Interesting thought: is it easier or more difficult to make a synthetic voice undistinguishable from a human one, compared to producing speech copying a real voice?
To me, at least, the voices I heard in Lyrebird's demo actually sounded more 'real' than Microsoft Sam for example.