Best descriptors for predicting noise annoyance and spectrum-based penalty of steady-state wideband noise

Many objective descriptors (i.e., single-number quantities) are available in the literature for the objective assessment of noise annoyance of wideband, steady-state sounds. However, there is no definitive understanding of which descriptors best correspond to subjective annoyance. Our purpose is to assess how different descriptors correspond with differences in noise annoyance resulting from changes in A-weighted sound level and spectrum. This investigation is based on recently published results from a psychoacoustic experiment conducted by Kuusinen et al. [1]. They investigated 23 steady-state wideband noises with various spectral shapes within 32-48 dB LAeq, and participants rated them in terms of subjective annoyance. The study revealed that penalty due to spectrum shape could be even 10 dB. Here, we present a follow up analysis focusing on objective noise descriptors. The results indicate that many descriptors are strongly associated with noise annoyance ratings, while Spectral Centroid and Sharpness appear to be the best candidates to predict spectrum-based penalty.