Subject: Re: place pitch and temporal pitch From: Christian Kaernbach <mailinglist(at)KAERNBACH.DE> Date: Fri, 19 Mar 2004 14:02:18 +0100Dear Eckard: >>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? > > To some extent yes, I think so. Pitch is obviously a unitary perception. That is an important achievement of our brain: To present to us as unitary perception what is deduced from different cues. Another example would be spatial hearing: There are intensity differences, interaural time (!) differences, spectral (!) filtering by the outer ear, and even cues due to involuntary small head movements that interact perfectly so as to give a single percept of stereolocation. So pitch being a unitay percept does not rule out its relying on separate mechanisms. (I am aware that there are other positions in psychology, such as direct perception etc., that would strongly object to such "constructivist" views, and I would not claim truth for this position, but it is one of my several working hypotheses [I switch them as I please ;) ]) > Furthermore, merely a faint pitch remains when you excluded the spectral > code. Not by chance, place code is clearly the dominant one. I did not say anything about relative importance. There seem to be two cues, both are in general present, and both will be exploited (and contribute to a unitary percept). >>In K&D 1998 (see below) we presented evidence against AC models >>based on AC of the raw sound waveform. > > Understanding your reasoning quite well, I can nonetheless not confirm > that. Your filters excluded the spectral code. Consequently, the admittedly > hidden two-stage autocorrelation of the raw sound waveform was also > excluded. This caused the discrepancy between your correct conclusion that > there is no autocorrelation and Peter Cariani's claim that autocorrelation > plays an important role. Do not be disappointed. You discovered what > remains without spectrally based autocorrelation. Oh, Eckard, I would not be disappointed if we two would continue to disagree. Such things happen. As to this attempt to "save" AC via a two-stage model including both spectral and temporal processing: No, it does not work. I said that natural stimuli do have both cues, but they don't need to have all of them along all frequencies involved. Natural stimuli will more often than not comprise unresolved harmonics with no spectral cues. And if we look on what is happening with resolved harmonics: There is simply no way to tell whether the mechanism dealing with them are able or not to do higher-order statistics. Resolved harmonics transmit only periodic approximations of the original stimulus, and in periodic sequences all-order and first-order statistics give the same information. It would be a parsimonious (and hence good) theory to assume that the temporal mechanism involved with resolved harmonics is operating very similar to that working with unresolved harmonics. I would not expect Peter to be especially upset about my last posting, as I was (I think so) very careful to give AC models their appropriate place, which is at least as important as those famous spectral models: All have their value to explain what they were made to explain, and all have their limitations. Best, Christian