The perceptual analysis measurement system for robust end-to-end speech quality assessment

The perceptual analysis measurement system (PAMS) is an objective model designed to evaluate the perceived speech quality of telephone networks. Two key network properties, linear filtering and variable bulk delay, made previous models unsuitable for end-to-end measurement. This paper outlines the innovations that allow PAMS to take these properties into account. Variable delay occurs in packet-based transmission such as voice over IP. It is addressed by a time alignment algorithm that identifies delay changes in silent periods and during speech. Linear filtering, typically caused by analogue interfaces, is dealt with using a transfer function estimation and equalisation technique designed for robustness to non-linear distortions. A further development is the use of a constrained non-linear regression method to map error parameters to predicted subjective quality. Applications are discussed including the selection, optimisation and monitoring of telephone systems. Results are presented performance of PAMS across 28 telephony subjective tests.