摘要语音是人类交流的重要手段，语音信号在传达语义信息的同时还传递着情感信息，而情感在人们生活和交流中起着重要作用，因此语音情感识别中的情感特征提取就变得相当重要。

本文主要分析三种情感状态：高兴，愤怒和正常。本文提取的情感特征包括能量，能量变化率，基音频率，基音频率变化率，相对语速，第一共振峰，第一共振峰变化率。提取特征参数之前要对原始语音信号进行预处理，包括加窗，分帧，预加重。提取基音频率前要进行语音信号的清浊音判断，本文采用的是基于双门限判别法的三参数清浊音判别，这三个参数包括短时能量，短时平均过零率，短时自相关函数。本文采用自相关法计算基音频率，运用线性预测法估计共振峰频率。67257

本文的核心在于MATLAB编程提取上述参数，并结合PAAT软件进行结果分析，综合MATLAB和PRAAT分析得出同一语句，相比高兴状态下的特征参数，愤怒状态下的平均能量变化率，平均基频，平均基频变化率，第一共振峰均值以及语速相对高，而愤怒状态下的第一共振峰变化率均值相对低。

毕业论文关键词  情感特征 短时能量 第一共振峰  基音频率 MATLAB编程  PRAAT分析

毕业设计说明书（论文）外文摘要

Title  Emotional features extraction from happy 

and angry emotion states of mandarin                     

Abstract

Speech is one of the most convenient means of communication between peoples. And it conveys emotion information according with its semantic information. As emotion plays an important role in communication, emotional features extraction of speech is becoming more and more important.

This paper studies three kinds of emotions: happiness, anger and neural. And the article  selects several emotional speech features to recognize the emotion state of a sentence ，including  energy, the rate of energy change , pitch , the rate of pitch change  , the relative speed, the first formant and the rate of the first formant change. Prior to the emotional features extraction, original speech signal must be preprocessed. The process contains adding windows, sub-frame and pre-emphasis. Before extracting the pitch, the work of voicing judgments must be done. The method of voicing judgments used in this paper is three parameters criterion based on the traditional dual-threshold discrimination. The three parameters consist of short-time energy, short-time zero-crossing rate, short-time autocorrelation function. The approach of calculating the pitch in this paper is autocorrelation function. The method of estimating the first formant is linear prediction.

The core of this paper is extracting above parameters via MATLAB programming. The paper also uses PRAAT to get and analyses the data, Combining the results of PRAAT and the analysis of MATLAB, the paper comes to the following conclusion that compared with happy emotion state,some parameters of angry emotion are higher , which include  the average rate of energy change, the average pitch, the average rate of pitch change ,the average first formant and the speed ,but the average rate of the first formant change is relatively lower.

Keywords  emotion feature   short-time energy    the first formant

pitch    MATLAB programming    PRAAT analysis

目 次

1  绪论 1

1.1  选题的背景及意义 1

1.2  国内外研究现状及面临的问题 2

2 情感语音库