Room Geometry Reconstruction Based on Speech and Acoustic Image Methodology

Abstract

Room geometry reconstruction is a process that aims to establish an understanding of the acoustic environment. We investigate the reconstruction problem of estimating the room size, such as the length, width, and height of the room surfaces, from signals obtained using a star format array. Specifically, confirming the room size is achieved by estimation of the time of flights (TOFs) of the first-order echoes by analysis of room impulse responses (RIRs). The Euclidean Distance Matrices (EDM) -based approach is an effective method for recognizing correct echo combinations. However, in practical cases, the EDM-based approach requires testing all of the echo combinations to confirm the TOFs of echoes corresponding to an image source, or equivalently to a room surface, then a time-consuming problem arises. To solve this problem, we propose an algorithm called threshold-based EDM which has high-resolution performance comparable to the EDM approach to narrow down the searching range of the TOFs of first-order echoes in the RIRs, thus accelerate the reconstruction. The proposed method is evaluated using simulated room under a variety of different controlled conditions, an example with an error less than 2cm when the signal-to-noise ratio (SNR) is 10dB, and the RT60 is 247ms is given to prove the practicality of the proposed method.