Intrinsic structure study of whale vocalizations

Whale vocalizations can be modeled as polynomial-phase signals, which are widely used in radar and sonar applications. Such signals lie on a nonlinear manifold parameterized by polynomial phase coefficients. In this paper, we apply manifold learning methods, in particular ISOMAP and Laplacian Eigenmap, to examine the underlying geometric structure of whale vocalizations. We can improve the classification accuracy by using the intrinsic structure of whale vocalizations. Our experiments on the DCLDE conference and MobySound data show that manifold learning methods such as ISOMAP and Laplacian eigenmap outperform linear dimension reduction methods such as Principal Component Analysis (PCA) and Multidimensional Scaling (MDS).

Duke Scholars

Author Douglas Nowacek Marine Science and Conservation

Author Xiaobai Sun Computer Science

Author Loren W. Nolte Electrical and Computer Engineering