01846nas a2200181 4500008004100000022001400041245011200055210006900167260001600236300001600252490000800268520124200276100002101518700002601539700002501565700002301590856005101613 2018 eng d a0001-496600aAn Empirical Mode Decomposition-based detection and classification approach for marine mammal vocal signals0 aEmpirical Mode Decompositionbased detection and classification a cJan-12-2018 a3181 - 31900 v1443 a
Detecting marine mammal vocalizations in underwater acoustic environments and classifying them to species level is typically an arduous manual analysis task for skilled bioacousticians. In recent years, machine learning and other automated algorithms have been explored for quickly detecting and classifying all sound sources in an ambient acoustic environment, but many of these still require a large training dataset compiled through time-intensive manual pre-processing. Here, an application of the signal decomposition technique Empirical Mode Decomposition (EMD) is pre- sented, which does not require a priori knowledge and quickly detects all sound sources in a given recording. The EMD detection process extracts the possible signals in a dataset for minimal quality control post-processing before moving onto the second phase: the EMD classification process. The EMD classification process uniquely identifies and labels most sound sources in a given environ- ment. Thirty-five recordings containing different marine mammal species and mooring hardware noises were tested with the new EMD detection and classification processes. Ultimately, these pro- cesses can be applied to acoustic index development and refinement.
1 aSeger, Kerri, D.1 aAl-Badrawi, Mahdi, H.1 aKirsch, Nicholas, J.1 aLyons, Anthony, P. uhttp://asa.scitation.org/doi/10.1121/1.5067389