| （摘要内容经过系统自动伪原创处理以避免复制，下载原文正常，内容请直接查看目录。） 情绪语音辨认作为语音旌旗灯号处置范畴的一个主要的研究分支，在继续传统的语音旌旗灯号处置技巧的特色的同时，也与人类心思学、语音学、声学等多个学科互相渗入渗出、穿插而构成语音处置范畴中的一个热点的研究点。所谓情绪语音辨认就是付与盘算机必定的智能，使其可以或许准确地断定出所输出语音的情绪状况。今朝，跟着盘算机迷信技巧和通讯技巧的疾速成长，情绪语音辨认在人机灵能交互方面也有侧重要的实际意义和应用远景。本文研究的重要内容是提取基于人耳听觉特征的谱能量特点，并在此基本长进行了几种优化性改良。论文中采取的数据库为TYUT情绪语音库和EMO一DB情绪语音库，个中TYUT情绪语音库包括中文和英文两种语种，EMO一DB情绪语音库只包括德语语种。后续辨认模子采取支撑向量机。本文起首引见并剖析比拟了LPCC、MFCC和ZCPA等几种经典的特点参数，德语毕业论文，然后分离就汉语、英语和德语三种语种设计了情绪辨认试验，并将试验成果作为后续所研究特点的参照。接着研究了根本的谱能量特点：AUSEES特点和AUSEEG特点，因为AUSEES特点和AUSEEG特点采取线性均匀频带划分办法，不相符人耳听觉特征，所以本文彩用模仿人耳听觉特征的Bark标准和ERB标准两种频带划分办法对根本的AUSEES、 AUSEEG特点停止了改良，获得基于人耳听觉特征的两类四种谱能量特点：AUSEES一Bark、AUSEEG一Bark和AUSEES一ERB、AUSEEG一ERB。将改良后的谱能量特点应用到情绪语音辨认中，试验成果注解，改良后的新特点的情绪辨认率显著进步，德语毕业论文，个中采取Bark标准频带划分的谱能量特点的情绪辨认率绝对最高，对分歧语种的辨认机能最稳固性。然后，本文后续任务以AUSEES一Bark、AUSEEG一Bark两种特点为重要研究对象提出了两种改良办法。起首运用LPCC参数重要反应声道呼应的长处对AUSEEG一Bark特点停止了赔偿改良，然后运用Teager能量算子对能量在分歧频段上的搬移感化对AUSEES一Bark、AUSEEG一Bark特点停止了优化改良。试验成果注解这两种改良办法都是有用可行的，改良后的新的谱能量特点也都具有更好的情绪分类后果，其情绪辨认率都有显著进步，个中，基于Teager能量算子的谱能量特点具有绝对最为满足的情绪分类后果。 Abstract: Speech emotion recognition as speech signal processing category a major research branch, continued in the characteristics of the traditional speech signal, disposal techniques at the same time, multiple disciplines, and the human mind to learn, phonetics and acoustics, etc. each other infiltration, interspersed with a speech processing in the field of a hot research point. The so-called emotional speech identification is to give the computer must be intelligent, which may accurately determine the emotion state of the speech output. At present, along with the computer science and technology and the rapid development of communication skills, emotional speech recognition can also have important practical significance of lateral interaction and application prospect in the human spirit. The important content of this paper is a study of the extraction of spectral energy features based on human auditory characteristics, and on the basic of several optimization improvement. This paper take the database for TYUT EMO DB emotional speech database and emotional speech database, the TYUT library includes two kinds of emotional speech Chinese and English languages, EMO DB includes only the German language emotional speech database. The subsequent recognition model of support vector machine to take. This paper first introduces and analysis comparison of the characteristic parameters of LPCC, MFCC and zcpa and several classical, then separated on Chinese, English and German three languages designed emotion recognition test and will test results as the characteristics of the follow-up research reference. And then discuss the fundamental spectrum features of energy: ausees characteristics and auseeg characteristics. Because of ausees characteristics and auseeg characteristics adopt uniform linear frequency division method, does not conform to the human auditory characteristics, so the color used to imitate human auditory characteristics of bark standards and Erb standard two frequency division method to improve the fundamental ausees, auseeg characteristics, obtained based on human auditory characteristics of two kinds of four spectral energy features: ausees bark, auseeg bark and ausees Erb, auseeg a Erb. Will be improved after the spectral characteristics of the energy applied to the speech emotion recognition, test results of the annotation, the mood of the new features in the modified identify significant progress rate, medium take bark standard frequency division of energy spectrum characteristics of emotional identification rate of absolute maximum, differences in language identification function the most stable. Then, this paper puts forward two measures to improve follow-up tasks by AUSEES Bark, AUSEEG Bark two is an important characteristic of the research object. Chapeau application LPCC parameters important reaction channel echo strengths on the features of the auseeg bark stop the improvement of compensation, then apply the Teager energy operator of the energy in different frequency band of the shifting action characteristics of ausees bark, auseeg bark stopped improving and optimizing. Test results note the two improved methods are effective and feasible and improved new spectral features of energy but also have a better mood classification consequences, the emotional identification rate have significant progress, medium, based on Teager energy operator energy spectrum characteristics is absolutely the most meet the mood of the classification results. 目录: |
