Speaker identification with user-selected password phrases