That's a fair point. Not at all like training on Wikipedia in which nearly every sentence has novelty to it.

Then again it would give you data on every accent in the country, so the holy grail for modelling human speech.