Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!warwick!zaphod.crihan.fr!univ-lyon1.fr!ghost.dsi.unimi.it!batcomputer!caen!usenet.cis.ufl.edu!usenet.ufl.edu!darwin.sura.net!howland.reston.ans.net!agate!overload.lbl.gov!lll-winken.llnl.gov!taurus.cs.nps.navy.mil!schievel
From: schievel@cs.nps.navy.mil (michael schievelbein)
Subject: Re: IBM ICSS (IBM Continuous Speech Series)
Message-ID: <CJxuut.4Dq@taurus.cs.nps.navy.mil>
Sender: news@taurus.cs.nps.navy.mil
Organization: Naval Postgraduate School, Monterey
References: <comp-speech-faq_758178011@GZA.COM> <1994Jan19.210726.22595@Princeton.EDU>
Date: Thu, 20 Jan 1994 17:20:53 GMT
Lines: 39

In article <1994Jan19.210726.22595@Princeton.EDU> 0428870@phoenix.princeton.edu (Devin Hosea) writes:
>Has anyone tried the IBM Continuous Speech Series Developer's Toolkit
>and Run-Time System for WINDOWS (or on any other platform for that matter?)
>
>If so, could you please comment on the veracity of IBM's claims:
>
>	- processes speech of different speakers with a wide variety
>	of accented speech
>
>	- accomodates natural or rapid rates of speech
>
>	- provides active vocabulary of up to 1000 words
>
>I am looking at several different speaker-independent systems, and
>I would appreciate any comments you have, and any suggestions you
>have about other systems.  Thank you for your time.
>
I have tried the IBM ICSS and currently have it running on my system.
I use it at my OS/2 User Group Meetings as a demo tool and people are
surprised that it requires no training.  IMHO the recognition after 
getting the microphone sensitivity set and things like that is still
topping out at about 80% and tops out at about 95% once you learn the 
rate and flexation that the programs are looking for.  

The system is designed for CONTINUOUS SPEECH and therefore ideal for 
normal speech rate, in fact if you speak in a pause type situation like 
the older recognition systems required, it will only mess things up.

I believe the vocabulary is way larger than 1000 words, in fact I think
it can be up to like 70K words.  I know that the 1000 words is an ideal
value for any given application with no delays on an average machine,
I have a editor beta that I am trying with a full dictionary and running 
on a 486/66 with no delay.

Michael Schievelbein

>


