Noise Reduction by Beamforming in Cochlear Implants: a Time Domain Study

EasyChair Preprint 9017

Abstract

Cochlear implants are devices with the aim of restoring communication capability in cases of severe to profound hearing loss. For this purpose, acoustic signals are transformed into electrical stimuli and then directly applied to the cochlea through a set of electrodes. Despite its efficacy in restoring communication under optimal conditions, its performance is severely degraded due to additive noise. In this work, a comparison between two variations of the most common beamforming method for noise reduction in cochlear implants is performed. Computational simulations with the Minimum Variance Distortionless Response beamforming (MVDR), using noisy-speech and noise-only correlation matrices were performed, and results for intelligibility and output signal-to-interference-plus-noise ratio are presented. The results indicate that the MVDR beamformer based on the noise correlation matrix provides higher intelligibility and acoustic comfort. This finding justifies the choice of the MVDR cost function in cochlear implant applications.

Keyphrases: beamformer, cochlear implant, noise reduction, speech processing.

@booklet{EasyChair:9017,
  author    = {Luciana M X Souza and Marcio H Costa and Renata C Borges},
  title     = {Noise Reduction by Beamforming in Cochlear Implants: a Time Domain Study},
  howpublished = {EasyChair Preprint 9017},
  year      = {EasyChair, 2022}}