End-to-End Magnitude Least Squares Binaural Rendering for Equatorial Microphone Arrays

Paper i proceeding, 2023

We recently presented an end-to-end magnitude least squares (eMagLS) binaural rendering method for spherical microphone array (SMA) signals that integrates a comprehensive array model into a magnitude least squares objective to minimize reproduction errors. The introduced signal model addresses impairments due to practical limitations of spherical harmonic (SH) domain rendering, namely, spatial aliasing, truncation of the SH decomposition order, and regularized radial filtering. In this work, we improve the processing model when applied to the recently proposed equatorial microphone array (EMA) to facilitate three degrees-of-freedom head rotations during the rendering. EMAs provide similar accuracy to SMAs for sound fields from sources inside the horizontal plane while requiring a much lower number of microphones. We compare the proposed end-to-end renderers for both array types against a given binaural reference magnitude response. In addition to anechoic array simulations, the evaluation includes measured array room impulse responses to show the method’s effectiveness in minimizing high-frequency magnitude errors for all head orientations from SMAs and EMAs under practical room conditions. The published reference implementation of the method has been refined and now includes the solution for EMAs.