An EMD Based Automatic Endpoint Detection Method for Cetacean Vocal Signals

In the ocean environment, there are many noises in passive acoustic monitoring (PAM) data of cetacean species. The manual detection of vocal signals from a large amount of PAM data is inefficient and time-consuming. Therefore, it is necessary to explore a more efficient automatic detection for processing PAM data. In this paper, an automatic detection via empirical mode decomposition (EMD) based endpoint detection method for cetacean vocal signals is proposed. Firstly, the PAM data is transformed into EMD signals. Then two adaptive thresholds are computed for automatic detection. Finally, the vocal signals are extracted from PAM data as well as the clutters are removed. This method grabs the acoustic features of vocal signals and distinguishes them from the background clutters of the ocean environment. It has been proved the effectiveness of the proposed method by comparing with the Teager-Kaiser energy operator (TKEO) detection method. We process 194 audio files from Marine mammal public databases, about 1.54G data. Five kinds of vocal signals from different cetaceans are extracted through the proposed method.