It's certainly possible in theory. The Microsoft Speech Recognition engine, which VoiceAttack uses, can be accurate enough for preset commands when properly trained(ideally in a quiet room with a good microphone, as with any voice recognition software).
I watched part of the demo section of the presentation video and replicated some short command phrases("slap", "delt", "pycon", "space", "quote", "yank"), which are actually recognized reasonably well(most between 96 to 86 confidence). However, it should be noted that recognition is generally more accurate with longer phrases, and some words work better than others(E.G. "quote" was recognized as "called" multiple times; note that I made no effort to train individual words, which could improve recognition).
Feature-wise, VoiceAttack should be flexible enough to do what was done in the demo(though the impromptu dictation may require a command beforehand), however the use of EMACS is important to note as it offers keyboard shortcuts for just about all its features. You may find that other editors are more difficult/involved to interface with.
It's up to you whether you want to invest time and money into either option, though the latter at least is an order of magnitude smaller for VoiceAttack.