel th e la sh of my wh ip" would be a great help,
especially if I can see the phonetics appear on the screen at the same
time I hear the sounds. With some practice, I imagine that with similar
clues I will have a much more fun time on the phone. It need not be 
perfectly accurate, but close enough. After all, I outstrip the computer
in parsing the semantics of English as I slowly move my lips and drag my
finger across the computer screen when I read. 

 Currently, I'm looking at IBM's ICSS package, which is a speaker
independent package with an active vocabulary of 1000 words. It runs
using a BNF tree, as I understand most packages of this type use. It
would be a neat trick to program the top 1000 greatest hits syllables,
and use the "context switching" feature for the other less popular but
still useful on the odd occasion syllables. 

 I've also heard of Sound Systems' setup, which includes a card. While
I am willing to sacrifice much for science, IBM's ICSS package is a mere
$350, while Sound Systems' is over $1000. Selfless that I am, I have a
sister who needs a kidney operation, see...

 If this is a fesible project, I do plan to get a hideously expensive
Pentium computer to play DOOM^H^H^H^H with some of the ideas I've read
about in this newsgroup. I am quite sure that if my telephone project is
a success, perhaps I might find other uses for it. Perhaps, even
listening to talk radio? But then again, mabye I do have an advantage of
never once listening to Rush Limbaugh.

 So how about it? Will getting a fast computer, ICSS, OS/2, and hooking
it up to the telephone make me understand phone conversations better,
given that the computer can certainly hear phenomes better than I can?


                                              Sincerely,


                                                 Derrick
-- 
-------------------------------------------------------------------------
- Derrick Williams   Rochester Institute of Technology | Insert snappy  -
- adw3345@ultb.isc.rit.edu     Computer Science        | quotation here -
-------------------------------------------------------------------------
