Collective excitations can be either localized or delocalized. For example, in a conductor the low-energy excitations are particles or holes near a Fermi surface, which are delocalized. But in an insulator, the excitations are often localized.
Several commentators have claimed that Bloch's theorem requires all collective excitations to be delocalized in a translationally invariant system. But this is incorrect; Bloch's theorem only applies to noninteracting systems. The Coulomb repulsion can result in an interaction-driven Mott insulator, whose low-energy excitations are usually very localized.
Localized excitations can still be legitimately thought of as "collective," because they're usually not perfectly localized to a single site in the microscopic lattice - they typically have an exponentially-decaying tail in space. E.g. this can arise from screening effects. These localized quasiparticles are not identical to the microscopic particles: they get "dressed" by their interactions with nearby particles - but not with those far away. These interactions can renormalize their mass, Lande g-factor, etc., as in Fermi liquid theory. Thus the presence of nearby microscopic particles is still important, even if the excitations are localized.
Regarding the finite speed of sound: for a lattice system whose interactions decay exponentially (e.g. because of screening), there are Lieb-Robinson bounds that mathematically limit the speed at which information and entanglement can spread. This is true regardless of whether the excitations are localized or delocalized.