Objective Measurement of Spatial Audio Coding Quality Based on MNLR Mapping Model

Objective audio quality of spatial audio coding systems is difficult to be measured because of the limited measurement capability under multiple channels especially at low bit rate. Conventional PEAQ tool computes the basic audio quality based on single channel signal and cannot describe the reconstruction effectiveness of the spatial information. The basic quality and the spatial effect are two major attributes in the subjective evaluation of multi-channel audio signal which should also be included in the objective measurement. This paper firstly compute the basic quality with canonical coefficient weighted PEAQ across different channels and then obtain the spatial distortion through comparing three kinds of spatial parameters between the original and the decoded audio signal. Multivariate Non-Linear Regression (MNLR) combined with Principal Component Analysis (PCA) is used to design the mapping model to predict the subjective score from the weighted PEAQ value and the spatial distortion. Formal listening tests are carried out to generate the training and test data with 5.1-channels spatial audio signals under different parametric coding schemes. Preliminary results with 5.1-channels audio show that the proposed objective measurement method can describe the spatial audio coding quality better than the conventional PEAQ method.