[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: AUDITORY Digest - 15 Jan 2003 to 16 Jan 2003 (#2003-13)

I would look at GA Millers work:

author={Miller, G. A. and Heise, G. A. and Lichten, W.},
title={The intelligibility of speech as a function of the context
       of the test material},
journal={J. Exp. Psychol.},
pages={329-335} }

,author={Miller, G. A.}
,title={Decision units in the perception of speech}
,journal="IRE Transactions on Information Theory"
,note_={Grammar = 4 dB of SNR; is it just a first order grammer?
       It is less than 5th.} }

,author={Miller, G. A. and Isard, S.}
,title={Some perceptual consequences of linguistic rules}
,journal="Jol. of Verbal Learning and Verbal Behavior"

I have e versions of the last two of these, but there are copyright
issues, I suspect.
I will trade my last two for anyone who can get me the first one in
scaned form.

Claude Shannon also did some work on this, but I doubt that is what you
are looking for here.

Automatic digest processor wrote:

Date:    Thu, 16 Jan 2003 09:37:22 +0100
From:    =?X-UNKNOWN?Q?T=F3th_L=E1szl=F3?= <tothl@INF.U-SZEGED.HU>
Subject: Re: Phoneme discrimination tests

On Thu, 16 Jan 2003, Hugo de Paula wrote:

Most of discrimination tests that allow the contruction of a phoneme based
confusion matrix uses nonsense words like DRT and MRT. Does anyone know of a
database and/or test procedure that allows us to find this kind of results
using rhyme words or sentences that are part of the language instead of using

I am just wondering whether nonsense speech testing has ever been extended
to sentences. That is, has anybody ever made an algorithm that
creates nonsense sentences "in a given language". By this restriction I
mean that
1. Some basic statistics of the generated sentences (e.g. phoneme
statistics, word-length statistics) should fit that of the language.
2. The generated sentences should obey the phonotactic rules of the
3. The edit distance between the words of the generated text and any real
word of the language should by larger than a certain threshold.

              Laszlo Toth
       Hungarian Academy of Sciences         *
 Research Group on Artificial Intelligence   *   "Failure only begins
    e-mail: tothl@inf.u-szeged.hu            *    when you stop trying"
    http://www.inf.u-szeged.hu/~tothl        *


End of AUDITORY Digest - 15 Jan 2003 to 16 Jan 2003 (#2003-13)

Jont B. Allen,   908/654-1274voice; 908/789-9575 fax
382 Forest Hill Way, Mountainside NJ 07092
http://auditorymodels.org/jba; jba@auditorymodels.org; JontAllen@ieee.org

Hegel was right when he said: "We learn from history that man can never
learn anything from history." - GB Shaw