Robust perceptual assessment of end-to-end audio quality

Abstract

Perceptual quality assessment models were initially developed to predict subjective quality of codecs. Experience with telephony applications has found that today's complex networks make assessment difficult. Analogue interfaces and variable delay are amongst the technologies used in current voice transmission systems-and often make the first generation of perceptual models produce inaccurate scores. It has been necessary to extend perceptual models for use in whole network and field applications. In this paper we describe the perceptual analysis/measurement system (PAMS), a model specifically designed for end-to-end assessment of complex telephone networks. We focus on the method used to estimate subjective quality, applying constraints to make robust predictions and enhance the model's generality.