Data-driven segmentation of audiometric phenotypes across a large clinical cohort