Construction of Ambience Bases from Weighing Matrices with Application in Spatial Audio Coding

摘要

Parametric spatial audio coding schemes, such as advanced joint channel coding in Dolby's next-generation audio coding system AC-4, achieve a higher data compression ratio as a result of a lower-dimensional intermediate signal representation, known as the downmix. During the inverse process, the upmix, which is guided by side information, the covariance between the source signals is reconstructed to preserve perceptually important cues such as ambience or source width. In this manuscript, a systematic approach for the construction of ambience bases from weighing matrices is presented. Furthermore, the basis vectors are generalized to accommodate for nonunitary mixing weights, and a new basis is derived. Round figures from internal listening tests are shared to underpin the utility of the approach.