Learning-based estimation of individual absorption profiles from a single room impulse response with known positions of source, sensor and surfaces
Résumé
In situ estimation of the individual absorption profiles of the surfaces in a room remains a challengingproblem in building acoustics. This work is aimed at studying the feasibility of this estimation in ashoebox room of known geometry, using a room impulse response (RIR) measured from a source anda sensor at known positions. This problem is tackled using supervised learning. Two neural networkarchitectures are compared. Simulated training and validation sets of RIRs featuring various types ofdistortions (surface di_usion, geometrical errors and additive white Gaussian noise) are generatedfor the considered geometry. Two extensive empirical simulated studies are carried out to determinethe influence of these distortions on the performances of learned models, and to determine whichcomponent of the RIR is most useful for absorption profiles prediction. Trained models are shown toyield errors significantly smaller than those of a naive mean estimator on every simulated datasets,including those featuring realistic perturbation levels. Our study outlines the benefit of usingconvolutional neural network layers, especially when geometrical errors exist. It also reveals thatfirst order specular reflections are the most salient feature of RIRs for absorption profiles predictionunder a fixed geometry.