Noise reduction for cochlear implants by enhancing the temporal envelope

Although the state-of-the-art cochlear implant (CI) devices have been very successful in recovering the hearing ability of the implantees in relatively quiet environments, the implantees are still struggling against the background noise. This paper presents a noise reduction algorithm for CI with a commonly implemented signal processing strategy, i.e., the Continuous Interleaved Sampling (CIS). Instead of denoising the acoustical signal input to the CI processor, the proposed algorithm applied a multi-band Wiener filtering for suppressing the noise components in the temporal envelopes that serve as the stimuli (after interleaved sampling by a biphasic pulse train) to the auditory nerves. To do so, the signal-to-noise ratios of the noisy envelopes were first estimated and then parametric Wiener gain functions were determined to suppress the noise components in the envelopes. A set of vocoder simulation experiments were conducted to find a perceptually optimized gain threshold, which is also essential for determining the gain function. Furthermore, several noise estimation methods were applied to examine their contributions to noise reduction performances with different kinds of noise sources. The performance of the proposed algorithm was compared with the baseline method in which a similar noise reduction process was applied to the acoustical sub-band signals. Results showed that the proposed algorithm is more effective than the baseline, in terms of the SNR of the temporal envelopes, the quality of the vocoded speech and the recognition rate for the vocoded speech as well.