PhD position in Automatic measurement and prediction of the distance to a sound source using microphone arrays for bird species detection

Supervisors: Théo Mariotte (LIUM), Manuel Melon (LAUM) Marie Tahon (LIUM, dir)
hosting teams: Laboratoire d’Acoustique de l’Université du Mans (LAUM), Laboratoire d’Informatique de l’Université du Mans (LIUM)
Site: Le Mans
Starting date : October 2026
Contact: Marie Tahon and Theo Mariotte (name.surname@univ-lemans.fr)
How to apply: You can submit your application (CV, motivation letter, ) on the dedicated platform:
https://amethis.doctorat.org/amethis-client/prd/consulter/offre/3083

 

Main topic

Biodiversity monitoring is a key challenge for the preservation of natural habitats and for low-impact urban planning. This monitoring often requires conducting an inventory of the species present. Methods for automatically detecting species from audio recordings (Passive Acoustic Monitoring) are being developed but face challenges in field deployment. This limitation stems from several factors, including (i) the models’ lack of robustness regarding the distance between the animal and the microphone, and (ii) the absence of a confidence score for the model’s detection. This thesis aims to address these two limitations by (1) recording acoustic data under real-world conditions with distance information, (2) using this data to estimate an animal’s distance, and (3) incorporating distance into a confidence score for automatic detection models, within a context of computational efficiency.

 

Candidate profile

MsC. in acoustics with strong interests in machine learning and deep learning or MsC. in machine learning with interests in acoustics and audio signal processing

 

Context and state of the art

Biodiversity monitoring addresses major challenges such as documenting biodiversity loss, implementing conservation measures, conducting pre-construction studies, and providing information to nature enthusiasts. This monitoring also aligns with the Net Zero Land Take (NZLT) objective2, which aims to reduce land consumption across the region. Accurate monitoring of natural habitats allows for the selection of the least sensitive areas in the event of land development. It also enables the documentation of the natural restoration of an area. Various local and national stakeholders (MNHN, CPIE) are responsible for this documentation and conservation mission.

Although necessary, on-site biodiversity monitoring operations require the involvement of a large number of field agents. These missions are one-off, primarily conducted during the day, preventing regular and precise monitoring of species presence within a study area. To overcome this limitation, naturalists are increasingly deploying automated monitoring methods that can take various forms [2]. An highly attractive option involves placing networks of microphones at study sites and then analyzing the recorded audio signal to identify present species or extract ecological indicators [3]. This approach, known as Passive Acoustic Monitoring (PAM), offers numerous advantages. Microphones can capture multiple sounds from all directions within a wide radius (up to 100 m). In recent years, numerous methods utilizing artificial neural networks have emerged to automatically detect active species in an audio signal [5], offering a promising outlook for the automation of monitoring and the estimation of ecological indicators. However, these detection methods have several limitations, such as their sensitivity to the recording environment and the distance between the source and the microphone [6]. These models also lack a confidence indicator for their detections, making their use challenging in an ecological research context.

 

Objectives of the thesis

The main objective is to calibrate automatic species detection models so that they provide a confidence score for their decisions. This confidence score will depend on the distance from the detected source. It is therefore necessary to be able to estimate this parameter. The first phase of the project is dedicated to establishing a measurement protocol using microphone arrays, followed by the acquisition of a dataset for measuring the distance to a bioacoustic source in a natural environment The second phase will consist of developing and evaluating acoustic methods for estimating the distance to a bioacoustic source. The third phase will involve using artificial neural networks to predict the distance. This estimator can be used to define a confidence score for detection models.

Research questions

• RQ1 : How can species detection models be calibrated to return a confidence score for their detection results based on distance?
• RQ2 : Is it possible to accurately estimate the source-to-microphone distance using frugal models with a single microphone?
• RQ3 : What are the limitations of source localization methods in real-world bioacoustic research conditions?

