Skip to Main content Skip to Navigation

Contributions en méthodes bioacoustiques multiéchelles : spécifiques, populationnelles, individuelles et comportementales

Abstract : The objective of this thesis is to make different methodological contributions in bioacoustics for the study of fauna. Bioacoustics is a recent multidisciplinary science and is very effective for studying and classifying an ecosystem. Many past studies have developed acoustical methods to analyze wildlife across (1) specific, (2) populational, (3) individual and (4) behavioral scales. The research presented in this thesis aims to study different case methods in the four scales of analysis listed above while also setting up tools from the setup of the acquisition material to the analysis of the data for all the aforementioned scales, and finally the discussion of the studies and putting them into perspective. In this study, (1) specific bioacoustics were illustrated by the automatic classification of orcas, sperm whales, and birds. The acoustic classification of orca clans were studied for (2) population analysis. Then the scale was refined and (3) individual acoustic emissions were studied through three different case studies : the individual locations of orcas, sperm whales, and birds. The last scale evaluated was (4) behavioral bioacoustics which aimed to correlate behaviors with acoustic emissions. In order to correlate certain behaviors with acoustic emissions, the influence of maritime traffic on pantropical spotted dolphins and the impact of chemical stimuli in humpbacks were evaluated and recorded. We deliberately chose to select a diverse pool of species that would produce a variety of different signals (stationary vs. transient) and had evolved in different environments (marine vs. terrestrial). This allows us to standardize analysis methods in order to facilitate the development of new studies in bioacoustics. Each case study showed interesting results in terms of bioacoustics and behavioral ecology. These results were compared with past studies which can be found in the bibliography. The results of each case study validated the methods proposed in this thesis. In particular, our study yielded excellent results in the evaluation of bird songs and is now a sound-recognition application available on any type of mobile phone, making it easy to identify bird species. The methodological contributions of this thesis, specifically the difference between stationary and transient signals and those of marine or terrestrial evolution, were synthesized, compared, and discussed. Supervised and unsupervised methods were also compared. These proposed methods have been tested and validated using massive data (several tens of Tera), which are unique. In conclusion, this thesis shows that supervised methods, in particular Deep Learning, are very well suited for the classification of stationary signals in specific and population-based bioacoustics for the terrestrial and marine environment. We also derived that unsupervised methods such as clustering and reduction of dimensionality, can be used within the framework of behavioral bioacoustics to identify signals of interest. Finally, individual bioacoustics can be translated into localization methods such as estimating the inter-sensor delay time which is feasible for transient signals and more complex for stationary signals.
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Thursday, July 15, 2021 - 4:29:10 PM
Last modification on : Sunday, June 26, 2022 - 5:58:53 AM
Long-term archiving on: : Saturday, October 16, 2021 - 6:54:33 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03287502, version 1


Marion Poupard. Contributions en méthodes bioacoustiques multiéchelles : spécifiques, populationnelles, individuelles et comportementales. Acoustique [physics.class-ph]. Université de Toulon, 2020. Français. ⟨NNT : 2020TOUL0015⟩. ⟨tel-03287502⟩



Record views


Files downloads