Note: This bibliographic page is archived and will no longer be updated. For an up-to-date list of publications from the Music Technology Group see the Publications list .

A Computational Model for Spectral Modulation and Speech Perception Performance of Cochlear Implant Users

Title A Computational Model for Spectral Modulation and Speech Perception Performance of Cochlear Implant Users
Publication Type Master Thesis
Year of Publication 2019
Authors Alvarez, F. Y.
Abstract In the development of new technologies for cochlear implants (CIs), there are many evaluation procedures that have to be done until it is implemented on hearing devices. The majority of such procedures consist of asking implanted volunteers to perform auditory tasks, but collecting the required data is time demanding. In this work, a computational model is presented to simulate two of the many tasks typically used to evaluate CI users performance: speech reception threshold (SRT), and spectral modulation threshold (SMT). The model uses a cochlear implant sound coding strategy, a peripheral auditory model to obtain the spike activity from auditory nerves and an adapter to extract the internal representation (IR). As backend, the simulation framework for auditory discrimination experiments (FADE) is implemented. The performance of three different CI sound coding strategies were compared, one with sequential stimulation and the other two with simultaneous stimulation with two and three channels respectively. Also, the impact of nerve fiber degeneration and number of functional fibers was studied. The model successfully performed SRT and SMT experiments resembling CI users. The CI sound coding strategy with triplet simultaneous stimulation performed the best in SRT experiments contrary to the empirical data, but the cause is discussed to be a mismatch in the overall firing rate between sound coding strategies. In SMT experiments the results were quite similar among the strategies. With worse nerve conditions, SRT performance decreased significantly while SMT performance barely changed. It is discussed that the model may be relying in the overall firing rate instead of the spectral shape to identify spectral ripple stimuli. Finally, It is discussed how different features affects the performance of the model, and whether is necessary to implement a different back-end to avoid modeling central auditory processes.
Final publication