Tracking marine species is critical for conversation efforts, therefore automatic detectors and classifiers play a significant role to facilitate such efforts. Previous work demonstrated the efficacy of the Empirical Mode Decomposition (EMD) classification capabilities in differentiating marine mammal vocalizations. However, EMD failed to separate Risso’s and Pacific white-sided dolphin pulsed signals due to their high spectral similarity and limitations of EMD filtering response. The Variational Mode Decomposition (VMD), a more advanced version of EMD, paired with an automated detector, provided a promising tool to tell such dolphins apart with accuracy up to 81.3% even under severe channel conditions. This level of accuracy on a sparse dataset that does not contain whistles is important for automatically classifying pulsed signals from dolphins that do not whistle, live in noisy environments, and/or are recorded in datasets with low duty cycles. Because many datasets collected to date in the Arctic Ocean are sparse due to conserving battery power over long deployments, the VMD method can help add to our ability to track dolphins using just their pulsed signals as they expand northward with warming waters.
