Having read the NYT article (with the PNAS paper still not available through a certain hub), I think a useful analytical framework would perhaps be to think of silence as a negative space. E.g., take some background noise (this could be the environmental noise averaged over some time scale) at certain overall intensity as "zero" (or reference level), then complete silence will have the same frequency content as that background but with negative intensity. From there one can start talking about various forms of "partial silence" as different spectral compositions of negative intensity. I'd even posit that some of the illusions they discovered would work in a similar fashion with positive intensity boost as well (e.g.two disjoint boosts vs one sustained boost). It is probably more about the frequency content than the intensity relative to the reference level.