Separating acoustic signal into underlying behaviors with self-exciting point process models