基于混合幅度差函数的基音提取算法  

Combined Magnitude Difference Function Based Pitch Tracking Algorithm

作　　者：刘建[1] 郑方[1] 邓菁[1] 吴文虎[1] 

机构地区：[1]清华大学信息科学技术国家实验室语音技术中心,北京100084

出　　处：《电子学报》2006年第10期1925-1928,共4页Acta Electronica Sinica

摘　　要：为了减少基音周期提取中的倍频和半频错误,进行更准确的基音周期估计,提出一种基于混合幅度差函数的基音周期提取方法.分析比较了不同幅度差和自相关函数估计基音周期错误率的分布,结合两类典型幅度差函数的优点定义了混合幅度差函数;进而基于混合幅度差函数,给出了使用历史信息进行校正的后处理方法.分析表明,所提方法可提高基音周期估计准确率,接近实时地确定基音周期,减少了传统基音周期估计因平滑处理而带来的误差或者动态规划处理带来的延迟.大量实验表明本文提出的基音周期提取方法比传统方法的错误率降低了13.8%.To reduce the halving and the doubling errors in pitch tracking,an algorithm based on a combined magnitude difference function is proposed in this paper. By analyzing the error distributions in different algorithms based on magnitude difference functions and autocorrelation functions, a combined magnitude difference function （CWIDF） is defined, which has the best performance compared with the others.Moreover, a post processing method is proposed which uses the history information to adjust the raw pitch estimated using the CMDF. The post processing method uses only the history information, thus the pitch tracking algorithm has low latency, which reduces the pitch estimation errors that often occur in traditional methods. Experiments show that the pitch tracking algorithm proposed in this paper can achieve a pitch error rate reduction of 13.8% compared with the baseline traditional algorithm.

关 键 词：语音信息处理 基音周期提取 混合幅度差函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]