谐波显著度的基频提取方法  

Pitch estimation based on harmonic salience

作　　者：宋黎明[1] 李明[1] 颜永红[1] 

机构地区：[1]中国科学院声学研究所中国科学院语言声学与内容理解重点实验室

出　　处：《声学学报》2015年第2期294-299,共6页Acta Acustica

基　　金：国家自然科学基金(10925419;90920302;61072124;11074275;11161140319;91120001;61271426);中国科学院战略性先导科技专项(面向感知中国的新一代信息技术研究;XDA06030100;XDA06030500);国家863计划(2012AA012503);中国科学院重点部署项目(KGZD-EW-103-2)资助

摘　　要：我们提出的谐波显著度的基频提取方法,目的是从语音信号中自动获取人声基频,该方法利用抑制因子计算出基频的谐波显著度谱,对各次谐波显著度加权求和之后进行基频轨迹跟踪确定语音的基频序列。在TIMIT掺噪数据集和音乐信息检索评测2005主旋律数据集上,谐波显著度方法的准确率分别达到了88.5%和73.3%,使倍频、半频错误相对降低了80%。实验表明,基于谐波显著度的基频提取方法增强了系统的抗噪性能以及抗倍半频错误的能力。A method based on harmonic salience is proposed for extracting the fundamental frequency from speech signal. It first calculates the harmonic salience spectrum by a inhibiting factor, and summarizes the weighted salience of every harmonic partial. Finally the pitch stream is determined by harmonic tracking. The experiment is conducted with various noised data of TIMIT database and polyphonic melody data of Music Information Retrieval Evaluation Exchange （MIREX） 2005 respectively. The result shows that the accuracy of 88.5% and 73.3% are achieved, and 80% of the half-frequency errors and multi-frequency errors are eliminated. It indicates that this method can effectively enhance the noise immunity and suppress the half-frequency errors and multi-frequency errors.

关 键 词：显著度 语音信号 轨迹跟踪 基音周期 抗噪性能 数据集 共振峰 加权求和 基音检测 宽度范围 

分 类 号：TN912.3[电子电信—通信与信息系统]