Re: Autocorrelation-like representations and mechanisms

[Christian Kaernbach <chris@xxxxxxxxxxxxxxxxxxxxxxxxxx>]

Peter Cariani wrote:
> most pitch phenomena do behave in a manner that is consistent with
> an autocorrelation-like representation
...
> discard all-order interspike interval representations on the basis
> of very limited evidence that is interpreted in the most shallow way.
...

Peter,

You can imagine that I am once more disappointed to see how evidence
against AC models is discarded as limited, and its interpretation as
shallow. You name quite a number of phenomena that are accounted for by
AC models. But science is not democracy: Evidence against a mechanism
can not be counter-balanced by any number of studies where AC models do
well. 

Here is a short summary of the evidence that is not compatible with AC
models (for references see below):
        Carlyon (1996) demonstrates that mixtures of unresolved harmonics do
not elicit two pitches, not even one, but just a crackle percept. AC
models would predict that both pitches would be audible.
        Kaernbach & Demany (1998) show that second-order intervals are much
less detectable than first-order intervals. AC models would not predict
such a difference.
        Kaernbach & Bering (2001) replicate that finding with much lower cut
frequencies, avoiding the criticism that this might concern only loci on
the basilar membrane where temporal processing might be less important.

It has often been said that our data disprove waveform AC, but not
realistic auditory nerve fiber AC. But why then don't AC modelers test
their models with second-order click trains and prove that they are
compatible with the above data data? Pressnitzer et al. (2002) compared
all-order interval (AOI) and first-order interval (FOI) statistics for
the stimuli of the K&D study. Their figure 5 shows clearly that there is
a big difference for ABX and X stimuli in both AOI and FOI statistics.
So neither FOI nor AOI of auditory nerve data explain the low
detectability of ABX stimuli. My guess is that if one takes into account
the volley principle of temporal coding, by implementing a coincidence
stage after the hair cell model, one would be back close to the waveform
AC, with FOI showing no difference between ABX and XXX.

This is not to say that AC models don't serve well for anything. They
may be regarded a convenient short-cut to describe temporal processing,
and as long as we don't have a detailed account compatible with all
evidence (including the one you dismiss) it might be considered the best
model we have. But we would have to keep in mind: It is a model, it is
not the truth. There is evidence that it can not be the truth.

Best regards,
Christian

        REFERENCES
        ==========

Carlyon, R. P. (1996). Encoding the fundamental frequency of a complex
tone in the presence of a spectrally overlapping masker, J. Acoust. Soc.
Am. 99, 517–524.

Kaernbach, C., Bering, C. (2001). Exploring the temporal mechanism
involved in the pitch of unresolved harmonics, Journal of the Acoustical
Society of America 110, 1039-1048.

Kaernbach, C., Demany, L. (1998). Psychophysical evidence against the
autocorrelation theory of auditory temporal processing, Journal of the
Acoustical Society of America 104, 2298-2306.

Pressnitzer, D., de Cheveigné, A. Winter, I.M. (2002). Perceptual pitch
shift for sounds with similar waveform autocorrelation. Acoustics
Research Letters Online 3, 1-6.
http://ojps.aip.org/getpdf/servlet/GetPDFServlet?filetype=pdf&id=ARLOFJ000003000001000001000001&idtype=cvips