一种改进的基于对数谱估计的语音增强算法  

An improved speech enhancement algorithm based on log-spectral amplitude estimation

作　　者：孔德廷 KONG Deting(Southwest China Institute of Electronic Technology,Chengdu 610036,Sichuan,China)

机构地区：[1]中国西南电子技术研究所,四川成都610036

出　　处：《声学技术》2020年第2期208-213,共6页Technical Acoustics

摘　　要：提出了一种基于对数谱估计的改进型语音增强算法。相对于传统语音增强算法,在语音信号存在不确定的条件下,利用软判决增益因子修正技术调正带噪语音信号的对数谱幅度,抑制背景噪声。引入的改进型先验信噪比估计和语音信号先验不存在概率估计方法,能够有效地估计得出语音信号的存在概率,进而求得语音信号存在时的谱增益因子函数,联合语音信号不存在时设定的增益因子函数加权求得谱增益函数。计算机仿真表明,即使在低信噪比条件下,输入背景噪声为高斯白噪声和粉红噪声等加性白噪声时,所提算法对噪声的抑制效果非常明显,且有效地克服了传统算法中引入的“音乐噪声”和语音信号畸变。An improved speech enhancement algorithm is proposed for suppressing the background group noise in noisy speech.Compared with the traditional noise suppressor,the proposed method uses the soft-decision modification method to modify the log-spectral amplitude of noisy speech under the condition of uncertainty in speech signal presence.In this paper,an improved estimator for the priori signal-to-noise ratio and an effective estimator for the priori probability of speech absence are proposed,by which the conditional probability of speech presence can be obtained.Simulation results confirm the superiority of this method in noise suppression and speech enhancement under the background of Gaussian white noise and pink noise.

关 键 词：短时傅里叶变换 语音增强 幅度谱估计 先验信噪比 先验语音不存在概率 

分 类 号：TN912.35[电子电信—通信与信息系统]