Ambient-Aware Sound Field Translation Using Optimal Spatial Filtering

In a previous contribution, we proposed a space-warping-based approach for sound field translation of non-reverberant higher-order Ambisonics signals with applications in spatial audio and virtual reality. In this work, we extend the concept of space warping in order to deal with ambient sound such as reverberation and diffuse noise by using spatially selective filtering. We propose a hard-decision and a soft-decision approach which both make use of the second-order statistics of the signal. The hard-decision variant yields improved performance with respect to the non-adaptive reference for low SNRs and is robust against covariance misestimates. The soft-decision variant is the solution to an optimal spatial filter derivation. It yields optimal performance for known covariances and easily outperforms the hard-decision and reference approaches also for moderate and high SNRs. We further derive expressions for the expected errors and relate our findings to the mathematically related problem of spherical-harmonics-domain noise reduction.