Cognitive factors influencing outcomes with advanced noise management

Abstract

Speech intelligibility in noise varies widely across hearing aid users despite noise management features. Traditional approaches, such as adaptive beamforming and single-channel noise reduction, can reduce background noise but may introduce signal distortions that degrade speech intelligibility, particularly for listeners with poorer cognitive abilities. Newer deep neural network-based noise reduction (DNN) can potentially improve signal-to-noise ratio while minimizing distortions. This study aims to characterize how individual cognitive abilities influence outcomes with DNN relative to adaptive beamforming across acoustic scenes. Listeners with hearing loss completed sentence intelligibility tasks using wearable hearing aids programmed with adaptive beamforming and DNN. Target speech and competing maskers were presented from on- and off-axis locations in co-located and spatially separated configurations, with masker types including gender-matched two-talker speech or diffuse noise. Participants also completed a cognitive test battery assessing working memory, processing speed, and attention control. Results-to-date indicate that on average beamforming results in higher intelligibility with two-talker maskers, whereas DNN provides better intelligibility in diffuse noise and for off-axis targets, with large individual variability. The presentation will discuss how variability in intelligibility across these noise management strategies relates to individual cognitive profiles and susceptibility to signal distortions.