MATLAB-Based Implementation of a Spatial Covariance Matching Spread Rendering Method

Abstract

A MATLAB implementation of the spatial covariance matching method proposed by McCormack et al. is developed for the case of binaural reproduction and implemented in an afSTFT framework using HRTFs, alongside baseline rendering (BL) and eigenvalue decomposition (EVD) and optimal mixing (OM). In the OM method, both prototype and residual mixing matrices are generated by the covariance-domain optimization process as introduced in the original method. The evaluation of per formance is done in terms of colouration, interaural level difference (ILD), interaural coherence (IC), RMSE analysis over 836 source directions at different source spreads, and signal distortion error (SDE). The results indicate that EVD improves coherence matching and spatial diffuseness and OM yields a compromise between spatial widening and signal preservation. The global performance is consistent overall with the reference implementation, but differences exist between the implementations for certain metrics and signal types in terms of quantity.