Not to stray off the topic of the thread [1] but I think you're neglecting some important bits of reality on this.
The basic issue is that there's a bunch of noise outside the car, including some generated by the car, that passes through (for example) the windows, and you have to listen to it, right?
View attachment 673498
... so one thing that Tesla can do is have better sound-insulating materials, such as "acoustic glass", to attenuate those sounds:
View attachment 673511
... that's OK, but not good enough, because this is an expensive car and if you can't hear a mosquito fart while driving 100mph, what's the point? So they're going to cancel the sound. ANC isn't anything new, and it's come a long way in the last couple of decades but basically if you have a sound wave, and add to it an identical but inverted sound wave, the result is very quiet [2]:
View attachment 673514
... if we take the black wave and invert it, then add the inverted wave to the original one, we get zero (let's ignore my sloppy image editing).
That's great, and for that to work, we need to get exactly the set of frequencies that you are hearing - as you say, right next to your eardrum - then invert them and play them back with perfect fidelity at the exact instant that the actual waves hit your eardrum.
That would be cool, but it's also physically impossible, because:
a) The speakers generating the anti-noise are some distance (maybe half a meter? less? more?) from your eardrums, and
b) There's additional delay because the DSP has to process the sound and then poop it back out to the car's amplifiers.
There's additional things at play here; the DSP, being clever, is also going to have some awareness of sounds that the speakers are not able to faithfully reproduce, so it may attempt to ignore those or treat them differently. It may also be on the lookout for specific frequency bands or types of sounds, e.g. emergency sirens, car horns, and people speaking, and not attempt to cancel them. As others have noted, some types of noise are relatively easy to cancel; others are much harder. Let's ignore all of that and just look at our basic oversimplification:
View attachment 673519
If we take the sounds right at your ears, send them back to a computer, have it invert them, then play them back from speakers that are half a meter from your ears, what you'll hear is a muddled mess because the anti-noise that reaches your ear is actually the inverse of something you heard a couple of milliseconds ago. That doesn't seem like a lot, but what you end up with is weirdly white noise with odd gurgling and jingling sounds, and other effects that you don't want.
If the microphone is just
before the speaker, then that delay is minimized (and, being known, can be accounted for) and the speakers are more likely to play something that's out of phase with the reality of
now, rather than the reality of a couple of milliseconds ago, relative to your ears:
View attachment 673520
This all seems very excessive so I'm going to stop now but ... I'm going to speculate that if there are microphones on the seats near your shoulders, they're not doing the primary work of active noise cancellation because they're on the wrong side of the speakers to be useful. They could be used to help the system distinguish sounds that shouldn't be canceled (e.g. people speaking), or be for something else altogether.
[1] who am I kidding?
[2] This is some high school level oversimplification but it works well enough. We're not going to the moon here.