语谱图二次傅里叶变换特定人二字汉语词汇识别  

Recognition of specific two-word Chinese vocabulary by applying Fourier transform twice to the spectrogram

作　　者：潘迪[1] 梁士利[1] 魏莹[1] 李广岩 许廷发[2] 王双维[1] 

机构地区：[1]东北师范大学物理学院,吉林长春130024 [2]北京理工大学光电成像与信息工程研究所,北京100081

出　　处：《东北师大学报（自然科学版）》2017年第2期95-100,共6页Journal of Northeast Normal University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61471111)

摘　　要：以语音信号的语谱图作为处理对象,提出了基于语谱图二次傅里叶变换对特定人二字词汇识别的方法.首先对语谱图二次傅里叶变换频域图的图像意义以及相应的语音特性表征进行了详细剖析;然后对语谱图频域图像进行二进宽度行投影,将投影值作为语音识别特征值,以支持向量机为分类器,进行特定人二字词汇语音整体识别.采用1 000个语音样本进行了仿真实验.结果表明,该方法正确识别率可达到92.4%,为汉语词汇整体识别提供了新的思路.This paper illustrates a method to recognize specific two-word Chinese vocabulary by analyzing speech signals using a spectrogram after Fourier transform is applied to it twice. First, we analyze the spectrogram in the frequency domain and its corresponding voice characteristics in detail after applying Fourier transform twice. Then, binary width zoning projection is carried out in the frequency domain. The projection value is treated as the characteristic value of semantic recognition feature and the support vector machine （SVM）is considered as the classifier for recognizing the semantics of specific two-word Chinese vocabulary. A total of 1000 voice samples were used in the simulation. The results using this method show a remarkable recognition rate of 92.4 %. The proposed method provides a new way for vocabulary recognition.

关 键 词：语谱图 二次傅里叶变换 支持向量机 二进宽度行投影 

分 类 号：TN7[电子电信—电路与系统]