一种基于动态平滑的实时基频提取算法

A dynamic smoothing based real-time pitch detection algorithm

基频是语音信号处理中的一个基本声学特征。传统的基频提取算法为了获得较好的检测效果,需要复杂的时频域计算。对于资源受限的应用条件,例如人工耳蜗等嵌入式实时系统,很难应用计算量大的基频提取算法。语音信号的基频具有短时平稳性,根据这个特点来确定基频候选值可以提高提取的准确性。据此,提出一种基于动态平滑的基频提取算法,使用此算法对汉语声调词库进行基频提取,并与另外两种基频提取算法进行比较。实验结果表明,新算法的基频绝对平均估计误差小于3Hz,优于另两种算法,能够准确地提取基频,同时算法计算量低,适合实时应用。

Fundamental frequency is one of the most important features in speech signal processing.Traditional pitch detection algorithms（PDA） can hardly be applied in the resource-limited hardware system due to the computation complexity.A dynamic smoothing based pitch detection algorithm is proposed in this paper.As pitch is a physical quantity that does not change rapidly,using continuous speech frames to decide the best candidate of fundamental frequency can improve the accuracy.An objective experiment was carried out to compare the pitch detection accuracy of the DSPDA with two other algorithms.The experimental results show that the averaged pitch detection error is 3Hz lower than that of other algorithms.