Perceptual and audiometric predictors of preferred digital noise reduction strength

Modern digital noise reduction (DNR) systems in hearing aids, including those using deep neural networks, suppress background noise by analyzing acoustic features of the environment. However, these systems do not incorporate listener-specific metrics such as Acceptable Noise Level (ANL) or Quick Speech-in-Noise (QSIN) scores, despite their clinical relevance. A key reason is the lack of validated correlations between these listener metrics and current DNR performance, making individualized tuning difficult. This study examines the relationship between ANL and pure-tone thresholds (PTA), addressing past limitations where PTA was averaged across frequencies. Such averaging may obscure frequency-specific patterns important for understanding noise tolerance. Principal component analysis is used to preserve spectral variation and improve the interpretability of audiometric data. Additionally, participants completed a forced-choice task in which they selected their preferred level of DNR while listening to speech in background noise. Stimuli varied across signal-to-noise ratios and reverberation conditions to simulate realistic environments. Preferences were analyzed in relation to each listener’s ANL, QSIN, and SSQ12 scores to determine whether noise tolerance or speech-in-noise ability predicts DNR preference. This work aims to inform listener-centered fitting strategies and support more individualized approaches to hearing aid signal processing in complex acoustic environments.