Apple already has a database of voice samples. Siri works by transmitting voice requests back to Apple's data centre where they are interpreted into text. Voice samples in the database, if it exists, may be indexed by identity, location, background noise, topic, and correlated activity. Not sure how feasible it might be to synthesise ad hoc utterances in the voice of a known person using that database, but it could certainly be used as a voice print comparison to identify individuals by sound.

ETA: speculation