Has anyone already applied A-weighting in PD?
How would you start?

https://en.wikipedia.org/wiki/A-weighting

Now I found A-weighting is not the same as equal loudness:
http://www.sengpielaudio.com/calculator-dba-spl.htm

There is no formula for calculating "equal loudness"

equal loudness is a measured curve, tested idealized with sine tones, with no reverberation:
http://www.sengpielaudio.com/Acoustics226-2003.pdf

What is best to adapt such a curve without degrading the sound? With a FIR filter?

@lacuna It is only "measured" in the sense that it is what a healthy human perceives as equal loudness, and it has been improved upon over the years.
You can hand draw your curve into a Hann window to use with a regular Fir filter, but you should be able to program (expr) the curve for the Hann window from the A weight formula.
It has to be easier than building an Eq to match an A weight curve.
http://www.katjaas.nl/minimumphase/minimumphase.html ........ the [pd spectfilter~] part.

But if you are using this with external components like a microphone and speakers then you need to compensate for their foibles and you could use 07.adaptive_equalisation from the adaptive library to make an overlay for the Hann A-weight curve.
I.e. multiply by the A weight Hann window and multiply again by the inverted curve from adaptive.
I don't think phase will be a concern, but that depends what you want to do with it.
You might be stuck with a download of Pd-extended to use adaptive without too much fuss..
David.

There is an object in the iemmatrix that implements the fletcher-Munson curve (equal loudness contour)... Take a look!