Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!doc.ic.ac.uk!uknet!pipex!howland.reston.ans.net!agate!iat.holonet.net!nickm
From: nickm@iat.holonet.net (Nicholas G. Marino)
Subject: Define Covariance matrix, please...
Message-ID: <CELszL.Bp9@iat.holonet.net>
Organization: HoloNet National Internet Access System: 510-704-1058/modem
Date: Sat, 9 Oct 1993 00:21:19 GMT
Lines: 14

I've often seen the term covariance matrix used in speech recognition
literature relating to distance (distortion) measures for feature vectors.
I'm familiar with LPC calc by the covariance method, but I think this
covariance matrix is something else. Can someone give a definition?

My practical purpose is to improve the acuracy of a vector quantizing
speech recognizer. Assume all my templates are represented by strings
of VQ indices. The recognizer then matches input feature vectors against
the templates by comparing each feature vector with the entire codebook.
Right now, I'm using a simple euclidian distance. Would computing a
covariance matrix for my codebook yield better results?

Many thanks in advance.

