Comparison of Series and Parallel LC Filters for Noise Reduction

In general, designing passive filters to work at the low, highly frequency dependent (i.e., reactive) and poorly-controlled impedances associated with things like headphones is not the best approach.

It would be better to attenuate the headphone signal as needed to run it into an active (opamp-based) filter with the desired characteristics, and then boost it back up to headphone levels at the output. The results will be much more stable and easier to fine-tune for specific applications.

In fact, you can purchase such units — both parametric and graphic headphone equalizers — off-the-shelf.

I believe I've answered this elsewhere but, regardless, here is the way to look at this. The transfer function for the RC HPF is:

\$\dfrac{j \omega RC}{1 + j \omega RC} \$

The "trick" is to look at the behaviour at low enough frequency such that \$ j \omega RC << 1\$. When this holds, the denominator is effectively just \$1\$ and the transfer function is effectively:

\$ j \omega RC\$

But, this is the transfer function for a differentiator with gain equal to \$RC\$. That's really all there is to it. For frequencies well below the corner frequency, the output is effectively proportional to frequency just as we would expect from a differentiator.

You're probably misunderstanding "stop band" in this context. A 1st order high pass filter has a gentle roll-off that is just about 20dB / decade. Signals aren't "stopped" below the corner frequency, they're increasingly attenuated.