Article
Deep Denoising Sound Coding Strategy for Cochlear Implants (de)
* Presenting author
Abstract:
Cochlear implants (CIs) are very successful in providing speech understanding to people with severe sensorineural hearing loss. Nowadays many users receive bilateral CIs (BiCIs), which provide further benefits in terms of spatial hearing. However, spatial hearing, which is very important to improve speech understanding in noisy environments, is limited with bilateral electric stimulation. For this reason, research focuses on new algorithms that keep the desired speech source while filtering out undesired background noises. New developments in the area of end-to-end speech denoising have been proposed, either as a front-end component or fully integrated in the CI sound coding strategy. This work proposes a deep-learning model that shares information between both hearing sides. Specifically, we connect two monaural end-to-end deep denoising techniques through latent fusion layers. These fusion layers share the latent representations from each side by multiplying them together resulting in enhanced speech and improving learning generalization. Results from objective instrumental measures demonstrate that the proposed fused BiCI sound coding strategy achieves higher interaural coherence, superior noise reduction, and enhanced predicted speech intelligibility compared to the baseline methods. Furthermore, our speech-in-noise intelligibility results in BiCI users reveal that the deep denoising sound coding strategy significnatly improves speech understanding in noise.