Missions

The post-doc recruited will participate in the development of new representations for the analysis of non-stationary signals, in particular bioacoustic signals. The main objective is to design, implement and evaluate a constrained neural architecture capable of producing interpretable and compact representations, based on the broken stationarity model. This work is part of a wider effort to gain a fine-grained understanding of complex acoustic signals for analysis or classification purposes.

Activités

- Development of a representation model based on deep learning, constrained by physical and mathematical considerations (spectrum of a stationary signal + deformation function).

- Implementation of the model in the form of a non-linear neural network, with a decoder reproducing the synthesis operation.

- Quantitative evaluation of the model on real databases, in particular the Watkins Marine Mammal Sound Database.

- Performance comparison with conventional time-frequency analysis methods (wavelets, Fourier).

- Contribution to scientific publications and participation in international conferences.

- Interaction with researchers in machine learning and bioacoustics to foster interdisciplinary synergies.

Compétences

- Solid background in signal processing and/or machine learning.
- Good level of scientific programming (Python preferred).
- Knowledge of or interest in time-frequency/time-scale representations.
- Ability to work independently in a collaborative environment.
- Taste for numerical experimentation, reproducibility and critical analysis of results.

Contexte de travail

The successful candidate will join the Signals, Systems and Physics (Sisyph) team of the Laboratoire de Physique at ENS Lyon. This team produces high-level research for methodological development in the field of information processing (from signal and image processing to machine learning).