Supervision: Hervé Lissek

Project type: Master thesis


In Switzerland, more than 1 million people are concerned with traffic noise, which may have many adverse effects on health. As a response to this environmental burden, regulations are reinforced to better monitor traffic noise in urban environments, relying on measurements (microphones/arrays) and propagation models, allowing evaluating the actual annoyance to the community ("evaluation level"). The latest model (sonROAD) requires a classification of the passing-by vehicles in 10 classes (buses, motorcycles, various classes of tourism cars, trucks, etc.) which are almost impossible to implement with a human supervision. Therefore, an automatic classification of such car categories is to be developed, based on the emitted noise (eventually combined with cameras).

The proposed project aims at developing and assessing a machine learning algorithm devoted to identify and classify car events based on the time/frequency characteristics of their pass-by noise. For that the following tasks will be undertaken:

  • development of an acoustic database of different car pass-by noise, including its manual annotation

  • the development of a signal processing strategy to prepare the samples for machine learning (denoising, car activity detection, mel spectrogram processing, etc.)

  • the development/assessment of a machine learning framework for car noise classification

  • the implementation of the machine learning framework on an existing acoustic array developed for traffic noise monitoring (acoustic goniometer allowing real-time noise source localization and tracking)

Profile: Electrical engineering, Micro-engineering, Physics, Mechanics

Prerequisites: Acoustics, Electroacoustics

*Learning outcome: Sound recording, acoustic signals analysis,

Context: Recording (20%), signal analysis (40%), Machine learning (40%)