Modeling Sound Quality from Psychoacoustic Measures

Sound quality is a complex perceptual measure depending on several factors, which can often be associated with more basic psychoacoustic measures such as roughness, sharpness, tonality or loudness. One frequently used method for modeling the overall quality of sounds is therefore to fit a linear combination of psychoacoustic measures to subjectively measured data of quality. The psychoacoustic measures are usually calculated using standards or descriptions from literature references. However, these standards or references may not always be accurate for the specific class of sounds for which sound quality shall be determined. In this study we therefore applied this method to subjectively measured psychoacoustic measures and compared the results to data from instrumental measures and to data from literature. By means of regression analyses model approaches were derived from parts of the data and used to predict another part of the data for validation. The results indicate that subjectively measured basic psychoacoustic measures are not always in line with standardized calculation procedures, and that subjectively measured annoyance can be predicted quite well by subjectively measured sharpness, loudness, roughness and tonality. These data serve as benchmarks for further model approaches and underline the general applicability of this method.