情感说话人识别技术的研究

 2022-02-11 06:02

论文总字数:31373字

摘 要

说话人识别是一种利用语音中提取出的特征来鉴别说话人身份的技术,在指纹识别等类似的识别技术快速发展的今天,说话人识别技术也有着广阔的应用与发展前景。而传统的说话人识别技术中,往往由于在模型训练和语音测试时都仅考虑了说话人提供的语音为中性语音这一种情形,而实际应用的情况中,说话人提供的语音常常可能因为外界的影响而提供情感语音,从而使得传统的说话人识别系统性能下降,严重影响使用。

本文在研究了现阶段说话人识别中常用的技术后,对学术界目前存在争议的“情感”的定义和分类作了简要探讨,并选择了一个比较适合本文研究方法的定义,以此为基础进一步探究语音中的情感因素对语音的基频、频谱、模式识别等的影响,说明了情感对说话人识别系统具体造成的影响。

最后,基于情感屏蔽的思路选择使用一种情感屏蔽的方法来对说话人识别系统进行优化,屏蔽测试语料中尽可能多的情感部分从而达到提升系统准确率的目的,并对得分选择方法中阈值大小的选取进行了必要的探讨。

关键词:说话人识别,情感语音,情感模型,模型识别,语音库,情感屏蔽

Research on Emotional Speaker Recognition Technology

Abstract

Speaker recognition technology is one kind of identification technology which makes use of the features of speech signals. With the rapid development of some biometric identification, such as fingerprint identification nowadays, speaker recognition technology has extensive application foreground. The traditional speaker recognition technology only considers using neutral utterances in training and testing speech, so the speaker being easily affected by many external factor in actual situations, many emotion states will affect speech, thus leading to the decrease of performance of the system. Such kind of recognition is called emotional speaker recognition technology.

After the study on the common technology of traditional speaker recognition technology, we further study the definition and classification of emotion, which are quite disputed in academic circles, and give the “emotion” an appropriate definition for this thesis. Based on the two jobs we have done, we further research on how emotions in speech have influence on the fundamental frequency, spectrum and pattern recognition of speech signals. Through experiment, we make the conclusion how emotion have the specific influence on speaker recognition system.

In the end, we propose a score selection method to solve this influence caused by emotions. Based on the thread of emotion shielding, this method is suitable for the situation where the testing utterance is mixed with neutral and emotional speech. Through this method, we can shield the more emotional part in the testing utterance as we can for the purpose of decreasing the emotional ratio, thus increasing the performance of emotional speaker recognition system.

Keywords:Speaker Recognition, Emotional Utterance, Emotional Pattern, Pattern Recognition, Utterance Base, Emotion Shielding

目 录

摘要 ……………………………………………………………………………………………Ⅰ

Abstract ………………………………………………………………………………………II

第一章 绪论 …………………………………………………………………………………1

1.1 引言 ………………………………………………………………………………1

1.2 情感说话人识别概述 ……………………………………………………………1

1.3 情感说话人识别常用解决思路 …………………………………………………1

1.4 本文主要研究内容 ………………………………………………………………2

第二章 说话人识别系统 ……………………………………………………………………3

2.1概述 ………………………………………………………………………………3

2.2 梅尔倒谱系数 ……………………………………………………………………3

2.3 高斯混合模型(GMM) ……………………………………………………………4

2.3.1 高斯混合模型 ……………………………………………………………4

2.3.2 GMM-UBM结构的说话人识别 …………………………………………6

2.4 说话人识别的性能评价标准 ……………………………………………………7

2.4.1 错误接受率、错误拒绝率 ………………………………………………7

2.4.2 等错误率和DET图 ……………………………………………………8

2.5 本章小结 …………………………………………………………………………8

第三章 语音信号中的情感因素 ………………………………………………………………9

3.1 情感的定义 ………………………………………………………………………9

3.2 情感的分类 ………………………………………………………………………10

3.3 情感语音数据库的建立 ………………………………………………………11

3.4 情感因素对说话人特征的影响 ………………………………………………12

3.4.1 情感因素对基音频率的影响 ………………………………………12

3.4.2 情感因素对频谱的影响 ……………………………………………12

3.4.3 情感因素对系统性能的影响 …………………………………………12

3.5 本章小结 ………………………………………………………………………13

第四章 基于得分选择方法的情感说话人识别 …………………………………………14

4.1 中性语音与情感语音的混合 ……………………………………………………14

4.2 情感比例与系统性能的关系 ……………………………………………………14

4.2.1 情感比例与说话人识别系统准确率的关系 ……………………………14

4.2.2 得分计算的分布规律 …………………………………………………15

4.2.3 实验总结 ………………………………………………………………16

4.3 得分选择方法的介绍 …………………………………………………………16

4.3.1 设计得分选择步骤 ……………………………………………………………16

4.3.2 阈值对性能提升的影响 ………………………………………………………18

4.4 本章小结 ………………………………………………………………………19

第五章 总结与展望 ………………………………………………………………………20

5.1 总结 ………………………………………………………………………………20

5.2 展望 ………………………………………………………………………………20

参考文献(References) ……………………………………………………………………21

第一章 绪 论

剩余内容已隐藏,请支付后下载全文,论文总字数:31373字

您需要先支付 80元 才能查看全部内容!立即支付

该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找;