Subject: Re: Gold & Pumphrey (Experiment I) From: Andrew Bell <andrew.bell(at)ANU.EDU.AU> Date: Mon, 20 May 2002 14:15:23 +1000Thankyou to Glenis Long for pointing out the importance of threshold microstructure. I agree with the implications of microstructure as interpreted by Dan Ellis. If G&P happened to have used frequencies near microstructural peaks, or their signal sidebands were more than 15 dB down anyway, then their Q calculations appear valid. Interestingly, microstructure was only recognised a decade after G&P's experiments (Elliott, 1958: 'A Ripple Effect in the Audiogram', Nature 181, 1076), but given G&P's concerns, it is surprising (with hindsight) that they didn't come across it themselves. There are some wonderfully detailed examples of microstructural threshold in Glenis Long's 1984 paper (Hear. Res. 15, 73-87). The figures show that the peaks and valleys are separated by about 100 Hz in the region near 2 kHz. Note that the 'valleys' (lowest threshold) are in fact the regions of peak sensitivity, and so the frequencies of the putative resonators corresponds to the frequencies of the valleys. The -3dB widths of the valleys are about 20 Hz, so the Q of these peaks is about 2000/20 = 100. This value is in line with the Q of 80 derived by G&P (although it is greater than the values of 15-20 derived by Christopher Shera with SFOAEs, which were discussed earlier on this list). The simplest interpretation of microstructural thresholds is that the ear is using a graded bank of resonators, the very notion that G&P were putting forward and the same perspective implied by Dan Ellis' comments. G&P were promoting Helmholtz's idea that sound is detected using sympathetic resonance, and this is a position to which I would also subscribe. Perhaps the strongest evidence supporting G&P comes from experiments on microstructural threshold conducted by Thomas in 1975 (JASA 57, S26). Amazingly, he found that, for stimuli just above threshold, the perceived _pitch_ of the tone stays _constant_ within the passband of each microstructural peak. That is, each microstructural passband is associated with a unique pitch, and as one steadily increases the frequency the pitch jumps quantally as one moves from one valley to the next. To my knowledge, this finding, pointing to a fundamental unit of pitch perception and identifying each passband with a single resonator, has not been followed up. Andrew. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Andrew Bell Research School of Biological Sciences Institute of Advanced Studies Australian National University Canberra, ACT 0200, Australia andrew.bell(at)anu.edu.au phone +61 2 6125 9634 fax +61 2 6125 3808 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ >>> Dan Ellis <dpwe(at)EE.COLUMBIA.EDU> 05/19/02 12:33am >>> >> [Glenis Long said] This proviso rules out most frequencies in a normal cochlea since >> there is almost always some threshold microstructure and often pronounced >> microstructure of up to 15 dB changes within 1/2 critical band. >This is an interesting twist, new to me. I'd like to know more details. > >Presumably it would be OK if the tests were done with the most >sensitive frequency in the locality - if that can be easily >established for a particular listener - since then we can be confident >that the sidelobes are below threshold. > >But, depending on the fineness of the microstructure, doesn't >microstructure in itself imply high-Q? If frequencies close to one >another are being processed differently (i.e. with different sensors >exhibiting different thresholds), then the individual sensor >structures (mechanical and neural) must be finely enough tuned to >separate these close frequencies, implying high-Q. Half a critical >band wouldn't give you a Q of 300, but the details might be telling. >