英语语音合理性优化识别建模仿真研究

摘要

In this paper,a modeling method for optimization recognition of English pronunciation rationality is presented based on a modified particle filter.Firstly,the different people pronunciation and lip-movement information were collected and the objective function of pronunciation intonation according was obtained to mutual information criterion in information theory and high-order statistics criterion in blind source separation.Then,subspace of pronunciation intonation difference integrated with dynamic time warping theory was built.Moreover,the variation speech under stress influence was eliminated and the main part of pronunciation intonation was recognized.Finally,the length alignment was made to feature vector of pronunciation intonation and the optimization recognition model of English pronunciation rationality was built.The simulation results show that the model has higher recognition precision using the modified model to recognize the pronunciation rationality.It provides certain scientific basis for pronunciation difference teaching.%对英语语音合理性进行准确识别,可以提高英语发音的准确性.在英语发音过程中,需要对在应力影响下出现变异语音进行过滤,但是传统的识别模型进行音准识别时,主要通过发音音准特征进行识别,无法消除应力影响下的变异语音,导致建模误差大的问题.提出基于改进粒子滤波的英语语音合理性优化识别建模方法.采集不同人的英语发音和其唇动信息,依据信息论中的最大互信息准则与盲源分离中的高阶统计量准则,获取英语发音音准合理性目标函数,并融合于动态时间规正原理,构造英语发音音准差别子空间,消除应力影响下的变异语音,对英语发音音准的主体部分进行识别,并将英语发音语音特征矢量进行长度对齐,组建英语语音合理性优化识别模型.仿真结果表明,利用优化模型进行英语语音合理性识别时识别精度较高,为英美发音差异教学提供了一定的科学依据.