It's mostly talking about the case where someone receives an encrypted message which is intended to later be published openly. If it was padded by adding stuff, an attacker can try to reconstruct the original plaintext by removing the flowery adjectives, whereas if things were deleted the attacker doesn't know what to add.
In particular, the length of a message is not encrypted when encrypting the text. So if the encrypted message is shorter, you know exactly how much to remove to get back the original, and then just need to guess what to delete. If the message is longer, it is much harder to guess whether to add flowery adjectives, a new sentence, change a pronoun for a name, or some other change.
It's not just about the actual sound effects, it's also the writing. The movie clip the article uses as an anchor shows people using silencers to have a gunfight in public with bystanders having no idea it was happening. That's not just hollywood sound direction, it's hollywood giving a completely wrong impression about how good silencers are and why they exist
NP-complete is indeed the intersection of NP-hard and NP.
If you can solve any NP-hard problem then you can solve any NP-complete problem (because you can convert any instance of an NP-complete problem to an NP-hard problem), so NP-hard problems are "at least as hard" as NP-complete problems.
(But an NP-hard problem might not be in NP, ie given a purported solution to an NP-hard problem you might not be able to verify it in polynomial time)
Someone who knew or guessed that you were doing it could find your key very quickly by trying out every beacon value in the time range your key was generated in
Whether family names are more differentiated depends on where you live.
The USA has a wide variety, but there are also places like Vietnam where only a handful of family names are in common use and more than 30% of people are Nguyens.