An efficient unit-selection method for embedded concatenative speech synthesis