 

Resources in the laboratories

The doctoral student will be enrolled in MaSTIC Doctoral School and will benefit from the expertise of both laboratories to successfully achieve the proposed objectives. The doctoral student will have access to an office on the LIUM premises, as well as the equipment necessary for their work (computer, monitor). The laboratories are committed to providing the necessary resources from their own funds (financial and material) to enable the student to publish their work in major conferences and journals in the field.

LAUM has research engineers capable of implementing complex acoustic acquisition systems that will be used for the measurement campaign. This expertise will also enable the development of versions of the acquisition system that can be deployed in the field by incorporating digital microphones (MEMS) controlled by programmable microcontrollers. The LAUM anechoic chamber will also be used during Phase 1 to validate the data acquisition system in a controlled environment.

LIUM has a computing server for training AI models. A technical team is available to assist students in getting started with this service. Students will also have access to national (Jean Zay) and regional (GLiCID) computing infrastructure for training and evaluating machine learning models. LIUM collaborates with the National Museum of Natural History in Paris (MNHN) as part of an ANR project. This collaboration will allow students to visit the MNHN’s CESCO laboratory and interact with ecologists.

Application: You can submit your application (CV, motivation letter, ) on the dedicated platform: https://amethis.doctorat.org/amethis-client/prd/consulter/offre/3083

Bibliography

[1]. S. H. M. Butchart et al., « Global Biodiversity: Indicators of Recent Declines », Science, vol. 328, no 5982, p. 1164‑1168, mai 2010, doi: 10.1126/science.1187512.
[2]. D. Tuia et al., « Perspectives in machine learning for wildlife conservation », Nat Commun, vol. 13, no 1, p. 792, févr. 2022, doi: 10.1038/s41467-022-27980-y.
[3]. S. R. P. ‐J. Ross et al., « Passive acoustic monitoring provides a fresh perspective on fundamental ecological questions », Functional Ecology, vol. 37, no 4, p. 959‑975, avr. 2023, doi: 10.1111/1365-2435.14275.
[4]. T. A. Rhinehart, L. M. Chronister, T. Devlin, et J. Kitzes, « Acoustic localization of terrestrial wildlife: Current practices and future opportunities », Ecology and Evolution, vol. 10, no 13, p. 6794‑6818, juill. 2020, doi: 10.1002/ece3.6216.
[5]. R. Schwinger et al., « Foundation Models for Bioacoustics — a Comparative Review », 2 août 2025, arXiv: arXiv:2508.01277. doi: 10.48550/arXiv.2508.01277
[6]. P. Somervuo, P. Lauha, et T. Lokki, « Effects of landscape and distance in automatic audio based bird species identification », The Journal of the Acoustical Society of America, vol. 154, no 1, p. 245‑254, juill. 2023, doi: 10.1121/10.0020153.
[7]. D. Stowell, « Computational bioacoustics with deep learning: a review and roadmap », PeerJ, vol. 10, p. e13152, mars 2022, doi:
10.7717/peerj.13152.
[8]. E. Verreycken, R. Simon, B. Quirk-Royal, W. Daems, J. Barber, et J. Steckel, « Bio-acoustic tracking and localization using heterogeneous,
scalable microphone arrays », Commun Biol, vol. 4, no 1, p. 1275, nov. 2021, doi: 10.1038/s42003-021-02746-2.
[9] D. A. Yip et al., « Sound level measurements from audio recordings provide objective distance estimates for distance sampling wildlife populations », Remote Sens Ecol Conserv, vol. 6, no 3, p. 301‑315, sept. 2020, doi: 10.1002/rse2.118.
[10] R. Schwinger, B. McEwen, V. S. Kather, R. Heinrich, L. Rauch, et S. Tomforde, « Uncertainty Calibration of Multi-Label Bird Sound Classifiers », 11 novembre 2025, arXiv: arXiv:2511.08261. doi: 10.48550/arXiv.2511.08261.