End-to-End Magnitude Least Squares Binaural Rendering of Spherical Microphone Array Signals

Paper i proceeding, 2021

Spherica1 microphone array (SMA) recordings are particularly suited for dynamic binaural rendering as the microphone signals can be decomposed into a spherical harmonic (SH) representation that can be freely rotated to match the head orientation of the listener. The rendering of such SMA recordings is a non-trivial task as the SH signals are impaired due to truncation of the SH decomposition order, spatial aliasing and the gain limitation of the employed radial filters. The perceptually most relevant consequence of this is an alteration of the magnitude transfer function at high frequencies. Previously, the magnitude least squares (MagLS) renderer for binaural rendering of SH signals was proposed to mitigate these effects under the assumption of ideal order-truncated plane waves, i.e., disregarding the influence of spatial aliasing as well as of non-ideal radial filters. Based on the MagLS renderer, we present a binaural rendering method for SMA recordings that integrates a comprehensive SMA model into the magnitude least squares objective. We evaluate the proposed end-to-end renderer by analyzing the reproduced binaural magnitude response. Our results suggest that the method significantly improves the high-frequency rendering mainly due to the inherent binaural diffuse-field equalization, while it achieves a slight improvement in the low and mid frequency range, where the error of the conventional method is already small. A reference implementation of the method accompanies this paper.