基于混合高斯模型的语种辨识技术的研究

论文总字数：21550字

摘 要

语种辨识技术是利用电脑或计算机等电子设备自动识别语音信号属于何种语言的技术，语种辨识技术是由语音识别衍生而来的。随着语种识别技术的飞速进步，语种辨识的重大意义，人们对其也有了更多的重视。语种辨识技术历经了一些年的发展，实现这一技术的发展，人们已经有了很多的方法。虽然这些方法各有各的优点，但还有很多方面不成熟，有待改善。就我国来看，语种辨识技术尚处于起步阶段。语种辨识与一般的语音识别有共同点也有不通点，语种辨识需要尽量剔除语音信号中个体发音的差异，强调在与文本无关和与说话人无关的条件下进行，并且尽量提取不同语音语种之间相异的声学特征，生成特征参数，使辨识效果有根本上的提升。

本文通过利用各类语音信号的美尔倒谱系数，建立混合高斯模型的，对不同语音信号进行语种辨识。一方面研究混合高斯模型的原理和算法，研究如何建立混合高斯模型以及如何使用；另一方面提取语音信号的特征参数，对不同语种之间所含的相异的特征信息进行深入的分析，来实现对不同语种的辨识。

本文通过试验，分析混合高斯模型以及其改进模型对不同语种的识别情况进行深入分析，总结和归纳经验，为今后的努力做出一定的贡献。

关键词：语种辨识；语音信号分析；特征参数；混合高斯模型

Research on language identification technology based on mixed Gauss model

Abstract

Language identification technology is the use of computer or computer and other electronic equipment automatic recognition of speech signals which belong to the language of technology, language identification technology is derived from the speech recognition. With the rapid progress of language identification technology, the significance of language identification, people have more attention to it. Language identification technology after several years of development, to achieve the development of this technology, people have a lot of ways. Although these methods have their own advantages, but there are still many aspects are not mature, to be improved. As far as our country is, the language identification technology is still in its infancy. Language identification and speech recognition in common or not, the language identification need to try to eliminate speech signal in individual pronunciation differences, the emphasis in text independent and speaker independent conditions, and try to extract different speech language between different acoustic features, generate the feature parameters, make the identification effect is fundamentally improved.

Through using Mel cepstral coefficients for speech signals and Gaussian mixture model is set up, and the different speech signals for language identification. Hand of Gaussian mixture model principle and algorithm, studies how to set up the Gaussian mixture model and how to use; on the other hand, extracting the characteristic parameters of speech signal, the characteristics of the dissimilarity between different languages contain information for in-depth analysis, to achieve the identification of different languages.

In this paper, through the experiment, the analysis of the mixed Gauss model and its improved model for the identification of different languages in-depth analysis, summary and induction of experience, for the future efforts to make a certain contribution.

Key words: language identification; speech signal analysis; characteristic parameter; mixed Gauss model

目 录

摘要 I

Abstract II

第一章 绪论 1

第二章 语种模型和模式匹配 3

2.1 语种信源模型 3

2.2 模式分类的具体方法 3

2.3 模式匹配和模型训练 4

2.3.1 基于失真的VQ方法 4

2.3.2 离散/连续各态经历HMM 5

2.3.3 混合高斯分布模型 6

2.4 本章小结 6

第三章 语音信号分析与特征参数提取 7

3.1 概述 7

3.2 语音信号的数字化 7

3.3 预处理 8

3.3.1 窗口的形状 9

3.3.2 窗口的长度 9

3.4 端点检测 9

3.5 本章小结 10

第四章 基于混合高斯模型（GMM）语种辨识 12

4.1 混合高斯模型（GMM）方法概述 12

4.2 GMM的识别 13

4.3 高斯混合模型训练的难点 14

4.4 本章小结 14

第五章 基于EM改进的GMM的语种识别 15

5.1 EM改进模型概述 15

5.2 改进的EM算法 15

5.3 将改进的EM算法应用于GMM模型 16

5.4 本章小结 16

第六章 数据库及对比分析（附程序） 17

6.1 数据库的建立 17

6.2 编程环境 17

6.3 代码说明 17

6.3.1 特征提取 17

6.3.2 参数求取 18

6.3.3 计算 18

6.3.4 识别 19

6.2 实验结果及讨论 20

6.3 本章小结 20

致谢 21

参考文献 22

第一章 绪论

语种辨识技术（Automatic Language Recognition）长久以来就是一个吸引力和困难度共存的研究课题。

剩余内容已隐藏，请支付后下载全文，论文总字数：21550字