Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!pipex!uunet!destroyer!sol.ctr.columbia.edu!usenet.ucs.indiana.edu!ectl-sub@snowhite.cis.uoguelph.ca
From: ectl-sub@snowhite.cis.uoguelph.ca (Electronic Communal Temporal Lobe  -  ECTL)
Subject: ECTL 2.10.01 (speech mail list)
Message-ID: <9210061858.AA01888@snowhite.cis.uoguelph.ca>
Sender: <ectl-sub@snowhite.cis.uoguelph.ca>
Organization: Indiana University
Date: Tue, 6 Oct 1992 18:58:16 GMT
Lines: 107


		Electronic Communal Temporal Lobe
		=================================



Issue: 2.10.01

========================================================================

(1) Roland Seidl, Telecom Australia Research Labs, Australia; r.seidl@trl.oz.au
       Subject: Performance Statistics Problem

(2) Peter Silsbee, Univ. of Texas, Austin, U.S.A; silsbee@vision.ee.utexas.edu
       Subject: References for small but difficult ASR tasks

(3) David Leip, U. of Guelph, Canada; ectl-request@snowhite.cis.uoguelph.ca
       Subject: A few notes of interest reqarding ECTL

========================================================================

1)

I tried sending this problem to sci.math.stats but with no luck. Perhaps the
ectl mailing list can help.

The problem is the following:

I have a database of approximately 240 speakers, saying a vocabulary of 19 words
twice each. Using say 200 of the speakers to train a speech recognizer, and the
remaining 40 to test the recognizer it is possible to derive a "performance"
figure. If a different set of 200 is used to train the recognizer and the
remainder to test the recognizer another "performance" figure can be obtained.

This procedure could be done 6 times, with 6 completely independent test sets. 
However the training sets are not independent.

The question is....
  How could the above performance figures be aggregated to derive a performance
measure and associated confidence level? How valid is it to merely assume that
the training sets are independent, and aggregate the statistics in the usual
way?

Any references would be appreciated!

Please e-mail replies to:     r.seidl@trl.oz.au

Thanks in advance

----------------------------------------------------------------------

2)

Hello folks,

        I would appreciate any references to ASR results for "small
but difficult" tasks such as e-set or minimal-pair discrimination
(speaker-dependent and independent).

Thanks in advance,
Peter

----------------------------------------------------------------------

3)

A few notes of interest:

ECTL has an anonymous ftp archive which is located at snowhite.cis.uoguelph.ca
[131.104.48.1] in /pub/ectl. Included in the archive are: all issues of ECTL,
as well as a list of subscribers, lists of speech related products (s/w & h/w),
and a list of speech related technical report abstracts. If you would like to
contribute to the archive, please mail: ectl-sub@snowhite.cis.uoguelph.ca
If you need information about how to ftp, or such things, please send mail to:
ectl-request@snoehite.cis.uoguelph.ca

The vote for comp.speech has passed, and comp.speech is now in existance. All
issues of ECTL will be carried there as well.

ECTL hasd now grown to over 550 subscribers from 29 countries, with 10 
redistribution sites (Australia-2, Germany-1, Italy-1, U.K.-4, U.S.-2).
Countries presently "served" by ECTL are:

Australia       Austria         Belgium         Brazil          Canada
Columbia        Denmark         Finland         France          Germany
Greece          Hong Kong       Ireland         Italy           Japan
Korea           Netherlands     New Zealand     Norway          Portugal
R.O.C.(Taiwan)  Russia          Singapore       South Africa    Spain
Sweden          Switzerland     United Kingdom  United States

=========================
End.


+-----------------------------------------------------------------------+
|									|
|		     Electronic Communal Temporal Lobe			|
|									|
|  Moderator:  David Leip, University of Guelph, Computing & Info. Sci. |
|		Guelph, Ontario, Canada  N1G 2W1.   			|
|		(519) 824-4120  ext.3709 (office) ext.4297 (lab).	|
|									|
|  Send submissions for ECTL to: ectl-sub@snowhite.cis.uoguelph.ca	|
|  Send all other requests to: ectl-request@snowhite.cis.uoguelph.ca	|
|  Anonymous ftp archive at: snowhite.cis.uoguelph.ca  in: pub/ectl     |
|									|
+-----------------------------------------------------------------------+
