Estimating the distortion component of hearing impairment from attenuation-based model predictions using machine learning

David Hülsmeier¹, Mareike Buhl¹, Nina Wardenga², Anna Warzybok¹, Marc René Schädler^1,3, Birger Kollmeier¹

¹ Medizinische Physik and Cluster of Excellence Hearing4all, CvO Universität Oldenburg, 26111 Oldenburg, Germany; ² Department of Otolaryngology and Cluster of Excellence Hearing4all, Hannover Medical School, Hannover, Germany; ³ Vibrosonic GmbH, Mannheim, Germany

Background: Hearing impairment affects the ability to understand speech. This can be described according to Plomp (1978) by an attenuation and a distortion component. The attenuation component affects speech recognition in quiet and is linked to the absolute hearing threshold. The supra-threshold distortion component affects speech recognition in quiet and noise while its origin remains speculative. Such supra-threshold deficits cannot be compensated by amplification and no “simple” measurement method exists. Yet, separating the attenuation component from the distortion component appears possible by using speech recognition models: When only using the attenuation component for modeling, differences between predicted and measured SRTs can be interpreted as an estimate of supra-threshold deficits.

Methods: Published speech recognition thresholds (SRTs) in noise of 315 hearing-impaired ears were predicted with the machine-learning-based framework for auditory discrimination experiments (FADE), the speech intelligibility index (SII), and a modified SII with a hearing-loss-dependent band importance function (PAV). Their attenuation-component-based prediction errors were interpreted as estimates of the distortion component.

Results: Overall, the models showed root-mean-square errors (RMSEs) of 7 dB, but for steeply sloping hearing loss FADE and PAV were more accurate (RMSE=9 dB) than the SII (RMSE=23 dB). Prediction errors of FADE and PAV increased linearly with the average hearing loss. This linear relation was used as distortion component estimate whose application significantly improved the accuracy of FADE’s and PAV’s predictions.

Conclusion: Simulations with FADE and PAV imply, that the supra-threshold distortion component increases linearly with average hearing loss. Accounting for a distortion component improves the model predictions and implies a need for effective compensation strategies for supra-threshold processing deficits with increasing audibility loss.