Simulation of unilateral and bilateral cochlear implant for speech recognition in spatialized noise

Authors: Mengchao Zhang1, Christine Du Plessis1

1Aston University

Background: This study examines the impact of sides of cochlear implantation and the impact of informational masking on recognizing speech in spatialized noise by testing the simulated effects of cochlear implant (CI) in normal-hearing listeners.

Method: 10 adults with hearing thresholds below 20 dB HL participated in the study. The effect of CI was simulated using the channel vocoder, SPIRAL, where dense tonal carriers are used (Grange et al., 2017). The experimental conditions were varied based on the implanted side (unilateral vs bilateral), spatial location of the masker (masker at -90, 0, and +90 degrees), and masker type (steady state noise, modulated noise, and single-talker interferer). The spatial effect of the stimuli was synthesized using the head-related transfer functions of behind-the-ear microphones derived from ARI BTE database. Target speech was always located in front of the participants. Unilateral implantation was simulated by disabling the input in the right ear. Speech recognition threshold (SNR at 50% intelligibility) was measured using adaptive procedure with masker level fixed at 65 dB SPL.

Results: Bilateral CIs outperform unilateral CI for all masker types when the masker is located on the left side (-90 deg, also the implanted side for unilateral CI). When the masker is located on the right (+90 deg) or at the front (0 deg), the difference between the implanted sides is smaller. Single-talker interferer produces the greatest interference among the three types of maskers, potentially due to informational masking.

Conclusion: Bilateral CI provides critical benefits to spatial listening in noise compared to unilateral CI. The bilateral benefit is also dependent on the type of masker. The results of the study also support past findings from real CI patients. The novelty of using a dense-carrier channel vocoder like SPIRAL for CI simulation provides an efficient way of studying the effect of CI in listening in complex environments.