Well, I mean the reason I'm suggesting this is precisely because txt to speech engines are so inadequate. And no matter how much they advance in the next year, they'll still never be able to put things in context like a human reader can (of course, many of the non-native English speakers won't be able to do that particularly well either).
Yes, TTS quality is far from real people reading. I think the service must be something like Amazon's Mechanical Turk but with a SLA since recordings must be quickly generated.
