Room Acoustical Parameter Estimation From Room Impulse Responses Using Deep Neural Networks

We describe a new method to estimate the geometry of room and reflection coefficients given impulse responses. The utilizes convolutional neural networks multilayer perceptrons coefficients. mean square error is used as loss function. In contrast existing methods, we do not require knowledge relative positions sources receivers in room. can be with only single RIR between one source receiver. For simulated environments, proposed estimation achieve an average 0.04 m accuracy for each dimension 0.09 real-world achieves 0.065 dimension.