Acoustic source localization is one of the interesting applications of sensor networks. Localization algorithms that use acoustic signal energy measurements at individual sensor nodes are proposed. The maximum likelihood (ML) algorithm is known as an algorithm for source localization. The ML algorithm has high accuracy to estimate source location, however the calculation amount of the ML algorithm is large. The expectation-maximization (EM) algorithm is proposed to realize the ML algorithm with less amount of calculation. An EM algorithm processed at a center, that is, the centralized EM algorithm is proposed in which each sensor node sends observed information to the center. The total cost of the centralized data processing is expensive, because the communication cost is large when the transmission distance from each sensor node to the center is long. On the other hand, the distributed data processing can reduce the transmission cost and thus the total cost, because data is processed in each sensor node and thus the amount of communication can be reduced. In this paper, we propose a distributed EM algorithm for acoustic source localization that is applicable to a source with unknown signal energy. We show that the proposed algorithm has high accuracy and reduces the communication cost.