A Deep Denoising Sound Coding Strategy for Cochlear Implants

Abstract

Cochlear implants (CIs) have proven to be successful at restoring the sensation of hearing in people who suffer from profound sensorineural hearing loss. CI users generally achieve good speech understanding in quiet acoustic conditions. However, their hearing ability degrades drastically in noisy backgrounds. To address this problem, current CI systems are delivered with front-end speech enhancement processors that can be beneficial for the listener, however, these perform well only in certain noisy environments, leaving quite some room for improvement in more challenging conditions. In this work, we propose substituting the CI sound coding strategy with a deep neural network (DNN) that performs end-to-end processing by taking the raw audio as input and providing a denoised electrodogram, i.e., the electrical stimulation patterns that are applied to the electrodes across time. We specifically design a DNN to emulate a common CI sound coding strategy; the advanced combination encoder (ACE). The DNN is designed not only to faithfully emulate the coding of acoustic signals that ACE would perform but also to remove unwanted interfering noises, when present, without sacrificing processing latency. The models were optimized using CI-specific loss functions and evaluated using objective instrumental measures and listening tests in CI participants. Results show that the proposed models achieve higher signal-to-noise ratio improvement and objective intelligibility scores when compared to the baseline models. Furthermore, the proposed deep learning-based sound coding strategy improved speech intelligibility in eight CI users.

Recovering speech intelligibility with deep learning and multiple microphones in noisy-reverberant situations for people using cochlear implants

“…Traditional methods can sometimes provide improvements in speech intelligibility in stationary background noise (Chen et al, 2015;Dawson et al, 2011;Hu et al, 2007;Loizou et al, 2005;Mauger et al, 2012;Wang and Hansen, 2018) but usually fail to improve intelligibility in more realistic non-stationary background noise (Baumgärtel et al, 2015;Bentsen et al, 2019;Bolner et al, 2016;Lai et al, 2018). In more realistic listening situations with non-stationary noise, deep neural networks have been shown to overcome previous limitations (Healy et al, 2023) and achieved significant improvements for speech intelligibility in background noise by CI listeners (Gajecki et al, 2023;Goehring et al, 2017Goehring et al, , 2019Kang et al, 2021;Lai et al, 2018). These findings for CI listeners are in line with results obtained for people with mild or moderate hearing loss, for example users of other assistive hearing devices such as hearing aids (Goehring et al, 2016;Healy et al, 2023Healy et al, , 2013Keshavarzi et al, 2019;Monaghan et al, 2017).…”