Speaker
Description
In virtual and augmented reality (VR/AR), spatial audio rendering often relies on limited-order spherical-harmonic (SH) representations. While SH truncation in head-related transfer functions (HRTFs) is known to introduce high-frequency phase errors that degrade perceptual quality, analogous issues for loudspeaker directivity have received less attention. This work investigates preprocessing strategies for transforming measured loudspeaker directivity into SH form in the presence of direction-dependent time-of-arrival (TOA) inconsistencies, particularly pronounced in multi-way loudspeakers with frequency-dependent acoustic centers. Such inconsistencies, when encoded at finite SH order, can cause spectral artifacts. We compare simple magnitude-oriented approaches (peak alignment, minimum-phase) with methods that aim to preserve physically meaningful phase via broadband and subband acoustic-center compensation, as well as per-driver encoding with crossover reconstruction. A MUSHRA-like listening test assessed perceptual similarity to a reference based on the original measurements for virtualized stereo reproduction in anechoic and semi-diffuse conditions. Results show that the simpler preprocessing approaches generally yield higher perceptual similarity than acoustic-center-preserving approaches, while per-driver encoding performs comparably to peak alignment at higher orders. An ERB-band spectral error correlates with the perceptual trends, underscoring the primary importance of magnitude fidelity under SH truncation.