The point is that context usually suffices to tell which word is which, same as with wait/weight, time/thyme, and so on. Speakers with the pin-pen merger do pronounce these words the same, and I've known someone who was surprised to learn that I didn't—more than a year into our friendship. Yet he'd heard me talk about our friend Jen and gin, possibly even in the same sentence. He just hadn't noticed that I said them differently.

That's just the way with mergers, though: we don't attend to differences that are unimportant in our personal grammars except through conscious striving. I'm the same way with marry/merry/Mary: I know there are people who say these differently, but even if I'm talking to someone who does, those words are all the same to me. The other speaker may be giving me one of three distinct tokens by their interpretation, but whichever one it is I interpret it as one of a set of identical triplets and rely on context to tell which meaning was meant.

Or to give another example: many speakers where I grew up (Pittsburgh) reduce the diphthongs in "tire" and "tower" so that both of these words sound like "tar." In other words, three words share the same token. But I've never misinterpreted a strong-accented Yinzer's "tar" as "tire," "tower" as "tar," etc. That doesn't mean a Texan would be wrong to say that some of us pronounce these words all the same: we do.

I don't think he's saying he can differentiate how they say it, rather that given the context of the situation he's never mistakenly given someone a pin when they were asking for a pen, and vice versa.

The people that study this kind of thing spend a lot of time listening to and transcribing human speech. Plus, you can see the vowel formants in a spectrogram of the audio.