Pre-thesis F/M: New Algorithms for Automated Room Acoustic Diagnosis

1000scholars; Strasbourg; On-site; EUR 10,000 - 15,000; Full time

Job summary

The research internship takes place in Strasbourg, France, at Cerema’s offices. It lasts one month and involves inverse acoustic parameter estimation through measurements, modelling, and data-driven approaches.

Supervisors include Antoine Deleforge (Inria) and Cédric Foy (Cerema), with contact emails provided. The project blends acoustics, signal processing, and machine learning, and offers hands-on Python programming opportunities.

Qualifications

• Master's level in computer science, signal processing, acoustics or applied math.
• Strong background in Python; MATLAB is a plus.
• Interest in acoustics and inverse problems.

Responsibilities

• Consolidate previous results on room acoustics.
• Write a scientific article for publication.
• Python programming tasks related to data and models.
• Bibliographical studies and literature review.

Skills

Optimization methods

Signal processing

Machine learning

Deep Learning

Education

Master's level in relevant field

Tools

Python

Matlab

Job description

Context

Location and duration of the contract

The research work will take place in Strasbourg (France) at Cerema's offices (11 rue Jean Mentelin, 67200 Strasbourg) and are expected to last 1 month.

Supervisors

Antoine Deleforge, Inria Research Fellow, MACARON Team (IRMA, 7 Rue René Descartes, 67000 Strasbourg)

[e-mail] (3 days/week at Cerema).

Cédric Foy, UMRAE Research Officer (Cerema, Univ. Gustave Eiffel, 11 rue Jean Mentelin, 67200 Strasbourg).

[e-mail]

Assignment

General description of the subject

Noise pollution is the main source of annoyance for the general public and is a major health and social issue. In buildings, the nuisance is often due to the poor acoustic quality of rooms caused by excessive reverberation (canteens, swimming pools, etc.). When it comes to acoustic refurbishment of rooms, proposing a solution requires a good knowledge of the geometric and acoustic characteristics of the existing building. To estimate certain unknown parameters (e.g. absorption of an old ceiling), acoustic engineers rely on in situ measurements of the sound field and on numerical (or analytical) acoustic models for which they calibrate the parameters iteratively in order to recover the measured sound field value. The entire diagnostic process is therefore long, costly and sometimes imprecise, depending on the models used. Given this situation, the development of so-called inverse methods capable of automatically defining the acoustic parameters of interest from the measurement would be a major advance for building acoustics, leading the path to the development of simpler, faster and more reliable tools for acousticians. This is a difficult non-linear inverse problem. We are planning an innovative, cross-disciplinary approach, with a team of researchers, each with their own expertise in acoustics (Cédric Foy, UMRAE, internship supervisor), signal processing and machine learning (Antoine Deleforge, Inria, internship supervisor; Sylvain Faisan, ICube) and mathematics (Yannick Privat, IRMA). This approach overcomes the existing barrier between the worlds of audio and acoustics. It involves solving the inverse problem using existing measurements and direct acoustic theoretical models and then, depending on the acoustic parameters to be determined, developing new algorithms that combine techniques from signal processing, optimisation and automated deep learning. In summary, we are attempting to answer the following question: « Is it possible, using punctual measurements of the sound field in a room, combined with partial and approximate knowledge of the room, the sources and the microphones used, to accurately estimate the acoustic and geometric parameters of the room? »

Bibliography

• [1] C. Foy, A. Deleforge, D. Di Carlo, Mean Absorption Estimation from Room Impulse Responses using Virtually-Supervised Learning, Journal of the Acoustical Society of America (JASA), 2021.
• [2] S. Dilungana, S. Faisan, A. Deleforge, C. Foy, Learning based estimation of individual absorption profiles from a single room impulse response with known positions source, sensor and surfaces, Internoise 2021.
• [3] S. Dilungana, A. Deleforge, C. Foy, S. Faisan, Estimation jointe des profils d'absorption des parois d'une salle à partir de réponses impulsionnelles, 16ème Congrès Français d'Acoustique, Marseille, France, 11-15 Apr. 2022.
• [4] S. Dilungana, A. Deleforge, C. Foy, S. Faisan, Geometry-Informed estimation of surface absorption profiles from impulses responses, Eusipco, 30th European Signal Processing Conference, Belgrade, Serbia, 29 Aug. - 2 Sep. 2022.
• [6] T. Sprunck, K. Chahdi, C. Foy, E. Franck, A. Deleforge, Reconstruction de la forme d'une pièce par super-résolution à l'aide de réponses impulsionnelles, 16ème Congrès Français d'Acoustique, Marseille, France, 11-15 ar. 2022.
• [7] T. Sprunck, Y. Privat, C. Foy, A. Deleforge, Room Shape Reconstruction Using Acoustic Super-Resolution, 24th International Congress on Acoustics (ICA), Gyeongju, Korea, 24-28 Oct, 2022.
• [8] T. Sprunck, Y. Privat, C. Foy, A. Deleforge, Gridless 3D Recovery if Images Sources from Room Impulse Responses, IEEE Signal Processing Letters, 2022.

Main activities

• Consolidating previous results obtrained on this topic through a master internship
• Writing of a scientific article to submit in a journal based on these findings
• Python programming
• Bibliographical studies

Skills

• Optimisation methods, Signal processing, Machine learning, Deep Learning
• Excellent level of
• Python
• programming (Matlab would also be a plus)
• Knowledge, experience or particular interest in audio or acoustics are highly desirable. Knowledge of partial differential equations (PDE) is a plus.
• Master's level (computer science, signal processing, acoustics, applied math...)