Résumé : We consider the problem of estimating the mean absorp-tion coefficients of a room from an impulse response us-ing supervised learning on simulated training sets. Twoneural network architectures and two training dataset de-signs are considered. The proposed approach is shown toyield smaller estimation errors than the classical Sabineand Eyring formulas, despite not relying on any geomet-rical information on the room. Simulated results demon-strate the robustness of the approach under different chal-lenging acoustic conditions
https://hal.inria.fr/hal-03045556
Contributor : Antoine Deleforge <>
Submitted on : Tuesday, December 8, 2020 - 9:38:50 AM Last modification on : Tuesday, January 5, 2021 - 4:17:46 PM
Corto Bastien, Antoine Deleforge, Cédric Foy. Mean Absorption Coefficient Estimation From Impulse Responses: Deep Learning vs. Sabine. Forum Acusticum 2020, Dec 2020, Online, France. ⟨hal-03045556⟩