Pervasive computing is fast becoming a reality with rapid advance in computing and networking technologies. It has the characteristics of scalability, invisibility, and the absence of the fixed infrastructure. Thus, existing security solutions do not fit this new computing environment. In many pervasive applications, devices collaborate with others within a particular area (such as a room or a building) and thus such applications need authentication for security. In this paper, we propose a new authentication technique which allows devices to authenticate each other when they are within the same audio region. The main advantage of our scheme is that there is no need of the pre-shared secret, the pre-existing trust relationship and human involvement. This paper takes a cyber-physical approach and generates features from the environmental sound. The idea is that this feature is used to verify the legitimacy of devices. We present the audio-based self-organizing authentication scheme and experimental results showing the achievement of requirements. We also analyze the computation cost, energy cost and security.