基于Hilbert—Huang算法的缺时检测问题的研究

随着普通话水平考试的普及，普通话水平测试系统的自动化变得日益迫切。为了建立自动的普通话水平测试系统，如何快速准确的对普通话测试命题说话内容进行缺时检测是一个关键问题。将语音信号分成若干窗口，采用EMD分解算法将窗口信号分解为若干IMF。在每一窗口上利用Hilbert-Huang变换获取每个IMF分量的瞬时幅值和频率。利用瞬时幅值和频率构造每个IMF分量的能频值。将所有IMF分量的能频值组成该段的能频值特征向量，利用该能频值特征向量来区分信号中的静音段和语音段。最后通过计算相邻静音段的最大时长判断是否缺时，并采用一种规则消除因强噪声信号而引起的某一段能频值过大的影响。实验表明，笔者所设计的算法对普通话水平考试中的缺时检测有良好的效果。

Detecting of lacks of speak based on Hilbert-Huang algorithm

With the popularity of mandarin test, automation of mandarin test system becomes increasingly urgent. How to quickly and accurately detect lacks of speak part in mandarin exam is a key issue in order to establish an automatic mandarin test system. Firstly, the whole speech signal will be divided into several small sections which called windows in this paper. Then divided the window signals into a number of IMF by using EMD algorithm. In each window, using Hilbert-Huang transform on each IMF component gets the instantaneous amplitude and frequency. Take advantage of instantaneous amplitude and frequency to structure each IMF component values. Using every IMF component of the frequency values make up the feature vector. And then distinguish between signal mute and voice based on the feature vector. Finally calculate the maximum length of time to determine the adjacent mute when the deficiency, and adopt a rule to eliminate too large frequency in certain section due to strong noise signal. Experiments showed that the algorithm in the design of this article has good effect when testing in National Proflciency Test of Putonghua with good results.