语谱图傅里叶变换的二字汉语词汇语音识别

摘要

A speech recognition algorithm of two-word Chinese vocabulary is proposed,which takes the spectrogram of speech signals as a processed object,and is based on binary width zoning-band projection feature fusion of the broad-band and narrow-band spectrogram images in Fourier transform domain.First,the image significance of Fourier transform domain image in the broad-band and narrow-band spectrogram and their corresponding speech characteristics are analyzed.Then,the binary width zoning-band column projection and line projection of the broad-band and narrow-band spectrogram frequency domain image are carried out respectively.The projected value is taken as the first and second feature parameter sets for speech recognition.The above two feature sets are fuzed according their features as the feature value of two-word vocabulary speech recognition.Taking the support vector machine (SVM) as a classifier to realize the speech recognition of two-word Chinese vocabulary.The experiment results show that the recognition rate of this method can reach to 96.8％ for specific persons and 98.8％ for non-specific persons.The proposed method provides a new way for vocabulary recognition.%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城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.