Morphologic clustering of earcanals using deep learning algorithm to design artificial ears dedicated to earplug attenuation measurement

Abstract

Designing earplugs adapted for the widest number of earcanals requires acoustical test fixtures (ATFs) geometrically representative of the population. Most existing ATFs are equipped with unique sized straight cylindrical earcanals, considered representative of average human morphology, and are therefore unable to assess how earplugs can fit different earcanal morphologies. In this study, a methodology to cluster earcanals as a function of their morphologies with the objective of designing artificial ears dedicated to sound attenuation measurement is developed and applied to a sample of Canadian workers’ earcanals. The earcanal morphologic indicators that correlate with the attenuations of six models of commercial earplugs are first identified. Three clusters of earcanals are then produced using statistical analysis and an artificial intelligence-based algorithm. In the sample of earcanals considered in this study, the identified clusters differ by the earcanal length and by the surface and ovality of the first bend cross section. The cluster that comprises earcanals with small girth and round first bend cross section shows that earplugs induced attenuation significantly higher than the cluster that includes earcanals with a bigger and more oval first bend cross section.