In this paper a scheme for peer-to-peer (p2p) multiparty voice communication is proposed. The paper considers realistic constraints on the upstream and downstream bandwidth of peers and their processing capacity. We propose a novel mechanism for distributed mixing of the audio streams that can improve the scalability of this service while meeting the capacity constraints. The paper also proposes an algorithm to form a multipoint-to-multipoint spanning tree (MMST) among the peers within a communication zone. The simulation results show that the performance of algorithm in terms of delay stretch and its impact on bandwidth reduction in comparison to multiple source rooted overlay multicast trees is quite acceptable.