【作者】 朱淑琴；

【导师】 裘雪红；

【作者基本信息】 西安电子科技大学 ， 计算机应用技术， 2004， 硕士

【摘要】 论文根据语音识别的一般流程，主要针对语音识别系统的关键技术进行探讨： (1) 首先对语音信号的预处理和特征提取问题进行讨论。分析了当前最常用的两种特征参数，MFCC和LPCC，在此基础上对语音识别系统预处理和特征提取作了一些改进，并给出相应的实验验证。 (2) 研究了一系列的语音识别算法：DTW、HMM和ANN。在深入研究每种算法的基础上，提出了许多改进和优化。DTW算法主要针对减小运算量和存储空间方面进行优化；HMM模型的初始化采用非平均分段的方法提高系统性能；将神经网络用于语音识别时做了一些修正改进。通过实验验证，这些方法效果良好。 (3) 论文最后针对不同方法，对语音识别的各个环节进行仿真，实现了一个语音识别演示系统，软件系统界面友好、操作方便。更多还原

【Abstract】 According to commonly steps of speech recognition, the key methods of speech recognition is discussed:a) Firstly, preprocessing and feature extraction in speech recognition is studied .We studied two important speech analysis methods and extracted two key features for speech recognition: MFCC and LPCC .On the base of the research we improve the algorithm and experiment with new method.b) Secondly, a serial of speech recognition methods, DTW, HMM and ANN, is analysised deeply. Much improvement and optimization is put forward. DTW algorithm is optimized in memory size and operation amount; Initialization of HMM adopts inequality segment; ANN must be modified before applying to speech recognition system. These methods have a good effect by experiment.c) We simulate the different methods and every parts .A speech recognition system is realized with friendly interface and convenience operation.更多还原