State-of-the-art room geometry inference algorithms estimate the shape of a room by analyzing peaks in room impulse responses. These algorithms typically require the position of the source wrt the receiver array; this position is often estimated with sound source localization, which is susceptible to high errors under common sampling frequencies. This paper proposes a new approach, namely using an array with a known geometry and consisting of both sources and receivers. When these transducers constitute a uniform linear array, new challenges and opportunities arise for performing room geometry inference. We propose solutions designed to address these challenges, but also designed to leverage the opportunities for better results.
Use and reproduction:
All rights reserved