As far as I can figure out, on chrome it's just a web service -- and doesn't work on Debian Chromium as they stopped shipping google api keys (which makes sense, part of the point of using chromium over chrome is to have some measure of independence from google).
Just thought I'd mention it here, in case anyone else were about to run off and try stuff like:
Yeah, Chrome implements it as a web service, but Google and Apple have been slowly working on a Web Speech API for a while (I suppose Mozilla and Microsoft see it as an unnecessary abstraction when the Web Audio API could do speech-related stuff).
I think Safari has only implemented the Synthesis part of the API because it can all be done offline and rely on the system voices on OS X and iOS. Google uses their own web APIs and thus also does Speech Recognition, but it doesn't work well offline or when you don't have Google API keys within Chromium/Chrome.
Just thought I'd mention it here, in case anyone else were about to run off and try stuff like:
http://updates.html5rocks.com/2014/01/Web-apps-that-talk---I...
only to be disappointed :-/
Note that on Debian/Ubuntu you can for example do:
if you need to get your computer voice fix.