Sound Field Model Violations in Parametric Spatial Sound Processing

Inhalt:
Describing a sound field with a single audio signal and parametric side information offers an efficient way of storing, transmitting, or reproducing the recorded spatial sound. In many approaches, the parametric representation is based on a simple sound field model assuming for each time-frequency instance a single plane wave and diffuse sound. When multiple sources are active, we thus have to assume that the source signals do not overlap much in the time-frequency domain so that the single-wave model is still satisfied. In this paper, we show that a single-wave model is easily violated even for sparse signals such as speech. When the parametrized spatial sound is reproduced, these violations degrade the spatial quality. We further study whether the well-known measure of approximate W-disjoint orthogonality can be applied to detect the relevant model violations. Index Terms — Spatial audio coding, parametric spatial audio processing,W-disjoint orthogonality