It's not pitch shifted at all, just granular synthesis with the grains sorted by pitch and filtered by voiced-ness.