()

Subject: 
From:     <>
Date:    


BAD MSG:
whom I've not yet been in touch), it seems that profile
nalysis has only been systematically studied up to about
1 frequency component per CB? For instance, in Green, D.M.
(1992). "The number of components in profile analysis tasks,"
J. Acoust. Soc. Am. 91, pp. 1616-1623 one finds that most
of the work was on up to 21 (and sometimes 24) log-spaced
components in a [200,5000] Hz interval, supplemented by
just two points at 41 and 81 components (his Fig. 1).

In my experience, a lot more than mere masking seems to
be happening when stepping from 41 through 81 components
on a log scale (where my default interval is [500,5000] Hz
to get a fair time-frequency compromise in case of dynamic
profiles - although I'm in this posting referring only to
static profiles). Not only does tonality appear to decrease
as one would expect, but beyond, say, 70 or so components
an unpleasant roughness becomes rather prominent, and going
to, say, 120 components sounds pretty bad to me.

Now I can imagine that this roughness can be explained
from intra-CB beats as one moves beyond the 1 component
per CB density. However, shouldn't this intra-CB beating
also have significant effects on profile discrimination
thresholds? Shouldn't this give strong dependencies, and
possibly more structure (maybe even minima an maxima)
than just a simple rising slope, for change detection
threshold as a function of the number of frequency
components as one moves in small increments from about
1 component per CB up to say 3 or 4 components per CB?

In principle I can imagine that the beating could help
to (learn to) perceive intensity changes in individual
components, because the beating changes with the intensities
of components within a single CB. Normally one would want
to separate the effects of beating from other frequency
detection effects for the sake of auditory research (the
notched-noise methods being based on such considerations).
In my quest it is the other way around: in the end I don't
really care what mixture of effects is used in hearing as
long as it gives maximum detectability of profile changes
(though I love to understand things like everybody else).
If intra-CB beats can help that's fine. Currently I use
about three components per CB, and would like to learn
more about profile analysis research around that density
range.

Have these things been systematically explored somewhere
in the context of profile analysis? Basically I'd like
to see more points in Fig. 1 of Green's interesting 1992
paper in the range between m=41 and m=81 sinusoids, and
to see an additional axis (nice 3D plot!?) showing the
threshold changes with training time.