大工至善|大学至真分享 http://blog.sciencenet.cn/u/lcj2212916

博文

[转载]【信息技术】【2006】生物力学信息下的非线性语音信号处理

已有 241 次阅读 2020-6-12 18:24 |系统分类:科研笔记|文章来源:转载

本文为英国牛津大学(作者:Max A. Little)的博士论文,共165页。

 

基于线性时不变系统理论的线性数字信号处理在语音处理中有着广泛的应用。由语音产生的线性声源滤波理论为使用线性技术提供了现成的生物力学依据。尽管如此,本论文所调查的生物力学研究显示出显著的非线性和非高斯性,对语音产生的线性模型提出了质疑。因此,为了测试线性系统假设对语音产生的适用性,可以使用替代数据技术。这项研究揭示了现有替代数据技术在设计和使用方面的系统缺陷,并通过新的改进,开发了一种更可靠的技术。对迄今为止与这一新技术兼容的最大一组语音信号进行排序,本研究接下来证明线性假设并不适用于所有语音信号。详细的分析表明,虽然健康受试者的元音产生不能用线性假设来解释,但辅音可以。线性假设也不适用于大多数患有语音障碍的病理受试者的元音产生。将这一新的经验证据与生物力学研究的信息相结合,最后得出结论:在一组统一的数学假设中解释所有这些发现的最简洁的语音生成模型是一个随机非线性、非高斯模型,它包含高斯线性和确定性非线性模型。作为一个案例,为了证明基于所提出的生物力学信息统一模型的非线性信号处理技术的工程价值,本研究探讨了无序语音测量的生物医学工程应用。设计了一种新的状态空间递推测度,并将其与现有的随机信号分形尺度特性测度相结合。使用了一种简单的模式分类器,这两种方法在检测病态元音和健康元音的大型数据库中的语音异常方面优于所有线性方法的组合,从而明确了这种生物力学信息的非线性信号处理技术的有效性。

 

Linear digital signal processing basedaround linear, time-invariant systems theory finds substantial application inspeech processing. The linear acoustic source-filter theory of speechproduction provides ready biomechanical justification for using lineartechniques. Nonetheless, biomechanical studies surveyed in this thesis displaysignificant nonlinearity and non-Gaussianity, casting doubt on the linear modelof speech production. In order therefore to test the appropriateness of linearsystems assumptions for speech production, surrogate data techniques can beused. This study uncovers systematic flaws in the design and use of existingsurrogate data techniques, and, by making novel improvements, develops a morereliable technique. Collating the largest set of speech signals to-datecompatible with this new technique, this study next demonstrates that thelinear assumptions are not appropriate for all speech signals. Detailedanalysis shows that while vowel production from healthy subjects cannot beexplained within the linear assumptions, consonants can. Linear assumptionsalso fail for most vowel production by pathological subjects with voicedisorders. Combining this new empirical evidence with information frombiomechanical studies concludes that the most parsimonious model for speechproduction, explaining all these findings in one unified set of mathematicalassumptions, is a stochastic nonlinear, non-Gaussian model, which subsumes bothGaussian linear and deterministic nonlinear models. As a case study, todemonstrate the engineering value of nonlinear signal processing techniquesbased upon the proposed biomechanically-informed, unified model, the studyinvestigates the biomedical engineering application of disordered voicemeasurement. A new state space recurrence measure is devised and combined withan existing measure of the fractal scaling properties of stochastic signals.Using a simple pattern classifier these two measures outperform allcombinations of linear methods for the detection of voice disorders on a largedatabase of pathological and healthy vowels, making explicit the effectivenessof such biomechanically-informed, nonlinear signal processing techniques.

 

1. 引言

2. 生物力学与语音学概述

3. 经典的线性数字语音分析

4. 非线性时间序列分析

5. 语音信号的非线性

6. 语音非线性的临床应用

7. 讨论与结论

附录A.1 声带管模型的数值解

附录A.2 杂项证明

附录A.3 修正TDMI估计量的推导


更多精彩文章请关注公众号:205328s611i1aqxbbgxv19.jpg




http://blog.sciencenet.cn/blog-69686-1237627.html

上一篇:[转载]【信息技术】【2008】医学图像配准的优化方法
下一篇:[转载]【计算机科学】【2018】组织学图像分类的深度学习结构

0

该博文允许注册用户评论 请点击登录 评论 (0 个评论)

数据加载中...

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2020-9-27 16:14

Powered by ScienceNet.cn

Copyright © 2007- 中国科学报社

返回顶部