Performance Comparison And Analysis Of Several Typical Noise Reduction Algorithms For Cochlear Implants

Cochlear implant(CI)is an implantable device that helps restore hearing in patients with severe hearing loss.With the implanted electrode array in the inner ear,CI stimulates directly the auditory nerves,which is done by the hair cells in normal hearing(NH),such to fulfill the acoustic-to-electric conversion.State-of-the-art CI technology has enabled CI recipients to enjoy a high level of speech understanding in quiet.However,in real-world speech communications,the target speech is usually corrupted by acoustic interference,which reduces the intelligibility of speech.The artificial electric hearing provided by CI is still far from satisfactory for noisy speech perception.Noise suppression has been one of the key techniques for CI signal processing.In this thesis,several typical noise suppression algorithms proposed for CIs are introduced and their performance are discussed.An improved version upon the classical parametric Wiener filtering(PWF)is proposed,in which sub-band optimal gain thresholds are experimentally determined and implemented for CI noise reduction.A set of experiments are carried out to evaluate the noise reduction performance of different algorithms.Since signal processing strategies of modern CIs mostly focus on extracting the temporal envelopes of the input acoustical signal,this study investigates the performance of different denoising methods when applied to either the acoustical sub-band signal or the envelopes in the sub-bands.Given the ideal noise estimation assumption,the proposed algorithm outperforms the classical PWF one with respect to the SNR improvement of the envelope.Furthermore,two noise estimation methods are implemented to investigate the performance of different denoising algorithms in unknown noise cases.Finally,Mandarin speech recognition by NH subjects with vocoder simulated sentences are conducted to evaluate the noise reduction performances.Results show that the proposed algorithm with sub-band optimal thresholds achieves better performance.For example,the IMCRA noise estimation method is used,the speech recognition rate by the proposed algorithm when applied to either the acoustical sub-band signal or the envelopes in the sub-bands,has been improved by 12% and 35.19%?...