Re: purely spectral pitch (Katrin Krumbholz )


Subject: Re: purely spectral pitch
From:    Katrin Krumbholz  <katrin.krumbholz(at)MRC-CBU.CAM.AC.UK>
Date:    Tue, 31 Oct 2000 17:46:23 GMT

Dear Peter, I think, I have to defend John's position a bit: Let me start off be explaining Roy and my experiments in more detail. Our study on pitch perception in binaural listening conditions was devided into two parts. The first part deals with dichotic pitches, namely Huggins pitch and its complex relative, Multiple-Phase-Shift (MPS) pitch. As you know, a MPS pitch can generated by introducing narrow phase transitions at harmonic frequencies in an otherwise diotic noise. The auditory nerve responses to a spectrally unresolved harmonic complex tone still mediate information about the repetition rate of the stimulus: when harmonics merge within the cochlear filters, they interfere and produce a temporal modulation (beating) at the output of the filters. This is not the case in a spectrally unresolved MPS stimulus (phase transitions merged within cochlear filters), because the signals at the two ears are just noises, so there is no beating between harmonics. Thus, even if dichotic and monaural pitches would be processed with different contiguity or integration windows, there would be no temporal information to process in a spectrally unresolved MPS stimulus! We measured the lower limit of pitch (LLP) for MPS stimuli and found that the pitch limit is about 65 Hz, which (as expected) approximately corresponds to Shackleton and Carlyon's (1994) estimate of the limit of spectral resolution in the relevant frequency region (between about 200 and 1200 Hz). For a comparison, we also measured the LLP for homo- and antiphasic (S0 and Spi) harmonic tones in a diotic noise masker. The loudness of the harmonic tones was matched to the loudness of the MPS pitch. The LLP for the homophasic harmonic tones was about 35 Hz, i. e. well below the limit of spectral resolution. In contrast, the LLP for antiphasic harmonic tones was essentially equal to the LLP for MPS stimuli. This finding is a bit astonishing, because the antiphasic tones should produce temporal modulation of the cochlear filter resonses when harmonics are unresoved (i. e. below about 65 Hz in our frequency region), and this modulation information should theoretically be interpretable by a temporal pitch mechanism. Due to the low SNR of the antiphasic tones, the temporal modulation would be burried under a lot of noise but there should be no reason why the temporal information shouldn't be recovered by binaural unmasking. Or should there? Yes, namely binaural sluggishness. In the second part of our study we measured the temporal modulation transfer function (TMTF) for homophasic and antiphase bandpass noise (filtered to the frequency region of interest, namely between 200 and 1200 Hz) in a diotic noise masker. The sensation level of the antiphasic noise was about 12 dB and the sensation level of the homophasic noise was set so that the modulation detection threshold at low modulation rates would be roughly equal in both interaural phase conditions. The result was that the TMTF for the antiphasic noise had a much lower 3-dB down cutoff rate than the TMTF for the homophasic noise, and in the antiphasic noise, modulation was completely inaudible at rates above 30 Hz, which would be relevant for pitch perception. Thus, the reason for the similar LLP for MPS stimuli and for antiphasic harmonic tones is that the temporal modultion in the cochlear filter resonses to a spectrally unresolved antiphasic harmonic tones is NOT recovered by binaural unmasking, becuase binaural processing is too slow. Now, if the binaural system is too slow to follow the envelope modulations in the cochlear filter resonses to unresolved harmonic tones, it is probably even more too slow to follow the temporal fine structure in cochlear filter responses to spectrally resolved harmonics. At least with the current concepts of binaural sluggishness in mind (temporal integration window with duration of the order of 100 ms), there is no reason to believe that envelope and fine-structure temporal information should be treated differently by the binaurla system. So under the assumption (and I concede that this is an assumtion) that fine-structure and envelope are both temporally smeared to the same extent, the conclusion would be that dichotic pitches and the pitch of binaurally unmasked harmonic (or pure) tones are PURELY spectral pitches. Finally just one short remark on your comments about frequency selectivity and the distinction between physiological and phsychophysical data: Ted Evans has produced a lot of quite compelling evidence that the frequency selectivity of the mammalian auditory system, as measured in phsychophysics, is already established at the level of the cochlear! Best regards, Katrin Krumbholz


This message came from the mail archive
http://www.auditory.org/postings/2000/
maintained by:
DAn Ellis <dpwe@ee.columbia.edu>
Electrical Engineering Dept., Columbia University