Text was fed into MS speech synth, once on a whisper and once on a normal voice. The whisper was subjected to various indignities; a copy was hugely time stretched, another was granulated, put through dFx's buffer override and then cut up. The whole lot was then put together and mixed.

What I /wanted/ to do, but couldn't get working and won't have time for before the deadline, is to make more use of the actual letters themselves by either 

a) Assigning a sample to each letter and treating a word as a sonic event made by layering these samples (dunno how timing would be handled). 

b) Using letters to represent parameters of a grain synth, again not sure how timing would have been represented. 

I'll continue to play with it...

Enjoy.