摘要描述语音信号激励源性质的特征参数非常多，基频就是其中最重要的特征参数之一，其应用十分广泛，在语音信号处理领域的地位不可撼动。高性能的语音信号处理系统离不开准确可靠的基频检测。42899

本文针对基频检测的现状，指出了目前基频检测存在的主要困难，将传统的基频检测算法分为时域、频域以及时频域混合三大类，从中选取了具有代表性的三种算法：自相关法（ACF）、倒谱法（CEP）和小波变换法（WT），分析了它们的优缺点，完成了这三种算法的Matlab仿真，获得了语音信号的基频轨迹。然后将信号分解中的经验小波变换法（EWT）应用于基频的解算，实现了语音信号的模态分解以及Hilbert变换，提取了语音信号的瞬时频率。并将EWT算法与传统基频检测算法相比较，得出了EWT算法的优越性。

关键词  基频检测  自相关法  倒谱法  小波变换法  经验小波变换

毕业论文设计说明书外文摘要

Title   Research on Pitch Detection Algorithm of Chinese  Speech Signal Based on EWT                                                

Abstract

Pitch is one of the important characteristic parameters describing the properties of speech excitation source, broadly applied to many areas. In the field of speech signal processing, pitch is in an unquestionable place. The accurate and reliable pitch detection is crucial for high performance speech signal processing system.

The major difficulties in pitch detection in terms of current situation are pointed out in this work. The traditional pitch detection algorithm is classified into time domain, frequency domain and mixed time-frequency domain, from which three representative algorithms, ACF, CEP and WT are selected. Through the analysis of the advantages and disadvantages of the algorithms and their simulation in MATLAB, the trajectory of speech signal pitch is obtained. Then instantaneous frequency of speech signal is extracted by the application of EWT of signal decomposition to pitch calculation and the modal decomposition and Hilbert transform into speech signal. The advantages of EWT algorithm has emerged in the conclusion compared to traditional pitch detection algorithm.

Keywords  pitch detection  auto correlation  cepstrum  wavelet transform  empirical wavelet transform

目   次

1  绪论 1

1.1  基频检测的定义 1

1.2  研究意义 2

1.3  国内外研究现状 2

1.4  基频检测的困难 3

2  语音信号的预处理 5

3  传统基频检测算法 8

3.1  自相关法 8

3.2  倒谱法 9

3.3  小波变换法 11

4  基于EWT的基频检测算法 15

4.1  经验小波变换法 15

4.2  Hilbert变换 19

4.3  Matlab仿真结果 21

5  算法比较 25

5.1  仿真结果对比 25

5.2  传统算法的局限性 29

5.3  EWT算法的优越性