一种联合广义旁瓣抵消麦克风阵列和MMSE-LSA的语音增强系统  

A Speech Enhancement System Based on GSC Microphone Array and MMSE-LSA Algorithm

作　　者：陈先宇[1] 

机构地区：[1]重庆交通大学,重庆400074

出　　处：《科学技术与工程》2014年第19期248-252,共5页Science Technology and Engineering

摘　　要：单通道语音增强算法自20世纪60年代以来有了长足的发展,但由于时频域处理的局限性,现有的单通道语音增强算法无法有效抑制背景噪声中的突发噪声成分。突发噪声通常表现为短时、能量强、时频域有纹理特征,在参数上无法和语音进行有效区分。但背景噪声中的突发噪声,在空间上通常具有方向性。因此,提出了一种联合空间和时频域的语音增强系统。即在语音采集的前端使用GSC麦克风阵列形成波束,使主瓣对准期望语音信号、旁瓣对准突发噪声从而从空间上抑制突发噪声,然后对采集到的语音信号进行时频域语音增强处理。选取MMSE-LSA作为时频域的处理算法,因其在保留语音的可懂度、自然度方面有突出的性能。实验表明,该系统可以有效地抑制含有突发噪声的背景噪声。Single-channel speech enhancement algorithm has great development since the 1960s.However,existing single-channel speech enhancement algorithm is unable to effectively suppress background noise which contains burst noise because of the limitations of time-frequency processing.Burst noise is short in time and strong in energy,time-frequency texture which can not be effectively distinguished from speech on the parameter.But usually burst noise is spatially directional.Therefore,a speech enhancement system combined space and time-frequency domain isproposed,i.e,using the GSC microphone array to form a beam in front of the speech acquisition.Its main lobe is aligned with desired speech signal and side lobe is aligned with burst noise in order to suppress the burst noise in the space domain.Then the collected speech signal is enhanced in the time-frequency domain.MMSE-LSA is selected as the time-frequency domain processing algorithms because of its good performance in keeping intelligibility and naturalness of speech.Experiments show that the system can effectively suppress background noise which contains burst noise.

关 键 词：语音增强 最小均方误差语音增强算法(MMSE-LSA) 广义旁瓣抵消器(generalized SIDELOBE canceller GSC) 麦克风阵列 突发噪声 

分 类 号：TN912[电子电信—通信与信息系统]