Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!iuvax!cica!tut.cis.ohio-state.edu!ucbvax!decwrl!shelby!csli!poser
From: poser@csli.Stanford.EDU (Bill Poser)
Newsgroups: comp.unix.wizards
Subject: Re: FCC doing it again...
Message-ID: <11637@csli.Stanford.EDU>
Date: 8 Jan 90 02:13:19 GMT
References: <21728@adm.BRL.MIL> <11416@csli.Stanford.EDU> <3883@levels.sait.edu.au>
Sender: poser@csli.Stanford.EDU (Bill Poser)
Reply-To: poser@csli.stanford.edu (Bill Poser)
Organization: Center for the Study of Language and Information, Stanford U.
Lines: 13

In article <3883@levels.sait.edu.au> CCDN@levels.sait.edu.au (david newall) writes:
>What comes out of this "theoretically ideal" technique would sound like me?
>Or is it not "theoretically" useful to be able to recognise someone's voice?
>Humbug!

None of the relatively highly compressed speech that I have heard seemed
to preserve much information about the speaker's identity. Unfortunately,
very little is known about the locus of speaker-specific information
in the signal, so it is impossible to make a good theoretical estimate
of the minimum amount of information necessary to transmit speaker
identity as well as the linguistic content. Presumably one could
set an upper bound by determining how many different voices a speaker
can distinguish. To my knowledge this has not been done.