一种适于计算声场景分析的混叠语音基音检测方法  

A Multi-Pitch Detecting Method Suitable for CASA

作　　者：赵鹤鸣[1] 朱美虹[1] 俞一彪[1] 陈雪勤[1] 

机构地区：[1]苏州大学通信与电子工程系,江苏苏州215021

出　　处：《电子学报》2003年第1期123-126,共4页Acta Electronica Sinica

基　　金：国家自然科学基金 (No 60 1 72 0 1 6)

摘　　要：本文提出了一种在混叠语音信号中检测各自语音分量基音信息的方法 .该方法采用小波变换作为基音检测模型中的滤波处理 ,并用广义自相关运算突出基音信息 ,用增强自相关累和消除冗余信息 ,并提出了用基音概率函数来预测并跟踪不同基音的变化以提高基音检测的准确性 .本文提出的方法可应用于计算声场景分析中 .实验结果表明 ,该方法对于混叠语音的基音检测是非常有效的 .This paper puts up a method suitable for multi pitch detecting under overlapping speech signals environment.In this method,wavelet transform is used as filtering analysis part of this pitch detecting model.Besides that,generalized autocorrelation function is used to strengthen pitch information and enhanced summary autocorrelation function is used to weaken redundant information.It is the most important that a pitch probability function is given to predict and tail after each pitch tracking to improve the veracity of pitch detecting.Above mentioned method could be applied to computational auditory scene analysis.From the experiment results provided,we can infer that this method is very useful and efficient.

关 键 词：声场景分析 混叠语音 基音检测 小波变换 

分 类 号：TN912.3[电子电信—通信与信息系统]