Modelling perceptual characteristics of prototype headphones

In this study the characteristics of compact loudspeakers in a stereo setup were investigated. Perceptual evaluations of eleven loudspeakers were conducted on the basis of six selected sensory descriptors, chosen by experienced listeners during consensus meetings. Based on an analysis of the perceptual evaluation, four of the descriptors were found suited for modelling, with the purpose of developing metrics for prediction of Bass depth, Punch, Brilliance, and Dark-Bright. Bass depth and Punch were modelled as one due to high correlation between them. The experimental setup included loudspeaker spinners, enabling fast positioning of loudspeakers. The prediction models were based on binaural recordings, processed using a loudness model, and developed on the basis of previous work on headphone modelling [1, 2]. They were trained on a subset of the data (66%) and validated on the rest. The resulting metrics had high correlations with the perceptual ratings of the validation dataset (r = 0.87-0.96).