Multiparty voice communication, where multiple people can communicate in a group, is an important component of networked virtual environments (NVEs), especially in many types of online games. In this paper, we present a new peer-to-peer protocol that uses Gabriel graphs, a subgraph of Delaunay triangulations, to provide scalable multiparty voice communication. In addition, our protocol uses positional information so that voice data can be accurately modeled to listeners to increase the immersiveness of their experience. Our simulations show that the algorithms scale well even in densely populated areas, while prioritizing the sending of voice packets to the closest listeners of a speaker first, thus behaving as users expect.