Ambient Assisted Living (AAL) has become a powerful alternative to improving the life quality of elderly and partially dependent people in their own living environments. In this regard, tele-care and remote surveillance AAL applications have emerged as a hot research topic in this domain. These services aim to infer the patients’ status by means of centralized architectures that collect data from a set of sensors deployed in their living environment. However, when the size of the scenario and number of patients to be monitored increase (e.g., residential areas, retirement homes), these systems typically struggle at processing all associated data and providing a reasonable output in real time. The purpose of this paper is to present a fog-inspired distributed architecture to collect, analyze and identify up to nine acoustic events that represent abnormal behavior or dangerous health conditions in large-scale scenarios. Specifically, the proposed platform collects data from a set of wireless acoustic sensors and runs an automatic two-stage audio event classification process to decide whether or not to trigger an alarm. Conducted experiments over a labeled dataset of 7116 s based on the priorities of the Fundació Ave Maria health experts have obtained an overall accuracy of 94.6%.