Authors: Matthias Inghels¹; Chuan Wen¹; Brent Nissens¹; Nele De Poortere¹; Guy Torfs¹; Sarah Verhulst¹
¹Ghent University
Background: Hearing loss compensation has traditionally focused on standard audiometric deficits, where reduced audibility is addressed through amplification and frequency-specific signal shaping. However, hidden hearing loss—often associated with cochlear synaptopathy and impaired neural coding despite near-normal audiograms—can substantially degrade speech perception in noise. These limitations motivate computational approaches that move beyond conventional gain-based hearing aid strategies to account for both threshold and supra-threshold deficits.
Method: We investigated computational closed-loop methods for personalized hearing compensation by combining auditory modeling and machine learning to estimate individual perceptual deficits and optimize signal processing. The evaluated approaches included speech enhancement, noise-robust feature extraction, and closed-loop fitting strategies informed by physiological measurements, with a specific comparison against the standard NAL-NL2 prescription rule.
Results: The proposed approach demonstrated improvements in HASPI v2 metrics compared with NAL-NL2. A MUSHRA test with 53 patients indicated improved perceived quality over NAL-NL2 and comparable quality to the unprocessed reference condition. Furthermore, preliminary performance testing of the outer hair cell (OHC) compensation algorithm in 14 patients showed an average improvement of 12.5% on a fixed-SNR 20 sentence matrix test.
Conclusion: Computational closed-loop compensation can improve personalization beyond conventional gain-based hearing aid fitting by integrating data-driven models with auditory-domain knowledge. These results support the development of next-generation assistive hearing technologies that better reflect the complexity of real hearing impairment, including both standard and hidden hearing losses.
