Binding through synchrony (was: An Auditory Illusion) (Alain de Cheveigne )

Subject: Binding through synchrony (was: An Auditory Illusion)
From:    Alain de Cheveigne  <alain(at)LINGUIST.JUSSIEU.FR>
Date:    Mon, 26 May 1997 03:55:28 +0200

At  4:37 PM 97.5.22 -0400, Al Bregman wrote:
>Here is a message sent to me to be forwarded to the list.
>From: Peter Cariani <peter(at)epl.meei.harvard.edu>
...
>An alternative (or complement) to binding-through-synchrony
>mechanisms is binding through common time (phase) structure.
>When we have a harmonic complex with a mistuned component
>or when we have two harmonic complexes (double vowels)
>with different F0's, the time (phase) relations within
>each object (complex vs. mistuned component; vowel1 vs. vowel2)
>are constant from one fundamental
>period to the next. The time (phase) relations across objects,
>however, are constantly changing. I think that any mechanism
>that groups by common time pattern from period to period should be
>able to segregate out multiple objects this way (Patterson's
>strobed temporal integration model, JASA 98(4);1890-4, 1995
>is in the right direction, but I'm not sure how well the
>triggering algorithm would handle multiple objects with
>different F0's).


Here's a bit of cyberpromotion of my own pet theory of harmonic
cancellation:

Suppose you have two periodic vowels mixed together.  The auditory
system could use the temporal regularity of repetition of one vowel to
add up neural information and enhance that vowel's representation,
relative to the other vowel whose correlates would be out of phase.
For example Patterson's STI model could trigger in synchrony with
the target vowel and accumulate an enhanced temporal representation
of the target vowel in its buffer.  This is an example of a class of
"harmonic enhancement" models that take advantage of the
harmonicity of a target to improve its representation.  So far I
know of little experimental evidence to support harmonic
enhancement, and there is quite a lot of evidence against it.

You can also use the regularity of the competing vowel to get rid of
it.  For example one might tune Patterson's model to strobe in
synchrony with the competing vowel, and subtract from the buffer
rather than add to it.   Segregation would thus depend on the
harmonicity of the competing vowel, and there is a lot of evidence
that that is indeed the case.  However there are problems with
repeated subtraction from the same buffer, and it is also perhaps
not so easy to derive a reliable strobe pulse from the mixture of
two vowels.

These problems can be solved by using a delay-and-subtract
network: the neural representation is subtracted from itself after
a delay equal to the competing vowel's period.  The remainder of
the subtraction represents the target vowel.  This delay-and-subtract
circuit is similar to that suggested by Licklider and incorporated in
Meddis and Hewitt's models (of pitch and concurrent vowel
segregation), but the excitatory interaction is replaced by inhibitory
interaction.  Remains the problem of estimating the appropriate
delay (the competing vowel's period).    It turns out that a simple
criterion of minimum output works quite well.  That is, an array of
delay-and-subtract circuits is scanned for minimum output, and
evidence of the target vowel is taken from the channel in which this
minimum occurs.

This is a workable example of a binding-through-synchrony
mechanism as suggested by Peter.

Alain

Alain de Cheveigne, CNRS/Universite' Paris 7, alain(at)linguist.jussieu.fr