Audio Signal Processing and Coding

12.1 INTRODUCTION

In many situations, and particularly in the context of standardization activities, performance measures are needed to evaluate whether one of the established or emerging techniques is in some sense superior to other alternative methods. Perceptual audio codecs are most often evaluated in terms of bit rate, complexity, delay, robustness, and output quality. Reliable and repeatable output quality assessment (which is related to robustness) presents a significant challenge. It is well known that perceptual coders can achieve transparent quality over a very broad, highly signal-dependent range of segmental SNRs ranging from as low as 13 dB to as high as 90 dB. Classical objective measures of signal fidelity such as signal-to-noise ratio (SNR) or total harmonic distortion (THD) are therefore inadequate [Ryde96]. As a result, time-consuming and expensive subjective listening tests are required to measure the small impairments that often characterize perceptual coding algorithms. Despite some confounding factors, subjective listening tests are nevertheless the most reliable tool available for codec quality evaluation, and standardized listening test procedures have been developed to maximize reliability. Research into improved subjective testing methodologies is also quite active. At the same time, considerable research is being devoted to the development of automatic perceptual measurement systems that can predict accurately the outcomes of subjective listening tests. ...