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

Re: Is there considerable phase locking up to 6 kHz?

At 3:08 PM +0100 03/17/2004, Christian Kaernbach wrote:
Dear Chen-Gia Tsai,

I use three stimuli:
A={400*9, 400*11, 400*13} Hz
B={400*11, 400*13, 400*15} Hz
C={400*13, 400*15, 400*17} Hz

If we listen to [ABC], we can hear an ascending melody,
I seem not to get your point. If ABC of the above three stimuli is
perceived as an ascending melody, this is surely due to their spectral
content being at increasingly high frequency values. The mid-frequncy of
A is 4400 Hz, of B 5200 Hz, of C it is 6000 Hz. Why should any simple
"place" model of pitch perception not predict that one perceives a
rising melody in ABC?

Temporal pitch extraction would extract 400 Hz in all three cases (if
operating flawlessly, and all effects of compression etc. put aside) .
Thus, if listening to your temporal pitch extractor (if you were
selectively able to do so) you would hear 400, 400, 400 Hz, i.e. no rise
in pitch (but change in timbre).

Christian Kaernbach

Your question is a good one, and points out how much we assume and
gloss over in these discussions.  First of all, there are at least
two relevant pitches.  The ascending melody is almost certainly just
the "place" pitch as you note.

The "ambiguous musical pitch" is lower, and is the one determined
(presumably) be temporal autocorrelation-like mechanisms.  As you
note, the period would correspond to 400 Hz.  But the envelope period
is 800 Hz.  Complexes of odd-numbered unresolved or
partially-resolved partials tend to have a somewhat ambiguous pitch,
fighting between the low "fundamental" and a couple of values close
to the "difference frequency" or "envelope frequency".  For very high
frequencies like these, the envelope is most salient, since there is
not enough fine structure to notice that after one envelope period
the fine structure is in reverse phase.

Hence the question.  If there is no phase locking or fine structure
synchrony at all, you expect a pitch corresponding to the envelope
frequency 800 Hz.  With some synchrony, you expect a pitch
corresponding to an autocorrelation peak that's shifted from the
envelope period by about a half-cycle of the center frequency (many
many experiments have shown such effects, and subtle variations on
them, but mostly at lower frequencies).

In the present question, it was noted, if I understood correctly,
that the "musical pitch" was not 800, but near there and somewhat
ambiguous.  This is exactly what you would expect if autocorrelation
mechanisms are at work on a small amount of information synchronized
to the fine structure near the envelope peaks.  But it you rule out
synchrony, you have to invent other mechanisms.  That's the step I
object to.

Some of the other responses to this discussion are real red herrings,
I think.  The ideal of "learning" at such a low level of the system
is too far fetched to take seriously.  And the idea that "FCT is the
only realistic cochlear transform" is just a mathematician trying to
force a biological system to be something he can analyze.