基于改进增益函数的电子耳蜗语音增强

Speech Enhancement for Cochlear Implant Based on Improved Gain Function

目前在安静环境下电子耳蜗编码技术已取得较高的语音识别率,但在噪声条件下听觉感知性能下降明显。针对该问题,提出基于改进增益函数的电子耳蜗语音增强算法。以组合编码算法为基础,采用约束方差的噪声谱估计算法进行噪声功率谱估计并应用于信噪比估计。结合人耳掩蔽阈值在子频带中自适应调节增益函数,将改进的增益函数与通道选择相结合,实现电子耳蜗语音增强。实验结果表明,与采用基本谱减法前端去噪和传统增益函数的电子耳蜗语音增强算法相比,该算法的语音平均识别率分别提高了53%和22%,在保留更多语音信息的同时能有效消除背景噪声干扰。

Currently,the Cochlear Implant （CI） coding techniques achieve a high speech recognition rate in quiet environment,but the auditory perception performance significantly decreases in noisy conditions.In order to solve this problem,this paper proposes an enhancement method in CI on the basis of improved gain function.Based on the combined coding algorithm,this paper makes use of the spectrum estimation algorithm of constrained variance noise to calculate the noise power spectrum estimation and applies it into Signal to Noise Ratio （SNR） estimation,and combines it with human ears＇ masking threshold to adaptively adjust the gain function in sub-band.The speech enhancement in CI is achieved by combining the improved gain function with the channel selection.Experimental results show that comparing with the methods of front-end de-noising spectral subtraction algorithm and the traditional gain function algorithm of the speech enhancement for CI,the proposed algorithm keeps more voice information and greatly removes the background noise.The average recognition rate of this method is respectively improved by 53％ and 22％.