不同表现状态下语音情感识别方法的研究

论文总字数：18308字

摘 要

在情感语言识别领域的研究方面，从现在开始有20多年的时间。在几十年的时间里，我们对人类的感官研究已经达到之前未有过的层次，与眼睛相关的视觉观察也要与皮肤相关的数字神经元的已经和我这篇所涉及到与我们的耳朵和大脑相关的对人类声音或者自然界声音或者声音音频的识别和处理，该文则强调对语音情感这一方面进行单个研究方向的研究，希望能刚接触这个方向的初学这有所帮助，语音识别无论是语音识别，还是情感识别，这意味着同样的事情，就是对感性声音信号的正确判断。而拥有感性的生命体不但包括人也包括动物和自然界其他生物，研究这一方向对结束自然界情感的诞生也有举足轻重的作用，情感中最为传统的情感语言识别是在机器的监督下进行的。在模型训练中，有一个控制使用这个模型来精确定义在情感状态测试状态的声音信号。

传统的机器学习专业术语上可以解释为情感模式识别，情感模型识别需要三个步骤来完成语言或音频情感模型识别：数据库，特征参数，识别网络。在这三个阶段，有文献包含许多研究领域的衍特征参数。特征的选择、特征的优化等。但是，文章提出了新的研究领域。以下是关于以下问题的初步审查和研究。不同数据库的结果是否会导致感情识别上的差异。

关键词：不同状态下；语音情感识别；方法

Research on speech emotion recognition in different states

Abstract

In the field of emotional language recognition, it has been more than 20 years since now. For decades. Our research on human senses has reached an unprecedented level, including visual observation related to eyes, digital neurons related to skin, and recognition and processing of human voice or natural voice or audio related to our ears and brain, This paper focuses on the study of a single research direction in the aspect of speech emotion, hoping that it will be helpful for beginners who have just come into contact with this direction. Whether speech recognition is speech recognition or emotion recognition, it means that the same thing is the correct judgment of perceptual sound signal. The sentimental life includes not only human beings, but also animals and other creatures in nature. The study of this direction also plays an important role in ending the birth of emotion in nature. The most traditional emotion language recognition in emotion is carried out under the supervision of machines. In model training, a controller uses this model to precisely define the sound signal in the emotional state test state.

The traditional machine learning terminology can be interpreted as emotional pattern recognition, emotional model recognition needs three steps to complete the language or audio emotional model recognition: database, feature parameters, recognition network. In these three stages, there are many literatures about the derivative characteristic parameters in many research fields. Feature selection, feature optimization, etc. However, new research areas are proposed. The following is a preliminary review and Study on the following issues. Whether the results of different databases will lead to differences in emotion recognition.

Keywords: in different states; Speech emotion recognition; method

目 录

摘 要

Abstract Ⅱ

第一章 引 言

1.1 语音情感识别的目的和 1

1.2 情感识别的现状 2

1.3 论文的工作安排 3

第二章语音情感数据库及情感分类

2.1 概述和数据库 4

2.2 识别网络SVM分类 6

第三章 语音情感识别特征的提取

3.1 语音情感的特征 8

3.2 提取方式 8

3.3 提取特征的流程 10

3.4 梅尔谱图的提取 11

3.5 基于卷积神经网络的特征提取 13

3.6 时域金字塔特征池化 14

第四章 实验结果和分析

4.1 实验配置和数据库 17

4.2 识别网络分析 17

4.2.1 正常状态下各情感识别准确率 19

4.3 混淆矩阵 19

4.4 损失函数和总体情感识别识别率的波形图 20

第五章 总结与展望 22

5.1 工作总结 22

5.2 未来展望 22

致 谢 23

参考文献（References） 24

第一章 引 言

剩余内容已隐藏，请支付后下载全文，论文总字数：18308字