The entire concept of unit selection rests on the assumption that speech can be broken down into a sequence of non-overlapping units, like the beads on a necklace. That assumption is too strong, but we can mitigate the problems of this assumption by using context-dependent units.
7 minutes 1 second