Parameter Wiener Filtering Cochlear Denoising Algorithm Based On Explicit Gain Threshold

Cochlear implant(CI)is currently the only device to restore the hearing capability for patients with severe hearing loss.It converts the input sound signal,with specific signal processing strategies,to a series of electric signals,which stimulate the residual auditory nerves and restore the hearing ability.Although the users with state-of-the-art CI devices can obtain good speech perception performance in relatively quiet environments,they are still struggling with background noise distortions.The deficiency might be attributed to the lack of frequency resolution and temporal fine structure in the stimulating signals.Therefore,it is important to develop an effective noise reduction algorithm for CIs.To deal with the abovementioned problem,this study proposes a parametric Wiener filtering denoising algorithm based on apparent gain function(a GT).Firstly,the background knowledge of CI and the existing noise reduction algorithms for CIs are introduced.Then,a successful CI denoising algorithm proposed by Mauger is described in detail.To overcome one of the problems in Mauger' algorithm,i.e.,requirement of massive experiments for searching the optimal gain parameters(?),this study converts the two dimensional searching problem to a one-parameter(a GT)optimization,thus significantly reduce the employees' workload for experimental adjusting the parametersFinally,the performance of the proposed algorithm is verified by a set of vocoder simulated experiment and CI experiment.The result of simulation experiments showed that most subjects can get the best speech recognition rate when the a GT is 5dB.Comparing with other three kinds of noise reduction methods,the proposed one achieves similar denoising performance as by method I and II where priori knowledge of the noise is given.Results also tells that the parametric wiener gain function might be more suitable than the ideal binary masking gain function for CI listeners.The result of CI experiment shows that CI users are more likely to choose gain threshold greater than 0dB comparing to normal hearing listeners who prefer negative thresholds,and most CI listeners obtain best word recognition rate at a GT of 5dB.