Mountain Lion includes a very basic continuous speech to text conversion tool. It relies on Apple's servers, and is widely assumed to use the Siri infrastructure. It is speaker-independent; the "calibrate" system preference option is to support command recognition, not speech engine customization.
It's not nearly as useful as the speech recognition functionality built into iOS or Android, but it has one great advantage for the use case I'm going to describe. Like Windows 7 speech recognition, it's built into a multi-user operating system and its free.
That means I could do this:
I set up a user account called 'French' on my MacBook and configured it for French use, adjusting even the language location and keyboard preferences. Then I activated speech recognition so that a function key double tap invokes the recognition microphone.
I'm using the simple Notes app to practice. The theory is that if the machine can recognize my mangled French, then so will mere humans.
There is very likely a way to do something similar with Android or iOS speech recognition, the advantage here is that I could set up a special user rather than changing system preferences. I suspect it would be relatively easy to create an Android app that would switch preferences for educational purposes (iOS is more closed).
(I have a long way to go. "Je suis un homme" became "je serai enorme".)