Waverly wrote:

[...] I see the weighting begins to change somewhat dramatically as 
nyquist rate is approached ( 1/2 sampling frequency ).  I've read 
that is to be expected but I am not sure that should be true always.

Dear,

there are several ways to explain this and I myself recently wondered:

"I wonder how the filter will look like for 8kHz sampling frequency,
i.e. at most 4kHz signal frequency, because it is defined up to at
least 20kHz."

A simple explanation is that no matter how the digital filter is 
realized, it is made to influence the signal samples, i.e. the values 
of the filter kernel must have the same sampling interval as your 
signal. That said, the filter kernel must conform to the same 
sampling conditions as the signal, i.e. it must at least be 
bandlimited to the Nyquist frequency.

 From my above "wondering" it is evident that any weighting filter for 
sampling frequencies, let's say, smaller than 96kHz, isn't a true 
A-weighting filter, because the latter is defined way beyond 20kHz.
Now the question arises: If your signal is sampled at 40.1kHz, then 
there should be no frequencies beyond 20kHz, so why worry about the 
A-weighting filter function beyond this limit?

A related question is, at what frequency was your audio signal 
sampled? Are you still free with choosing the sampling?
If yes, use 96kHz or more to get the best digital approximation of 
the A-weighting filter. (Don't forget to properly bandlimit your 
signals _before_ the sampling, i.e. by an analog lowpass!)
If not, you may consider digital up-sampling of your signals which, 
if done correctly, doesn't hurt your signals, but provides "spectral 
room" for the high-frequency roll-off (slope) of the A-weighting 
filter.
You must decide if this additional accuracy in the spectral region 
near the Nyquist frequency is important.

HTH
-- 

                   Herbie

          ------------------------
          <http://www.gluender.de>