基于软语音存在概率的噪声互功率谱估计  

Noise Cross Power Spectral Density Estimation Based on Soft Speech Presence Probability

作　　者：罗瀛 曾庆宁[1] 龙超[1] LUO Ying;ZENG Qing-ning;LONG Chao(Key Laboratory of Cognitive Radio and Information Processing of Ministry of Education,Guilin University of Electronic Technology,Guilin 541004,China)

机构地区：[1]桂林电子科技大学认知无线电与信息处理教育部重点实验室

出　　处：《科学技术与工程》2019年第23期126-130,共5页Science Technology and Engineering

基　　金：国家自然科学基金(61461011);广西自然科学重点基金(2016GXNSFDA380018);桂林电子科技大学研究生科研创新项目(2017YJCX16,2017YJCX20)资助

摘　　要：针对目前已有的改进相干滤波语音增强系统中噪声互功率谱估计方法运算效率低、准确性不足的问题,提出一种基于软语音存在概率的噪声互功率谱估计方法。通过计算语音信号的固定先验软语音存在概率代替语音活动检测器,得到噪声互功率谱的无偏估计值,以改善估计的准确性,同时还可避免复杂的偏差补偿值计算,使算法计算量得以减小。仿真实验结果表明,所提出的噪声互功率谱估计方法在应用于改进相干滤波语音增强系统时有更好的感知语音质量评价得分,且运算用时更短。According to solve the problem of slow operation efficiency and inaccuracy of existing noise cross-power spectrum estimation methods in the improved coherent filtering speech enhancement system,a noise cross-power spectrum estimation method based on the soft speech presence probability is proposed.By calculating the soft speech presence probability with fixed priors instead of voice activity detector to obtain the unbiased estimation of the noise cross power spectrum to improve the accuracy of the estimation,and also avoid the calculation of the complicated bias compensation,so that the algorithm calculates can be reduced.The experimental results show that the proposed noise cross-power spectrum estimation method has better perceptual evaluation of speech quality score and shorter computation time when applied to the improved coherent filtering speech enhancement system.

关 键 词：相干滤波 无偏估计 软语音存在概率 噪声互功率谱密度 

分 类 号：TN912.35[电子电信—通信与信息系统]