The Virtual Sound Decoder (or "VSDecoder" or "VSD") is a new module that I am creating to add to the JMRI project. In brief, it's a module that provides sounds in response to throttle inputs, just like an actual hardware throttle does, except the sounds are played through your computer's speaker system instead of through a tiny speaker on the locomotive.
The sound models themselves are stored in a "VSD File", which is conceptually similar to a Digitrax SPJ file in that respect. The user can select one of several different sets of sounds, play sounds for multiple locomotives, and so on.
The VSD project isn't even far enough along to call it "Alpha" code yet, and hasn't been included in a JMRI release, though it's in the SVN repository, if you like.