本科毕业论文
题 目 语音基音周期检测算法比较 学 院 管理科学与工程学院 专 业 电子信息工程 班 级 081信工(1)班 学 号 200883082 姓 名 周刚 指导老师 段凯宇 讲师
二〇一二 年 六 月
语音基音周期检测算法比较
摘要
基音周期作为语音信号处理中描述激励源的重要参数之一，广泛的应用于语音合成、语音编码和语音识别等语音信号处理等技术领域。准确可靠的对基音周期进行检测将直接影响整个语音处理系统的性能。 常用的基音检测算法对于纯净语音信号都能达到较好的检测效果。然而，实际当中的语音信号不可避免的会受到外界背景噪音的影响，使得这些检测算法的检测效果都不是很理想，为此本文用两种基本算法对语音信号滤波前后进行基音周期检测，在进行比较。
论文首先介绍了语音基音检测算法的研究背景极其重要意义。其次对现有的基音检测算法进行了归纳和总结，并详细的介绍本文将用的两种基本基音检测算法的基本原理及实现。最后在Matlab上对语音信号进行基音周期检测。
论文还完成了算法的程序设计，在Matlab7.0仿真环境下，对上诉算法进行仿真验证，并且在滤波前后做了对比实验。实验结果表明，经过滤波的语音信号基音周期检测的更加准确。
关键词：基音检测；自相关函数法；平均幅度差函数法；基音轨迹
Abstract
Pitch as in speech signal processing is one of the important parameters to describe the excitation source, widely used in speech synthesis, speech coding and speech recognition speech signal processing technology. Accurate pitch period detection will directly affect the performance of the speech processing system.
Commonly used algorithm for pitch detection for clean speech signal to achieve good detection effect. However, the actual speech signal will be inevitably influenced by external effects of background noise, so the detection algorithm to detect the effect is not very ideal, the paper use two basic algorithms before and after filtering the speech signal pitch detection, in comparison.
The thesis first introduces the research background of speech pitch detection algorithm is very important. Next to the existing algorithm for pitch detection are summarized, and a detailed introduction to this article will use the two kinds of basic pitch detection algorithm is the basic principle and realization. Finally in Matlab on speech signal pitch period detection.
The paper also finished programming algorithm, in the Matlab7.0 simulation environment, the algorithm is validated by simulation, and the contrast experiments were done before and after filtering. The experimental results show that, after the filtering of the speech signal pitch period detection is more accurate.
Keywords: pitch detection; autocorrelation function; the average
magnitude difference function method; pitch contrail
目 录
第1章 绪论······················································ 1
1.1 引言·························································· 1 1.2 基音检测的定义和意义·········································· 1 1.3 基音检测的研究现状············································ 2 1.4 论文的结构安排················································ 3
第2章 基音检测的常用算法······································ 4
2.1 引言·························································· 4 2.1常用的基音检测算法及原理······································ 4 2.2.1 自相关函数法············································· 4 2.2.2 平均幅度差函数法········································· 6 2.3本章小结······················································· 7
第3章 基于Matlab基因周期的检测································· 9
3.1 语音文件的录制与读入·········································· 9 3.2 语音信号的采样与分帧·········································· 9 3.3 语音信号基音周期的检测······································· 11 3.3.1 短时自相关函数法检测基音周期····························· 11 3.3.2 短时平均幅度差函数法检测基音周期························· 12 3.4 基音周期检测的后处理········································· 14 3.5 滤波后的基音周期检测········································· 15 3.6 本章小结····················································· 18
第4章 总结与展望··············································· 19 参考文献························································· 20 致谢······························································ 21 附录······························································ 22
安徽财经大学管理科学与工程学院本科毕业论文
第1章 绪论
1.1引言
语音是语言的声学表现是人类相互之间进行信息交流时所用的重要载体和基本手段，随着现代信息科学技术的飞速发展，尤其是计算机网络技术的日益普及和完善，是的语音信号处理技术发挥着越来越重要的作用。目前，语音信号处理的一系列技术及其应用已经成为信息社会中不可或缺的重要组成部分，对促进信息社会的发转具有重要意义。
语音信号处理是语音学的数字信号处理相结合而形成的一门综合性学科，包括语音合成、语音识别、语音编码和说话人识别等四大分支[1]。语音信号处理和认知科学、心理学、语言学、计算机科学、模式识别和人工智能等学科联系紧密。语音信号处理技术的发展依赖于这些学科的发展，而语音信号处理技术的进步也会促进这些学科的发展[2]。
在语音信号处理中，语音信号参数提取的准确性非常重要。只有获得准确的可表征语音信号本质特征的参数，才有可能利用这些参数进行高效的语音合成，语音识别以及语音压缩编码解码处理，其中基音周期的提取在语音信号众多参数提取中尤为重要，其提取的准确与否直接影响到合成语音是否真实再现原始语音信号的频谱[3]。
1.2 基音检测的定义和意义
人在发音时，根据声带是否震动可以将语音信号分为清音跟浊音两种。浊音又称有声语言，携带者语言中大部分的能量，浊音在时域上呈现出明显的周期性；而清音类似于白噪声，没有明显的周期性。发浊音时，气流通过声门使声带产生张弛震荡式振动，产生准周期的激励脉冲串。这种声带振动的频率称为基音频率，相应的周期就成为基音周期[4]。
通常，基音频率与个人声带的长短、薄厚、韧性、劲度和发音习惯等有关系，在很大程度上反应了个人的特征。此外，基音频率还跟随着人的性别、年龄不同而有所不同。一般来说，男性说话者的基音频率较低，大部分在70～200Hz的范围内，而女性说话者和小孩的基音频率相对较高，在200～450Hz之间。
基音周期的估计称谓基音检测，基音检测的最终目的是为了找出和声带振动频率完全一致或尽可能相吻合的轨迹曲线。
1