A deep neural network was developed and trained to identify the sounds of pomacentrids (damselfishes) in the National Park of American Samoa. Four years of continuous, passive acoustic data were recorded by a single stationary hydrophone. Deep neural networks enable the full utilization of such large datasets by automating laborious manual sampling tasks. The recordings were analyzed using this novel application of neural networks to detect the vocalizations of damselfishes. These data provide a rare opportunity for long-term monitoring of seasonal and diel vocalization patterns. Vocalizations of damselfishes are presumed to change in occurrence over time, so understanding the temporal variation is necessary to examine the effects of environmental cues and pressures on these fishes and their marine habitats. The vocalizations were found to vary on daily and annual timescales, most likely in relation to wind, water temperature, and other environmental factors. Damselfishes can be used as indicators for coral reef health and as a proxy to assess the populations of commercially important fishes, such as snapper and grouper. The remote nature of the location reduces potential interference from anthropogenic sound sources, such as cargo shipping, revealing meaningful interactions between the fish and their environment. The field recordings were quantified using manual and deep neural network techniques and assessed for indications of significant change. This research has broad implications for state-of-the-art acoustic analysis, which promises to be an efficient, scalable asset in ecological research and conservation management plans.