Remarks on Optimal Scores for Speaker Recognition