Is it considered a self-plagiarism to reuse (published) abstracts for talks?

In papers themselves, there is absolutely no reason why you should ever copy any previously published text (including yours) without explicit acknowledgment, along the lines of "The background material in this section is a nearly verbatim adaptation of Section 3.2 of X". Assuming you make it perfectly clear what you have copied and from where (not hiding this information in a note later in your paper, for example) and you have permission from the copyright holder, then this is ethical, while it's not ethical without these conditions.

Of course talk abstracts are not quite the same. Let's assume we're talking about relatively ephemeral abstracts. I.e., they might appear on the web or in the conference program, but they aren't carefully archived, citable contributions to the research literature. (This distinguishes them from "extended abstracts" in CS conferences, which are actually short research papers, and there may be other intermediate cases.) These sorts of abstracts generally don't list any references within the abstract, and they aren't considered published or treated nearly as formally as published material.

In mathematics, I doubt anyone would get upset about recycling a paper abstract for this kind of talk abstract. Customs vary greatly between fields or sorts of abstracts, so you should seek advice from colleagues in your area, since "someone on the internet said it was OK" is not a compelling argument.

If you are worried about self-plagiarism, you can simply append something like "(adapted from the abstract of paper citation)" at the end of your talk abstract. However, that might stand out in its formality.

It's probably a good idea in any case to rewrite the abstract at least a little, since a talk abstract has different goals from a paper abstract.

In general, do not copy from your previous publications. At the same time, I would say that the severity of "self-plagiarism" depends on what you copy. If it is a description of a tool or site, it is not very problematic (after all there may be only so many ways to describe it). If you copy conclusions and items of creative importance then I would say it is more severe. I personally rewrite everything just because I am fine with that, even site descriptions. At some point I will have to go through and see if I unknowingly copied myself in those trivial sections at some point.

So in short, don't copy, rewrite. Who knows, you may find that you improve your formulations that way.

Uhm... discipline specific tag missing??? In economics, you can often see an acknowledgement like "I would like to thank [the list of 15 specific people, may be a Nobel prize winner or two thrown in], audiences at [10 universities in which this talk was given] and [4 more international conferences] for their helpful comments". So an economist won't even get what it is that you are asking; presenting the same stuff over is their daily business.

Generally, I think it is basically your own self-discipline and commitment to moving research further with new work (as opposed to just selling your name with one or two successful papers that everybody keeps wanting to hear about). I usually present the same research two-three times, which means copying and pasting the abstract, but once I hit all the major audiences with it, I move on to something different.