Improvement in speech-in-noise listening with deep neural network enabled hearing aids
TL;DR
The results reported here demonstrate the benefits of unique noise reduction features to improve listener performance across multiple objective measures, and listener variability may in part be explained by suprathreshold and cognitive abilities.
Abstract
Despite technical innovations in amplification and signal processing, hearing aid users still frequently report difficulty understanding speech in noise. Recent advances in hearing aid technology using deep neural networks (DNNs) promise to expand benefits of amplification beyond traditional noise reduction features. The present study evaluated the efficacy of a commercially available DNN-enabled hearing aid in a free-field speech understanding task using the coordinate response measure corpus with noise background. We compared multiple configurations of the DNN-enabled device (DNN-on versus DNN-off) as well as comparison devices with traditional noise reduction technologies. Results showed a significant benefit of the DNN-based noise reduction represented by higher CRM accuracy at all target locations, with the highest behavioral performance improvement observed at lateral target locations (±120 deg) for the DNN-enabled device compared to the same device with the DNN-off or the comparison devices. Furthermore, individual behavioral and cognitive measures were shown in some cases to have direct associations with this benefit. The results reported here demonstrate the benefits of unique noise reduction features to improve listener performance across multiple objective measures, and listener variability may in part be explained by suprathreshold and cognitive abilities.