Non-intrusive speech-quality assessment using vocal-tract models

Abstract

It is increasingly vital that there are effective quality of service (QoS) metrics to describe the performance of telecommunications networks. Speech quality is a major contributor to users' perception of QoS, and the ability to design for and monitor this quality is paramount. The authors describe work towards a non-intrusive speech quality assessment algorithm, capable of making predictions of the speech quality received by a customer, utilising the in-service signal. Modern telecommunications networks contain complex nonlinear elements that cannot be assessed with traditional engineering metrics. A novel use of vocal-tract modelling techniques is described, which enables predictions of the quality of a network degraded speech stream to be made. Details of the algorithm's adaptation to different talker characteristics are presented, together with a summary of the performance of the system.