基于VQ大学生语识别算法研究.docVIP

LANZHOU UNIVERSITY OF TECHNOLOGY 毕业论文 题 目： 基于VQ的大学生语音识别算法研究 College Students Speech Recognition based on VQ Algorithm 摘 要 语音识别主要研究的是使机器能够准确的听出说话人语音内容的问题，即语音识别的最终目的是使计算机能够准确的识别任何人、任何内容的讲话。语音识别技术是一项集声学、语言学、计算机、信息处理、人工智能等领域的综合技术，在计算机、信息处理、通信与电子系统、自动控制等领域中，以及工业、军事、交通、医学、民用等诸多方面有着广泛的应用。 语音识别系统从本质上说是一种模式识别系统，其基本结构与常规模式识别系统一样，包含有特征提取、模式匹配、参考模式库等基本单元。矢量量化技术在语音识别中占有很重要的地位，其包括码书设计、码字搜索和码字索引分配，前两者尤为重要。本文主要讲了矢量量化过程中最佳码书设计算法--LBG算法的设计和实现，利用MATLAB工具进行仿真的实现。比对实验数据结果可看出LBG算法是一种下降算法，在仿真过程中，每次迭代的平均失真具有单调不增特性（或者大小至少保持不变），可用来改进训练序列的初始码书从而生成最佳码书，同时通过仿真过程还可看出LBG存在运算量和存储空间大，易得到局部最优码书的特点；最后运用仿真工具，通过改变数据量、码书维数、量化压缩比、最小失真门限等参数进行了大量仿真实验，比对仿真数据结果，对LBG算法的实现过程和参数性能特点进行了分析。 关键词:语音识别；矢量量化（VQ)；LBG算法 Abstract The main speech recognition is to accurately machine can hear the speakers voice content problem that speech recognition is the ultimate goal enables a computer to accurately identify the person, any speech content. Speech recognition technology is a set of acoustics, linguistics, computer, information processing, artificial intelligence and other areas of integrated technology, computer, information processing, communications and electronic systems, automatic control and other areas, as well as industrial, military, transportation, medical , has a wide range of civilian and many other applications. Speech recognition system?is essentially a?pattern recognition system,?the basic structure?and the conventional?pattern recognition?system,?including?feature extraction,?pattern matching reference model?library?and other basic?units.Vector Quantization in speech recognition technology plays a very important role, which includes codebook design, code word search and codeword index distribution, the first two particularly important. This article is mainly about the best course of vector quantization codebook design LBG algorithm and implementation using MATLAB simulation tool implementations. The results of the experimental data than can be seen L