The underlying thesis behind AsTeR (Audio System For Technical Readings) and later Emacspeak is that information is display-independent. This leads to the insight that producing auditory renderings of information starting from the true source of that information often produces better renderings than those that result from working from a modality-specific representation; thus, attempting to speak visually rendered information can often prove sub-optimal. AsTeR applied these ideas to documents authored in LaTeX; Emacspeak generalized these ideas to user interfaces.
Emacspeak was therefore designed from the ground-up to enable applications generate their own spoken feedback, rather than having an external software program construct the spoken feedback by responding to events in its environment.
In Emacspeak, theory meets practice to deliver a working implementation; Emacspeak leverages the power of Emacs and its embedded Lisp interpreter to inject spoken feedback into applications that run within Emacs. For a detailed overview on how the advice mechanism in Emacs is used, see the original Assets96 paper, as well as the chapter on Emacspeak in the OReilly publication entitled Beautiful Code.