Prediction Model of Multi-channel Audio Quality Based on Multiple Linear Regression

Perceived audio quality is an important metric to measure the perception degradation of multi-channel audio signals especially for coding and rendering systems. Conventional objective quality measurement such as PEAQ (Perceptual Evaluation of Audio Quality) is limited to describe both the basic audio quality and the spatial impression. A novel prediction model is proposed to predict the subjective quality of 5.1-channels audio systems. Two attributes are included in the evaluation including basic quality and surround effects. Multiple Linear Regression (MLR) combined with Principal Component Analysis (PCA) is used to establish the prediction model from the objective parameters to subjective audio quality. Data set for model training and testing is obtained from formal listening tests under different coding conditions. Preliminary experiment results with 5.1-channels audio show that the proposed model can predict multi-channel audio quality more accurately than the conventional PEAQ method considering both the basic audio quality and the surround effects.