一种语音增强中新的噪声预估计算法  

Novel Noise Pre-Estimation Algorithm for Speech Enhancement

作　　者：何鑫 高勇[1] HE Xin;GAO Yong(College of Electronics and Information Engineering,Sichuan University,Chengdu Sichuan 610065,China)

机构地区：[1]四川大学电子信息学院,四川成都610065

出　　处：《通信技术》2018年第10期2320-2324,共5页Communications Technology

摘　　要：语音增强算法的核心在于噪声估计的是否准确。现有噪声估计算法的噪声估计初值由前导段功率谱计算均值得来,此种方法假定语音前导段都为噪声。但实际中,输入信号前导段可能一开始就是带噪语音,此时由传统的预估计算法得到的噪声估计初值将偏大,导致语音失真。基于此,以OM-LSA算法为语音增强算法的框架,提出了一种新的噪声预估计算法。实验表明,可以使输入信号前导段为带噪语音时得到的噪声估计初值更加准确,从而保证增强后的语音音质,且可懂度更好。The core of speech enhancement algorithm lies in whether the noise estimation is accurate. The initial value of the noise estimation for the existing noise estimation algorithm is calculated by means of the power spectrum of the leading segment. This method assumes that there is an unvoiced segment in the preamble speech. In actual situations, however, the input signal may initially be noisy speech. At this time, the initial value of the noise estimation obtained by the traditional pre-estimation algorithm would be too large, leading to speech distortion. Based on this, a novel noise pre-estimation algorithm is proposed, and the OM-LSA algorithm is used as the framework of the speech enhancement algorithm. Experiments indicate that the initial value of the estimated noise can be more accurate when the input signal is preceded by noisy speech, thus ensuring the quality and better intelligibility of the enhanced speech.

关 键 词：语音增强 噪声预估计 OM—LSA 语音失真 

分 类 号：TN912.35[电子电信—通信与信息系统]