Signal Processing

TitleSignal Processing
Publication TypeBook Chapter
Year of Publication2016
AuthorsMellinger, David K., Roch Marie A., Nosal Eva-Marie, and Klinck Holger
EditorL. Au, Whitlow W., and O. Lammers Marc
Book TitleListening in the Ocean
PublisherSpringer New York
CityNew York, NY
ISBN Number978-1-4939-3176-7
EndNote Rec Number11030

We examine some methods commonly used for analyzing marine bioacoustic recordings. Filtering techniques are used to prevent aliasing, to remove certain types of noise, to flatten the spectrum of ocean noise before recording, and so on. Filter design necessarily requires making choices that affect trade-offs among various desirable filter properties. Detection and classification are used for analyzing large data sets. They often start with signal conditioning, which can adjust the spectrum, standardize signal level, and remove some types of noise. They proceed by calculating numerical acoustic features and using them to decide whether a given sound is present (detection) or to choose which of several categories a vocalization belongs to (classification). A variety of methods for detection and classification are briefly described, with the choice depending both on the nature of the sound(s) and the noise as well as on the task to be solved. Detectors operate in the time domain or on a time–frequency representation, with different ones appropriate for different call types. Classifiers are characterized as either generative or discriminative, as parametric or nonparametric, and as supervised or non-supervised. Performance of detection and classification can be evaluated in several ways, including receiver operating characteristic curves and precision/recall statistics. Localization of calling animals is usually performed using time differences of arrival of sounds at several hydrophones; a variety of methods are available, with the best choice depending on the characteristics of the sound and the acoustic environment. The most accurate localization methods use acoustic propagation modeling to estimate travel times. Several software packages are reviewed for filtering, detection, classification, and localization.