"Computational passive bioacoustics: Will AI really solve animal communication?"
When: January 20st - 11 am - 12 am
Where: Salle des Voûtes
Paul Best (Eq. Multisense & Body) has been awarded an 2024 ILCB post-doc grant to work on the topic of spatialised bioacoustics for the analysis of turn-taking in non-human interactions. Paul’s original background is in computer science and machine learning. During his PhD and post-doc in Toulon university, he worked on automating the analysis of non-human vocalisations using neural networks. Paul has applied these methods to a range of species with a variety of research questions, always working with long-term acoustic data recordings of free-ranging animals. Some of these projects include analysing the presence patterns of sperm whales, characterising the evolution of song structure in fin whales, and linking communicative complexity to sociality in orcas. Currently at CRPN, Paul is working with recordings of pilot whales and cao-vit gibbons, with a focus on how contextualising passive acoustic data with spatial information (the location of the vocalising animal) contributes to a better understanding of their vocal behaviour.
Abstract:
For this seminar, we will look at current research in passive bioacoustics, especially regarding opportunities and limitations of using machine learning to study non-human communication. With the increasing use of long-term monitoring systems such as autonomous acoustic recorders, the relatively young and growing field of computational passive bioacoustic is promising in collecting naturalistic data to contribute to ethological and evolutionary questions. Drawing on my research experience and project in this domain, I will present what is and could be possible with such approach, including with the use self-supervised learning, case studies of cetacean communication systems, the benefits of sound localisation and the relevance of integrating a musical perspective in the study of non-human communication.