NICOP - DIARIZATION APPLIED TO DETECTION AND CLASSIFICATION OF LARGE WHALE SPECIES, REGIONAL DIALECTS AND INDIVIDUALS
Abstract
AbstractPassive Acoustic Monitoring (PAM) is now a widely spread method for studying the distribution, movements, and behavior of"" cetacean species worldwide. PAM delivers large datasets, which subsequently require powerful analytical tools to process and analyz"e often tens of thousands of hours of acoustic data containing signals of interest but also ambient noise. Developing such tools and developing new methods for estimating animal density or absolute abundance from PAM data are one of the biggest challenges in marin"e bioacoustics at the moment.Whales not only present species-specific acoustic signals, but within a species, several acoustic pop"ulations can be determined. The patterned sequence of calls known as a ~song~ is made up of individual sounds or ~units~ that occur" in repeated ~phrases~ and make up a song bout, where they are repeated over a scale of hours. Blue whale regional song types (or di""alects) are generally distinguished based on differences in song phrasing, inter-unit time interval, inter-phrase time interval, phr""ase duration, and song unit characteristics. At present, a number of methods exist to automatically detect species of whales and aco""ustic groups within species, like spectrogram cross-correlation and pitch tracking, however there is no way of determining how many" individual whales are singing in a given recording. The latter is a key step towards being able to determine animal density or abso"lute abundance. Moreover, very little focus has been given to variation in song characteristics between blue whales that produce the"" same regional song type. Only recently has speaker diarization, a widely used technique in the field of speech processing, been app"lied for the very first time to biodiversity monitoring to determine bird species richness with some success. Diarization corresponds to the process of annotating an input audio channel with information that attributes temporal regions of signal energy to their sp"ecific sources. To the best of our knowledge, speaker diarization has not been applied to acoustic data from the marine environment.""The focus of this proposal is to develop a method to detect and classify different acoustic signals produced by large whales, in o""rder to discriminate between species, acoustic groups/regional dialects, and individuals, using the principles of speaker diarizatio""n. The acoustic data set that will be used in this study is from the blue whale feeding ground in Chilean Patagonia, collected by th""e COPAS Sur-Austral program at the University of Concepcion as of January 2016 in the southern part of the Corcovado Gulf, and curre"ntly available. An alternative acoustic data set will also be available for this analysis obtained by Marine Autonomous Recording Units. The research objectives are: 1) determine large whale species diversity using the principles of speaker diarization; 2) determi"ne blue whale song type diversity (e.g. Antarctic vs. southeast Pacific) using the principles speaker diarization; and, 3) determine" individual variation in temporal and frequency characteristics of song units and inter-unit intervals from single blue whale song b"outs using the principles of speaker identification modelling. Objectives 2 and 3 focus only on blue whales as a model species, howe""ver we expect to apply this resulting method to other large whale species in the future. To achieve objectives 1 and 2, statistical" models of the vocalizations for each species and each blue whale dialect present in the data will be combined with diarization tech"niques to determine the number species according to their dialects in a given set of recorded whale songs. For objective 3, intra-in""dividual statistical models for single-blue whale song bouts will be constructed in combination with diarization principles, and the"n applied to acoustic data in order to determine the number of individuals in a given time period. Duration modelling will be applie"d in order t
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 05, 2017
- Source ID
- N629091712013
Entities
People
- Néstor Becerra Yoma
Organizations
- Office of Naval Research
- United States Navy