Coclear vs. psychoacoustic models

[Lars Bramsløw <LBRAMSLOW@xxxxx>]

Dear Jont + List,

A few belated replies to your comments, received july 11:

> Isn't it also important to simulate recruitment and
> the variation, if any, of the critical ratio with level?
Simulation of recruitment and critical ratio with level was dealt with
in the model, indirectly.  The roex-filters were widened according to
level, giving loudness recruiment and reduced frequency selectivity.
Critical ratio is a result of 1) frequency selectivity, which was
simulated and 2) what Glasberg and Moore term 'detection efficiency',
i.e. at what S/N ratiowithin the band the tone is audible.  Zwicker
terms this 'Schwellenfaktor'.  This 2nd factor was not modified, if
that's your question.

> What about Fletcher and Munson's 1933 loudness model paper,
> Munson and Gardner's 1950 forward masking loudness model paper, and
> Steinberg and Gardner's 1937 model of recruitment?
> These give ample information to do what you want, dont they?
I don't know.  I have not consulted those papers during the model work,
as I wasn't aware of them.

> I guess I still dont know the difference between an ERB and a critical
band.
> Can somebody help me?
The concept of the bark and the ERB scale is the same, however based on
the roex-filter model, the ERB scale has more bands, roughly 32 in the
range 100  - 10 kHz.  There are 22 critical bands in the same range. The
ERB bandwidth is ERB(Hz) = 24.7(4.37*fc(kHz) + 1), thus the bandwidth
minimum is 35 Hz (Critical band is 100 Hz) and the high-frequency
bandwidth is roughly 10% of the centre frequency (Critical band is appr.
20 %).  This is just the short version, others may fill in.

> This is an excellent starting point if you are trying to make
> objective measures as applied to hearing aids. If you want to
> model the cochlea, this model doesn't help much.
No, of course not!  That wasn't the intention either.

> This is the data that was used to set the parameters of this
> model. I should hope that it would do a good job.
It sounds simple, but it isn't, when the complete model is re-used to
predict its design data, since the roex-model was combined with a mix of
outer-ear corrections and the Zwicker loudness growth function.

> Are you saying that the cochlear models dont work in real life?
Not for this type of task.  As I recall Jim Kates made a cochlear model
with hearing loss with percentage loss of inner and outer hair cells,
but the catch is to specify these percentages correctly to simulate the
psychoacoustic performance of a patient.  Do you have any
counter-examples?



Lars Bramslow responds:

> What problem are you trying to solve? Do you want to build a
> model of the auditory system, or make a new hearing aid?
> I sure didnt try to build a better cochlear model while working
> on a hearing aid design, I was too busy.

I was applying a model of the auditory system for the purpose of
predicting  subjective quality.  There was no hearing aid design
involved whatsoever, despite the affiliation with a hearing aid company.
I was quite busy doing listening tests, auditory modeling, statistical
analysis, data fitting etc!

> The basic research in how the cochlea and auditory system works
> is a difficult puzzle. We do not all agree on how it works.
> But without differing opinions, we cannot make progress.

> The purpose of a cochlear model is to obtain a basic understanding of
> what is going on, not provide a simulation that can be used to
> simulate a hearing aid.

> The Cambridge (Moore et al.) model is based on Zwicker. The Zwicker
model
> is based on Fletcher and Munson (1933). As far as I can tell, the
> Fletcher and Munson model is based on Wegel and Lane (1928).

> The user of a model should not try to "compete" with the designer
> of a good cochlear model, nor should the modeler try to compete
> with the hearing aid designer. Each has an important place, and
> plays a role.

I agree!  The coupling between hearing aid design and more basic
research was absent in my project, just as intended.  I do think, as
previously stated, that the published cochlear models are of limited
PRACTICAL use, but this does not make them less significant as research
tools.

Regards,

Lars Bramslow, August 19, 1997