Loudspeaker localization in a reverberant room is essential for ensuring high-quality immersive rendering with spatially distributed loudspeakers. Traditional approaches rely on beamforming or deep-learning-based feature extraction for a single source localization at a time. This paper applies simultaneous deconvolution (SD) towards multi-source localization where loudspeaker distances are first estimated from impulse responses deconvolved {\em simultaneously}, and the distances are then used towards estimating loudspeaker signal Direction-of-Arrival (DOA).
We compare geometric angle estimators, supervised and unsupervised machine learning (ML) models, and traditional signal processing approaches involving cross-correlation and pseudo-spectrum. The Trapezoid estimator and supervised ML models are the top two techniques for use with SD for loudspeaker localization, factoring both synthetic and listening room test sets.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=22999
(1073KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
