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

Re: need MDS discussion



Hi everybody,

Along the line I recommend Marozeau's paper which present a technique (using
MDS) enabling to compare timbre of music instruments independantly of their
pitch ( to a certain extend, that is mainly inside the same octave) :

Jeremy Marozeau, Alain de CheveignÃ, Stephen McAdams and Suzanne Winsberg
(2003), The dependency of timbre on fundemental frequency, JASA 114-5, 2003
http://recherche.ircam.fr/equipes/perception/marozeau/marozeau2003.pdf


FrÃdÃric Maintenant
http://www.cdmc.asso.fr/biographies/m_q/maintenant.htm

----- Original Message -----
From: "Amit Gal" <hobbit@xxxxxxxxxxxxxxxxxxxxx>
To: <AUDITORY@xxxxxxxxxxxxxxx>
Sent: Tuesday, April 05, 2005 8:41 AM
Subject: Re: need MDS discussion


hi,

personally, i'm not fond of MDS techniques for modeling similarity ratings.
as
far as i experienced and read, similarity judgements tend to be too much
context dependent (so sometimes the judgements you get for "a is similar to
b"
and "b is similar to a" are different, and it may be sensitive to
"irrelevant
options" and sensitive to the experimental methodology itself). you can read
more about this in many of robert goldstone's papers on similarity
for example you can look at:
http://cognitrn.psych.indiana.edu/rgoldsto/pdfs/similarity2004.pdf

however, for timbre, there are some known applications of mds for example:
Grey, J. M. (1977). Multidimensional perceptual scaling of musical timbres.
Journal of the Acoustical Society of America, 61, 1270-1277.

some of stephen mcAdams researches on musical timbre also use MDS. stephen
mcAdams told me that forthe specific case of timbre most of the complexities
of
similarity do not occur, and that he thinks it is safe to use it.  however,
i
didn't see evidence on that, so i can't comment whether i believe it or not.

mathematically speaking, the concept of MDS is a tricky one, as there are
many
ways to play with the number of dimensions without distorting too much the
original "distances" (that came from examinees responses). this means that
there
is no one "true" answer even about the number of dimensions needed to model
the
results.

on the other hand it does give you a simple graphical way to understand your
results and interpret them.

so if you ask me, i would advise you to use mds with lots of caution.

hope this helps (and not too discouraging)

amit.

Quoting beaucham <beaucham@xxxxxxxxxxxxxxxxxxxxxx>:

>
> We are looking into doing a multidimensional scaling test
> on a group of sounds and would like to read about the pros
> and cons of using subject rating of the similarity of pairs
> of sounds (e.g., on a 0 to 10 scale) vs. judging which of
> two pairs presented in rapid succession is most similar.
> Please let me know if you know of a reference which discusses
> this point.
>
> Jim Beauchamp
> Univ. of Illinois at Urbana-Champaign
> jwbeauch@xxxxxxxx
>




----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.