dB@Tidmarsh

Aims

Ecoacoustics has not yet been involved in the validation of habitat restoration programs, that is conservation actions that aim at recovering the initial state of a natural environment after a strong perturbation. Here, we will develop ecoacoustic methods to track the dynamics of animal populations and communities recolonizing a natural environment in restoration. A platform developed by MIT (Cambridge, Massachusetts, USA) offers a unique opportunity to monitor biodiversity restoration of a freshwater wetland. An important dataset has already been collected by a large network of acoustic and environmental sensors opening a challenge for AI. The first objective of the PhD consists in adapting AI supervised methods to automatically detect and identify vocalizations. The second objective is based on unsupervised methods to detect novelty or anomaly events (e.g. arrival of new (unknown) species) directly from the raw audio dataset. The last and third objective aims at describing, discovering and predicting (unexpected) spatio-temporal patterns and spatial species distribution. The expected results should help to inventory, monitor and preserve natural environments using non-invasive and expert free methods, providing AI techniques to citizens and managers.

Participants and collaborations

Félix Michaud, MIT Media Lab, Living Observatory

Funding

Sorbonne Center for Artificial Intelligence Sorbonne University

Localization

Sound sample

MIT Media Lab

Pictures