ISSC 2007, Derry, Sept 13-14 Comparison of Temporal Masking Models for Audio Quality Assessment Dermot Campbell, Edward Jones and Martin Glavin. _______________________________________________________________________________ Abstract— This paper investigates the use and performance of some recently proposed models for temporal masking in wideband audio quality assessment algorithms. Three published temporal masking models are incorporated into the ITU-standardised PEAQ (Perceptual Evaluation of Audio Quality) algorithm, to assess the accuracy of such models in predicting subjective audio quality. The PEAQ algorithm already includes a temporal masking model, however, the paper shows that the performance of PEAQ may be improved for certain distortions by using a different temporal masking model. Furthermore, a novel temporal masking model is presented in this paper, that appears to further improve the perceptual accuracy of the PEAQ algorithm for most distortions examined. Keywords – PEAQ, Temporal Masking, Forward Masking, Audio Quality Assessment. _______________________________________________________________________________ I INTRODUCTION Wideband audio quality assessment tests are used in a number of applications in consumer electronics and telecommunications, including the development and testing of multimedia devices, codecs and networks. Furthermore, they can be used to assess and compare various devices and audio coding algorithms. Traditionally, tests using human listeners have been used for audio quality assessment. However, these subjective listening tests are time consuming and expensive as they require expert listeners and specialised equipment. To remove the need for such tests, objective (computer-based) audio quality assessment algorithms are now used. These algorithms are normally perceptually based and consist of a psychoacoustic model to model the human auditory system and a cognitive model to model how the average person grades the quality of audio signals with impairments. A generic block diagram for such models is shown in Figure 1. The most commonly used audio quality assessment algorithms are intrusive, in that they require both a reference and a degraded signal. Figure 1: Generic Block Diagram for Audio Quality Assessment Algorithms. The ITU (International Telecommunications Union) has published a standard for audio quality assessment called PEAQ (Perceptual Assessment of Audio Quality) [1],[5],[6],[7]. It is used to assess the quality of signals with very small impairments by comparing the impaired (i.e. degraded) signal to the original (reference) signal. Since PEAQ was standardised there have been a number of investigations into the performance of the algorithm [5],[8] but there has been little work done in attempting to improve the accuracy of the standard itself. Temporal masking is a perceptual phenomenon whereby one sound (the masker) can influence the perception of another sound (the maskee), at the same or a similar frequency, which occurs either before or after the masker. The standardised PEAQ Department of Electronic Engineering , National University Of Ireland Galway, Galway, IRELAND dermot.campbell@nuigalway.ie , edward.jones@nuigalway.ie , martin.glavin@nuigalway.ie .