Re: Rationale for Critical Bands

["Richard F. Lyon" <DickLyon@xxxxxxx>]

A follow-up, in case anyone still cares...

I got a copy of Ehret and Romand's book "The Central Auditory 
System", with Ehret's chapter "The Auditory Midbrain, a 'Shunting 
Yard' of Acoustical Information Processing", to see what he says 
about critical bands in 1997.  It's a fascinating chapter, with lots 
of good stuff.
On p.277, "In accordance with psychophysical terminology, these 
bandwidths of spectral integration or spectral resolution are called 
critical bands."  That is, he adopts the term critical bands, for 
something that is not psychophysical, and is not measured with an 
approach that much resembles the paradigms for the psychophysical CB 
measurement.
and "the tuning curves for complex sound analysis of most neurons 
(62%) in the ICC of the cat remain rather constant in width when the 
intensity of the sound is changed."  This seems to be the association 
of CB as level independent, but the case is not made that the CB and 
this curve, whatever it means, are in some sense the same phenomenon. 
It he meant to claim that, he probably would have said so.
On p.278, "Whether neurons in more peripheral auditory nuclei have 
tuning curves for complex sound analysis with the features mentioned 
above remains to be determined."  That is, he has backed away from 
the earlier conclusion that such curves (whatever they are, that he 
associates with CB) arise first in ICC.
Then he refers to the "receptive field size (the critical bandwidth) 
of ICC neurons in the spectral domain."  Again, mixing the 
physiological and psychophysical terminology, as if they're the same, 
but not really making a claim that they are.
It's quite possible that these neurons do extract some kind of 
spectral feature in a fairly level-independent way, and that may be 
closely related to some kinds of critical band phenomena.  But for 
other phenomena, such as detection of tones in notched-noise maskers, 
and other psychophysical tests, the bandwidth must vary a fair bit 
with level to fit the data, and the filter shape is nothing like 
rectangular, which is what he seems to be suggesting the shapes of 
the ICC tuning curves show.
In my opinion, applying ideas like spectral domain, that are so 
intimately tied to the dynamics of linear systems, to the nonlinear 
neural processing several stages beyond the periphery is bound to be 
ambiguous and unsatisfying for understanding the function.  In the 
periphery, where the nonlinearities can be understood and accounted 
for well enough, linear systems concepts such analysis by sinusoids 
and noise bands can be applied with reasonable success, if care is 
taken in not extrapolating to different levels and such.  But in the 
ICC, observations about the tonotopy and response curves really don't 
have any directly interpretable relationship to things like critical 
bandwidths.  That opinion probably explains why I always associated 
critical bands with peripheral filtering -- that is, after all, where 
the frequency response, and its level dependence, is shaped, as Ehret 
said in the bit I quoted before.  I believe that's the prevailing 
interpretation:  the CB reflects cochlear filtering, and is 
moderately level dependent.
Dick

ps. I got no replies to my query for finding the most relevant source 
from Ehret, at the send of this snippet, so I'm still hoping someone 
has it:
At 11:43 AM -0700 6/15/10, I wrote:
> I've been reading up on the concept of critical band in IC, which is 
> what you've been talking about.  I have this book on IC, which one 
> can read parts of online:
> http://books.google.com/books?id=Ift_WTgSKqMC&pg=PA327
> The section "Origins of Spectral Tuning and Resolution" in the 
> chapter by Ehret and Schreiner starts off with something that I 
> think we can all agree on:
> "Cochlear filtering represented in the RFs of auditory nerve fibers 
> is the basis for spectral tuning in higher centers of the auditory 
> pathways."
> But then they get into an interpretation of "critical bands" that's 
> one I'm not familiar with, but probably is what you are talking 
> about, when they say on p.328 "The origin of neural critical 
> bandwidths with properties of psychophysical critical bands is 
> associated with ICC neural filtering properties" and "The steep 
> slopes of the critical-band borders...".  On p.325 they say 
> "Perceptual critical-band properties are absent in auditory nerve 
> fibers (Ehret 1995)"; so I'll guess I'll have to find that and see 
> what they mean.  It's in a book that I can't find a copy of:
> Advances in Hearing Research: Proceedings of the 10th International 
> Symposium on Hearing : Swabian Conference Centre, Irsee, Bavaria 26 
> June-1 July, 1994, Geoffrey A. Manley, G. M. Klump, C. Koppl, H. 
> Fastl, H. Oeckinghaus (eds.); anybody have a copy?