Vocalizations of cetaceans form a key component of their social interactions. Such vocalization activity is driven by the behavioral states of the whales, which are not directly observable, so that latent-state models are natural candidates for modeling empirical data on vocalizations. In this paper, hidden Markov models are used to analyze calling activity of long-finned pilot whales (Globicephala melas) recorded over three years in the Vestfjord basin off Lofoten, Norway. Baseline models are used to motivate the use of three states, while more complex models are fit to study the influence of covariates on the state-switching dynamics. The analysis demonstrates the potential usefulness of hidden Markov models to concisely yet accurately describe the stochastic patterns found in animal communication data, thereby providing a framework for drawing meaningful biological inference.