Skip to Main content Skip to Navigation

Antennes microphoniques intelligentes : localisation de sources acoustiques par Deep Learning

Abstract : For my PhD thesis, I propose to explore the path of supervised learning, for the task of locating acoustic sources. To do so, I have developed a new deep neural network architecture. But, to optimize the millions of learning variables of this network, a large database of examples is needed. Thus, two complementary approaches are proposed to constitute these examples. The first is to carry out numerical simulations of microphonic recordings. The second one is to place a microphone antenna in the center of a sphere of loudspeakers which allows to spatialize the sounds in 3D, and to record directly on the microphone antenna the signals emitted by this experimental 3D sound wave simulator. The neural network could thus be tested under different conditions, and its performances could be compared to those of conventional algorithms for locating acoustic sources. The results show that this approach allows a generally more precise localization, but also much faster than conventional algorithms in the literature.
Complete list of metadata
Contributor : Abes Star :  Contact
Submitted on : Wednesday, February 24, 2021 - 1:44:37 PM
Last modification on : Friday, May 21, 2021 - 10:40:01 AM
Long-term archiving on: : Tuesday, May 25, 2021 - 6:36:29 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03151039, version 1



Hadrien Pujol. Antennes microphoniques intelligentes : localisation de sources acoustiques par Deep Learning. Génie mécanique [physics.class-ph]. HESAM Université, 2020. Français. ⟨NNT : 2020HESAC025⟩. ⟨tel-03151039⟩



Les métriques sont temporairement indisponibles