论文部分内容阅读
语音识别是让机器自动识别和理解语音信号,并把语音信号转变为相应的文本或命令的技术。通过对特定人孤立词语音特点的研究,在对语音信号进行预处理的过程中,选择过零率与短时平均能量两项指标作为对语音信号端点检测的依据,提取语音线性预测系数,通过计算分析后获得线性预测倒谱系数,作为语音特征参数。选择动态时间规整法为模板匹配算法,并针对传统匹配算法中计算量大的特点,作出改进,采用全局限制的方法以减小匹配过程中的计算量。采用上述算法设计了一种基于特定人的孤立词语音识别系统,并对该系统进行了多种背景条件下的M atlab仿真研究。仿真实验结果表明,此算法对于特点人孤立词的语音识别能达到较好的识别效果。
Speech recognition is a technology that allows the machine to automatically recognize and understand speech signals and convert the speech signals into corresponding texts or commands. Through the research on the speech characteristics of isolated words in a particular person, two indicators of zero-crossing rate and short-time average energy are selected as the basis for the endpoint detection of speech signals in the process of preprocessing the speech signal to extract the speech linear prediction coefficient, After the calculation and analysis, the linear predictive cepstrum coefficient is obtained as a speech feature parameter. The dynamic time warping method is selected as the template matching algorithm. In order to reduce the computational complexity in the matching process, aiming at the large amount of calculation in the traditional matching algorithm, the method of global limitation is adopted. Based on the above algorithm, a speech recognition system of isolated words based on a specific person is designed and the Matlab simulation of the system under various background conditions is studied. Simulation results show that this algorithm can achieve good recognition effect for speech recognition of isolated words of characteristic people.