一种自适应变阶谱减降噪算法在电子耳蜗中的应用

由于传统谱减语音增强存在残留的“音乐噪声”,因此基于传统谱减法降噪的电子耳蜗(CI)感知的声音品质也会受到影响。为提高CI的抗噪性,本文提出了一种自适应变阶谱减算法,并将该方法应用于电子耳蜗的语音增强中。根据CI电极对应的频带关系,该算法先对采集的带噪声音信号功率谱进行Bark子带划分,并在每个Bark子带中根据信噪比的变化进行谱减阶数和系数的自适应调节,使各子带噪声更均衡地去除,基本消除了传统方法存在的“音乐噪声”。基于该算法的电子耳蜗ACE仿真实验及测听结果表明,与传统谱减法相比,改进的算法能更好地抑制背景噪声和残留噪声,仿真得到的CI合成音感知更好和更清晰。

An adaptive spectral subtraction algorithm for speech enhancement in cochlear implant

On account of traditional spectral subtraction speech enhancement method usually exists scrap 'musical noise' phenomenon,the auditory perception speech quality from CI users will be affected if the traditional spectral subtraction algorithm is used in Cochlear Implant(CI)speech enhancement.To improve the noise immunity of CI,this paper introduces an adaptive order spectral subtraction algorithm and applies it to CI speech enhancement.According to the relationship between CI electrodes and frequency bands,in this algorithm the noisy speech power spectrum is divided into Bark subbands,and the coefficients and orders of the spectral subtraction in each subband are calculated according to the S/N rates.As a result,the improved method suppresses noise more balanced in each sub-bands and ultimately eliminates the 'musical noise' produced by traditional spectral subtraction.The ACE synthesis simulation based on the proposed method and subjective listening test results show that,compared with traditional spectral subtraction,the proposed method can suppress the background noise and the residual noise more effectively,and the CI synthesized speech is apperceived much more comfortable and clearer.