Fully convolutional network (FCN) model to extract clear speech signals on non-stationary noises of human conversations for cochlear implants

摘要

Cochlear implant (CI) electronically stimulates the nerve to help those with severe hearing lost. However, under noisy backgrounds, speech perception tasks have remained difficult for CI users. Therefore, speech enhancement (SE) is a critical component to improve speech perception examining through different noise scenarios. In this study, we developed the fully convolutional network (FCN) model to extract clear speech signals on non-stationary noises of human conversations in the background, and further compare the model's performance with previously developed log power spectrum (LPS) based Deep neural network (DNN) model's performance by conducting hearing test of enhanced speech which simulated in CI.