结合语音增强的基音检测改进方法  

Improved pitch detection method for speech combined with speech enhancement

作　　者：徐昕[1] 张天骐[1] 石穗 张亚娟[1] 

机构地区：[1]重庆邮电大学信号与信息处理重庆市重点实验室,重庆400065

出　　处：《计算机工程与设计》2015年第3期699-704,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61371164;61071196;61102131);信号与信息处理重庆市市级重点实验室建设基金项目(CSTC2009CA2003);重庆市杰出青年基金项目(CSTC2011jjjq40002);重庆市自然科学基金项目(CSTC2012JJA40008);重庆市教育委员会科研基金项目(KJ120525;KJ130524)

摘　　要：针对低信噪比情况下,语音信号传统的基音检测方法鲁棒性较差的问题,提出一种结合语音增强的基音检测改进方法。通过基于听觉掩蔽的多频带谱减法减小带噪语音信号背景噪声,得到较干净的语音;将增强后的语音作为基音检测的待处理语音,利用能零积和能零比的多门限法对其进行清浊音判决;在平均幅度差函数(AMDF)加权自相关函数(ACF)的基础方法上进行改进,实现精确的基音检测。理论与仿真结果表明,在信噪比为-10dB时,该方法依然能够精确检测基音周期,鲁棒性明显提高。For poor robustness issues of pitch detection of noisy speech,the improved pitch detection method was proposed for noisy speech combined with speech enhancement.Firstly,to reduce background noise and receive relatively clean speech,the multi-band spectral subtraction and the masking properties of human auditory system were used to work the noisy speech,then the enhanced speech signal was treated as pending speech,and the energy and zero-crossing rate＇s product and quotient were used to adjudge the voice,finally,the improved average magnitude difference function（AMDF）weighted autocorrelation function（ACF）method was used for accurate pitch detection of the voice.Theoretical and experimental simulations results show that,the method can detect pitch accurately in low SNR,and the robustness is improved significantly.

关 键 词：语音增强 多频带谱减 自相关函数 平均幅度差函数 基音检测 

分 类 号：TN912.3[电子电信—通信与信息系统]