Assessment of Multichannel Acoustic System Identification Using a Spectral-Importance Weighted Misalignment

Modern multimedia applications demand a high degree of spatial realism and aim at providing the audience with an immersive listening experience. Achieving this ambitious goal calls for robust and fast adaptive multichannel audio reproduction schemes. A fundamental prerequisite for most of these techniques is a linear description of the listening environment by means of point-to-point transfer functions. The on-line identification of multichannel acoustic systems, however, is severely hindered by the presence of coherence in the input signals. Unfortunately, commonly used solutions to this problem require decorrelation preprocessing which introduces distortions and renders those approaches unsuitable for high quality audio reproduction setups. In this paper, we show that already the assessment of multichannel system identification is difficult due to the simultaneous effects of non-stationarity, spectral shape, and coherence of real-world system excitations. We therefore propose a spectralimportance weighted misalignment for an improved evaluation of the pure effect of coherence for signals encountered in practical setups such as music or film audio tracks.