基于分频带自相关函数的混叠语音基频分离提取新算法  

Pitch Extracting New Method for Mixed Speeches Based on Multi-band Autocorrelation Function

作　　者：黄秀轩[1] 季飞[1] 韦岗[1] 

机构地区：[1]华南理工大学电子与通信工程系,广州510640

出　　处：《信号处理》2004年第5期490-493,共4页Journal of Signal Processing

基　　金：国家自然科学基金资助课题(69871011;60172048)

摘　　要：混叠语音的基频分离提取问题是听觉场景分析系统的重要一环。以往的分频带自相关函数的混叠语音基频分离提取方法都是基于频带只受混叠信号之一支配的假设,而事实上,频带常常同时受两个信号影响,为此,本文提出了一种混叠语音基频分离提取新算法,算法在寻找可能的频带组时采用了闭环自适应频带选取模块,根据频带组的基频及其周期度确定两个潜在基频,提高了搜索潜在基频的鲁棒性;利用两个潜在基频重新判断频带的归属来分离信号提取基频,提高了提取基频的精度。实验结果证明新算法具有较高的有效基频提取精度。Extracting pitches from mixed speech is an important part in auditory scene analysis. The former mixed speech pitch extracting method using multi-band autocorrelation function is based on the assumption that frequency channel is dominated by one of the mixed signal, but actually, frequency channel is often dominated by both of the two signals. This paper presents a new bitch extraction method, which uses close-loop adaptive frequencies picking block to pick out the possible frequencies group, selects the two potential pitches base on the frequencies groups pitches and the relation between their periodicity so as to improve the robustness of finding potential pitches. By separating all the frequency channels according to the two potential pitches, the accuracy of the extracted pitch is improved. Experiments show that the new method has higher pitch accuracy than former method.

关 键 词：混叠语音 基频 分频 频带 听觉场景分析 自相关函数 信号 新算法 鲁棒性 搜索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN912.3[自动化与计算机技术—计算机科学与技术]