place pitch and temporal pitch

[Christian Kaernbach <mailinglist@xxxxxxxxxxxx>]

Uwe Baumann wrote:
> The ongoing discussion about temporal pitch emerged the statement
>
> "There is no physiology of a "place pitch", and this concept should be
> given up." by Martin Braun, which was supported by several others.
I saw these statements, and I said to myself...

"But don't I teach to my students that there is clear-cut evidence for
both, for place pitch and for temporal pitch?"

Evidence for place pitch: A sinusoid of 5 kHz elicits undoubtedly a
perception of pitch. There is no phase locking in mammals at that high
frequency, and as it is a sinusoid there is no envelope, so this can
only be place pitch.

Evidence for temporal pitch: A high-pass filtered low-pass masked
periodic click train (e.g. F0 = 125 Hz, Fc = 2000) elicits undoubtedly a
perception of pitch. There is no pattern anymore in the cochlea as the
remaining harmonics are unresolved, so this can only be temporal pitch.

Both these pitches suffer.
- The jnd for filtered click trains goes up from 0.5% to 2.5% (or 1.3%,
if you believe in our results, K&B 2001) as soon as Fc is high enough to
cut away all resolved harmonics. I don't know about studies of jnd for
high-frequency sinusoids. (Surely somebody on the list does.)
- The musicality of filtered click trains is existent, but weak (see
K&B, 2001). I don't know about studies of the musicality for HF
sinusoids, but I have often heard reports of musicians that it is
difficult to hear melodies with these stimuli. (Some report, however,
that it can be learned.) It is interesting to notice that the range of
pitches used in orchestras or any type of instrument usually stops at 4 kHz.

So in conclusion I usually state that the brain would exploit all cues
to pitch that offer themselves, and place as well as time are there so
they are used. And I conclude that pitch is best if the brain can
exploit both cues, as is usually the case with most natural signals. Is
there anything wrong with this view? Please tell me so urgently, as
thousands of students have already been exposed to this view, and more
thousands are to follow...

Let me add a note on modeling. Dick Lyon stated;
> As far as I know, temporal models have succeeded
> more than failed (that is, temporal models of
> processing in and beyond the cochlea, not to be
> confused with temporal processing of raw sound
> waveforms).  Spectral models, while widely used,
> often run into limitations that make them "fail".
"I'm trying to decide whether that comment is directed at me." In K&D
1998 (see below) we presented evidence against AC models based on AC of
the raw sound waveform. With our stimuli (filtered click trains) there
is not much difference between raw waveform and processed waveform, and
those difference that there are are not relevant to our findings. I urge
everybody how doubts this statement to pass our stimuli through his
preferred model of cochlear processing (the different choices one has
there discouraged us to do it ourselves, as surely somebody would have
said "But your cochlear model lacks XYZ") and to see whether he can
explain our results. To my knowledge, no temporal model ever succeeded
to explain our result.  ---   This is a dangerous situation. Modelers
just like to see their models succeed, and they don't jump voluntarily
on paradigms where they sense a certain risk that their model could
fail. The situation is quite similar to that some decades ago: Place
models were the models of choice, and those who demonstrated that they
"succeeded more than failed" would simply use stimuli that worked well
with their model. My idea of science is different: One simply has to
look for supporting as well as for non-supporting evidence, and one has
to try to learn from both.

I am not going to say that AC models are "wrong". All models are "wrong"
in that they make simplifying assumptions. The Ising model of
ferromagnetism (and the equivalent lattice model of van der Waals gases)
are very rude cartoons of reality, but they do a good job to explain
phase transitions. However, it is wise to find out (and remember) the
limitations of a model. Nobody would assume a Hopfield model to explain
the complex wealth of phenomena in human memory. E.g., our ability to
remember sequences of patterns clearly rules out symmetrical
connections. Nobody would expect an AC model to explain all phenomena in
pitch perception. E.g., our difficulty to hear second-order
periodicities clearly rules out all-order statistics. But both, Hopfield
and AC, have deserved their place in history and are still useful for
studying phenomena were [sequences of patterns / higher-order
periodicities] don't play a role.

Best regards,
Christian

References:

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.
http://www.uni-leipzig.de/~kae/publications/2001_kae&ber_jasa.pdf

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.
http://www.uni-leipzig.de/~kae/publications/1998_kae&dem_jasa.pdf