Acoustic simulation of cochlear implant sound to approximate the perceptual experience of electric hearing

TL;DR

This study created German sound samples to simulate how cochlear implant users hear, finding that optimized simulations closely matched their perception but varied with different sentences.

Contribution

The study introduces a method to simulate the sound of cochlear implants in German and evaluates its accuracy across different speech samples.

Findings

Optimized simulations achieved high similarity scores (9.7/10) for the initial sentence.

Similarity scores dropped significantly for different sentences, indicating variability in simulation accuracy.

Some participants showed no change in similarity scores across different sentences.

Abstract

The electrical signal processing of cochlear implants (CIs) is thought to partially explain the unnatural and unfamiliar sound impressions in recipients. To date, there are no accurate German-language sound samples available that represent how CI users perceive their CIs. The primary aim of this study was to create German-language sound samples for the sound of Cochlear Nucleus implants (Cochlear Ltd., Sydney, Australia). Furthermore, we investigated whether the simulation parameters best matched for one sentence were also accurate approximations of the CI sound when applied to two further sentences spoken by the same male speaker. Fifteen patients with single-sided deafness who had at least two years of experience with their CI were included in this study. The participants rated ten simulations based on the similarity to the sound perceived with their devices. The simulation with the best similarity score served as the starting point for optimization using a software sound-tool. On average, a score of 9.7 ± 0.5 on a scale of 1 (no similarity) to 10 (signals are identical) was achieved for the optimized simulations. Most often, a low-pass filter or a comb filter was required to optimize the simulation. The sound samples optimized for each study participant are provided in the Supplemental Material. When the parameters of the optimized simulation were applied to two further sentences (variable in content and phonemes), the similarity scores were significantly worse (mean sentence 2: 8.4 ± 1.5, Z = 3.015, p = 0.003, mean sentence 3: 8.9 ± 1.3, sign test: Z = 2.268, p = 0.048). However, when the similarity scores were considered individually for each participant, there were 4 (sentence 2) or 7 (sentence 3) participants, respectively, for whom the speech material had no influence on the similarity score.

The online version contains supplementary material available at 10.1038/s41598-025-25711-z.

Figures