Assessing subjective perception of audio quality by measuring the information flow on the brain-response channel

In this paper, we use mutual information (MI) as a measure to quantify the subjective perception of audio quality by directly measuring the brainwave responses of human subjects using a high resolution electro-encephalogram (EEG). Specifically, we propose an information theoretic model to interpret the entire 'transmission chain' comprising stimulus generation, brain processing by the human subject, and EEG measurements as a nonlinear, time-varying communication channel with memory. In the conducted experiment, subjects were presented with audio whose quality varies between two quality levels. The recorded EEG measurements can be modeled as a multidimensional Gaussian mixture model (GMM). In order to make the computation of the MI feasible, we present a novel approximation technique for the differential entropy of the multidimensional GMM. We find the proposed information theoretic approach to be successful in quantifying audio quality perception, with the results being consistent across different subjects and distortion types.