In this paper, we derive the threshold of the maximum likelihood (ML) decision rule, assuming the reliability In this paper, we propose a maximum likelihood detector measure can be modeled with Rayleigh probability density for reliable sound source localization system. It is based on fnction (pdf) and show that the performance of the making a measure of reliability of estimation results. The reliability measure in a fixed talker environment and a reliability can be reduced from waterbed effect of source moving talker environment. localization algorithm. If the calculated reliability measure This paper organized as follows. In Section 2, we introduce has a lower value than a predefined threshold, the estimated the GCC-PHAT DOA estimator. In Section 3, we discuss direction-of-arrival (DOA) is regarded as a wrong result and the definition of the reliability measure. In Section 4, we subsequently discarded. We determine the threshold for derive the threshold for reliable source detection with ML reliable estimate selection using maximum likelihood rule. decision rule. In Section 5, we discuss the experimental Some experiments show that the proposed method can reject results. Finally, we present our conclusions in Section 6.