Constant Power Psychoacoustic Spectrum Optimization for Loudness and Sharpness with Application to Vehicle Interiors

Abstract

Sound quality assessments are an integral part of vehicle design. Especially now, as manufacturers move towards electrification, vehicle sounds are fundamentally changing. By improving the quality of the interior sounds of a vehicle, consumers’ subjective evaluation of it can be increased. Therefore, the field of psychoacoustics, which is the study of human perception of sound, is broadly applicable here. In fact, the perceived quality of a sound signal is influenced by several psychoacoustic indicators, including loudness, sharpness, and roughness. Of particular utility is identifying in advance how to distribute audible frequency content in a way that optimizes psychoacoustic metrics as this can help automotive engineers obtain specific design targets that optimize vehicle noise, vibration, and harshness (NVH). In this article, a novel modified gradient-based optimization technique (MGOT) is developed to optimize psychoacoustic loudness and sharpness. The new technique is applied to identify targeted adjustments to a measured vehicle interior sound signal that keep the signal energy constant but reduce loudness and/or sharpness. The MGOT numerically approximates the objective function gradient for small changes in the signal power distribution for which constant overall signal power is maintained. These gradient calculations identify power spectrum one-third octave band trades that minimize a sound signal metric that is a weighted sum of loudness and sharpness while conserving the total signal power. A trade consists of a reduction of power content from a one-third octave band designated as a source together with a simultaneous addition of that power to another receiver one-third octave band. In the MGOT, a one-third octave band that is at any time identified as a source can never later become a receiver of power. The MGOT results and execution times are compared with two widely available general-purpose optimization routines (a standard gradient-based optimizer and a “genetic,” non-gradient optimizer) are used to achieve identical optimization objectives. In comparison to existing optimization techniques, MGOT is found to identify spectrum modifications that produce a superior minimization of the objective function for comparable or even reduced execution times. The resultant sound spectrum modifications can guide vehicle structural or calibration design recommendations that realize a preferred frequency distribution for enhancing the vehicle driving experience.