基于DWT的AMDF基音检测改进算法

Modified AMDF Pitch Detection Algorithm Based on DWT

平均幅度差函数(AMDF)法是一种常用的基音检测算法,但由于AMDF存在严重的均值下降趋势,使得用这种算法在提取基音的过程中易产生错误.为克服该缺点,提出了一种基于离散小波变换(DWT)的AMDF改进算法(DW-AMDF),该算法首先求取每帧浊音信号的AMDF,利用DWT提取出AMDF中的下降趋势,去除下降趋势后得到DW-AMDF用于检测基音.实验结果表明,与传统的AMDF算法及其几种典型的改进算法相比,该算法的检测精度有了明显的提高.

As a commonly used pitch detection algorithm,Average Magnitude Difference Function（AMDF）is used to estimate the pitch from the speech signal effectively. However,owing to the falling trend of the AMDF curve,it is easy to make the estimated pitch erroneous when AMDF algorithm is performed. In order to overcome the drawback,this paper presents a modified AMDF algorithm based on discrete wavelet transform（ DWT） called DW-AMDF. In this algorithm, conventional AMDF of each voiced frame is firstly calculated. Then the falling trend of AMDF is extracted by DWT. Finally DW-AMDF is obtained by removing the falling trend from AMDF and the true pitch can be detected from DW-AMDF. Lots of experiments show that the performance of the proposed DW-AMDF outperforms AMDF and other AMDF-based modified algorithms.