Can current room-acoustics indices specify the quality of aural experience in concert halls?

Abstract

It is a dream of acousticians to identify indices that can be used to specify the quality of aural experience in concert halls but can nevertheless be measured with physical instruments. A battery of such indices has been standardized by the International Organization for Standardization, ISO (ISO3382-1, 2013, 2009), particularly concerning the role of the hall in the quality-formation process. Yet, even considering improvements as proposed in the meantime, for example, by Bradley (2010), these and related indices are derived from physical data and not from perceptual ones, namely, perceptual properties of aural objects and scenes. Consequently, they disregard interindividual perceptual differences. To analyze the problem, one should take the perspective of concert goers, sitting in a hall and listening to a musical performance. There is no doubt that the listeners do not care about any physics-based indices in this situation. They judge on what they actually perceive: that is, on what they hear and see, on vibrations they may feel, and maybe also on smells that they sense. This conglomerate of percepts is set in relation to their expectations, namely, to internal references that are not directly accessible to external observers (Raake & Blauert, 2013; Raake & Egger, 2014). The extent to which these expectations are met will then determine their quality judgment (Jekosch, 2005a). The internal references are individual and depend, among other things, on the actual music played and the knowledge and experience that the listeners have with the particular kind of music. It has, however, to be stressed at this point that only experienced listeners are able to judge separately on the contribution of the hall to the overall quality of aural experience in it.Without doubt, various physics-based indices show significant correlation with perceptual features; for example, the reverberation time is related to reverberance and the lateral-energy fraction to the spatial extents of aural objects. Yet, all of these suffer from the fact that they abstract from concrete individual observations and thus disregard individual perceptual particulars. Further, the correlation of such indices with the corresponding aural features is limited in principle. Ergo, there is a problem of validity when it comes to the prediction of quality-of-aural-experience judgments in general and, even more so, judgments by individual listeners.In an attempt to increase the amount of correlation of the results of physical measurements with aural features, algorithms have been designed that mimic elements of peripheral auditory signal processing. Yet, although this may increase the correlation between acoustics and aural perception, these algorithms do not directly measure perceptual features. At best, they provide estimates of (primitive) perceptual attributes such as loudness, roughness, or source direction. These estimates incorporate some understanding of the peripheral auditory system but still suffer from the disadvantage of abstracting from individual listeners. Thus, although the validity problem has been reduced, it still persists.It goes without saying that sound quality is an important component of aural experience in performance spaces. As mentioned above, quality judgments require internal reference of the listeners. To get access to these references by external observation is very tedious, as they are concepts in the listeners' minds. As a consequence, developing models that functionally mimic human quality formation is a complex task. Hence, respective more sophisticated instrumental approaches to sound-quality assessment, as, for example, ITU-PEAQ (a standardized algorithm for assessing soundquality degradation due to transmission [e.g., Rumsey, 2010; Thiede et al., 2000]), do not incorporate internal references. In fact, algorithms like this still abstract from individuality. Consequently, they also do not really measure perceptual quality but only provide generalized estimates of it. …