Reverberant speech recognition with actual cochlear implants: verifying a pulsatile vocoder simulation method
* Presenting author
Classic channel vocoders have been widely used for simulating cochlear implants (CIs). However, some important features of state-of-the-art CI strategies were usually ignored, such as pulsatile current, broad current spread, and low quantization resolution of intensity. To solve this problem, a novel vocoder was proposed, which is combined from Gaussian-shaped acoustic pulses each corresponding to an electric pulse and can directly transformed CI-electrodograms to sounds. Compared with the classic channel vocoders on a reverberant speech recognition task, this new vocoder was considerably more sensitive to the reverberant condition (published in APSIPA-2018). Moreover, its performance is comparable to actual CI users in previous CI-reverberation studies. Here, the same paradigm as in the APSIPA-2018 paper was tested on actual CI users. We found that under reverberant conditions the CI users got comparable scores to the simulated performance from the new vocoder simulation in the NH cohort, while the classic vocoder predicted significantly much higher performance This study demonstrated that the classic vocoders overestimated the task performance and further lent great support to the usage of the novel vocoder for the simulation of reverberant speech recognition in adult post-lingually deafened CI users.