i would also be interested by an "A-weighting" filter as a max external. No one ?

If not, perhaps, trying to approach the curve - the curve is here, in blue : http://en.wikipedia.org/wiki/A-weighting - using two filtergraph/biquad~.

The way to do this is:

1 - apply A weighting filter to the input

2 - measure the level in any appropriate method (with or without averaging)

For 1 you have a few different options as with any filter - you can either try an approximate using a IR filter (like biquad~ etc.), or you could use a FIR filter which can probably be made more precise, but I think you'll incur some latency as your filter will be centred on half the impulse length, rather than around your input samples- For an FIR filter you can either use a convolution based approach (using buffir~ or similar), or use an FFT (which will also introduce latency). Both the later approaches will be fairly expensive using standard max methods, but probably more accurate, depending on how closely you want to match the A weighting curve.

Regards,

Alex