摘要
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。
Voice - identification is a kind of technology that is using computer to transfer the voice signal to an associated text or command by identification and understand. Zero - crossing rate and short - term energy are selected as the basis in the endpoint detecting in preprocessing of speech signal through the studying of the characters of single - word. LPC （ Linear Prediction Coefficient） is extracted from the signal, then cepstral coefficient is obtained as the speech characteristic parameters. DTW （ Dynamic Time Warping） algorithm , which is improved to reduce the amount of data in the matching process by using global constraint, is used for the matching of the model. Based on the algorithm, a speaker- depended isolated -word speech recognition system is designed, and the simulation and analysis is carried on by using Matlab in a variety of background conditions. The experiment shows that the algorithm for speaker - depended isolated- word speech recognition can achieve good results.
出处
《控制工程》
CSCD
北大核心
2011年第3期397-400,404,共5页
Control Engineering of China
基金
国家自然科学基金资助项目(60443008)
