Authors: Rutendo Chatiza¹; Lidea Shahidi¹; Robert P. Carlyon¹; Francois Guérit¹
¹Cambridge Hearing Group, MRC Cognition and Brain Sciences Unit, University of Cambridge
Background: Cochlear implant (CI) users typically understand speech in quiet conditions, but in noisy conditions even the most successful struggle. One contributing factor is channel interactions, whereby the electrical fields generated by multiple CI stimulating electrodes overlap, leading to unintended co-stimulation of auditory nerve fibres. Auditory-nerve computational models could help us design and evaluate stimulation methods that reduce channel interactions, allowing rapid testing of varied stimuli. However, these models are inspired from animal data and are rarely validated against CI psychophysical literature. Here, we validate a computational model by comparing it to several published CI psychophysical datasets.
Method: We select papers from the CI psychophysical literature with within-participant manipulations (e.g. the effect of temporal and spatial separation between two pulse trains on loudness summation) that investigated interactions both within and across channels. We then simulate them via the model by Brochier et al. (2022) and test different metrics to compare the data against the modelled neural activation patterns. Finally, we investigate the impact of changing the model selectivity and time constants (refractoriness, central integration) on the fit between modelled neural activity and psychophysical data.
Results: Ongoing results show modelled outputs that are overall in range of the data found in the literature, as well as expected and unexpected differences between the modelled and published within-participant effects. This suggests necessary improvements in the model parameters and/or improvements in the metrics used to map modelled neural activity to psychophysical data.
Conclusion: By improving the comparison between the output of CI computational models and known within-participant effects, computational models can improve our understanding of the complex impact of channel interactions on speech perception with CIs.
