Automatic segmentation of speech ir on imporrontpmblem that is useful in speed recognition, synthesis and coding. We explore in this paper: the mbust parameter set, weightingfunction and distance measure for reliable segmentation of noisy speech. It is found that the MFCCparometers, successful in speech recognition. hal& the best promise far mbusr segmentation also. We also explored a variery of symmetric and asymmetric weighting lifrer, fmm which it is found that a symmetric lifrer of the formis mmt effective. With regard to distance measure. the direct La norm irjound odequote.