语谱图傅里叶变换的二字汉语词汇语音识别

发布时间：2018-04-11 21:31

本文选题：傅里叶变换 + 语谱图　；参考：《现代电子技术》2017年16期

【摘要】：以语音信号的语谱图作为处理对象,提出一种基于宽窄带语谱图傅里叶变换频域图像二进宽度分带投影特征融合的二字汉语词汇语音识别算法。首先,对宽窄语谱图傅里叶变换频域图的图像意义以及相应的语音特性进行分析;然后,分别对宽窄带语谱图频域图像进行二进宽度分带列投影和行投影,将投影值作为语音识别的第一个特征参数集合和第二个特征参数集合,将以上两个特征集进行特征融合作为二字词汇语音识别的特征量,以支持向量机为分类器实现二字汉语词汇语音识别。实验结果表明,该方法对特定人二字汉语词汇语音的识别率可达96.8%,对非特定人二字汉语词汇语音的识别率可达98.8%,为解决汉语词汇整体语音识别提供了一种新的思路。
[Abstract]:Taking the speech spectrum of speech signal as the processing object, this paper presents an algorithm of two-word Chinese lexical speech recognition based on Fourier transform frequency domain image binary width banding projection feature fusion.Firstly, the image meaning and the corresponding phonetic characteristics of the Fourier transform frequency domain image are analyzed, and then, the dyadic width banding column projection and the row projection are carried out on the frequency domain image of the broad narrow band spectrum image respectively.The projection value is taken as the first feature parameter set and the second feature parameter set in speech recognition, and the above two feature sets are fused as the feature quantities of two-character vocabulary speech recognition.Support vector machine (SVM) is used to realize two-word Chinese vocabulary speech recognition.The experimental results show that the recognition rate of the new method can reach 96.8 for the Chinese vocabulary speech of a specific person and 98.8 for the non-specific Chinese vocabulary speech, which provides a new way of thinking for the whole speech recognition of Chinese vocabulary.
【作者单位】：东北师范大学物理学院;北京理工大学光电成像与信息工程研究所;
【基金】：国家自然科学基金项目(61471111)
【分类号】：TN912.34

【相似文献】