[bdonlan]

UTAU and Vocaloid do indeed use pitch-shifted samples (UTAU even lets you build your own sample libraries). A more recent product, CeVIO, uses modeling IIRC.