基于谱幅度估计的麦克风阵列语音增强算法

Microphone Array Speech Enhancement Algorithm Based on Spectral Amplitude Estimation

麦克风阵列接收到的语音信号常常被噪声信号污染,为了从观测语音信号中去除噪声信号,本文分别提出了基于谱幅度最小均方误差估计和最大后验估计的麦克风阵列语音增强算法。假设语音信号谱和噪声信号谱都服从复值高斯分布以及语音信号谱幅度和谱角度相互独立。在这些假设下,推导出纯净语音成分谱幅度的最小均方误差估计的表达式。通过遗传算法来最大化纯净语音成分谱幅度的后验概率,得到语音成分谱幅度的最大后验估计的次优解。仿真实验结果表明本文提出的两种算法在三种经典的客观评价指标下优于两种经典的算法。

The speech signal received by microphone array is often polluted by noise signal.In order to remove noise signal from ob-served speech signal,two microphone array speech enhancement algorithms based on minimum mean square error estimation and maxi-mum posterior estimation of spectral amplitude are proposed respectively in this paper.It is assumed that both speech signal spectrum and noise signal spectrum obey complex value Gaussian distribution and the amplitude and Angle of speech signal spectrum are inde-pendent of each other.Under these assumptions,the expression of minimum mean square error estimation of spectral amplitude of pure speech components is derived.Genetic algorithm is used to maximize the posterior probability of pure speech component spectrum am-plitude,and the suboptimal solution of maximum posterior estimation of speech component spectrum amplitude is obtained.The simula-tion results show that the two algorithms proposed in this paper are better than the two classical algorithms under three classical objec-tive evaluation indexes.