基于支持向量机的声源定位研究

发布时间：2018-05-11 18:10

本文选题：声源定位 + 互相关函数　；参考：《南京邮电大学》2017年硕士论文

【摘要】：随着多媒体技术的进一步发展,麦克风阵列技术在声源定位中有着日益广泛的应用。然而现今许多的基于麦克风阵列的定位技术有一些缺陷,表现为:在噪声大、有混响的环境下定位性能不佳、系统实现所需成本高、计算复杂等。为了解决上述的问题,本文提出了一种新的基于支持向量机(Support Vector Machine,SVM)的声源定位算法。SVM是一种基于统计学习理论和结构风险最小化原则的机器学习方法,其诸多参数的选择直接影响到SVM的性能。为此,本文主要做了以下工作:1.对语音信号的预处理工作。预处理工作是十分必要的,因为只有在完成对麦克风所接收到的语音信号进行对应的预处理工作后,才能在算法改进的工作基础上,继续提高声源定位算法的准确率。通过将非平稳的、宽带的语音信号转变为平稳的、窄带的语音信号,这样能够便于对后续定位算法的处理。2.现今,许多麦克风阵定位系统所使用的方法的是GCC-PHAT时延估计的定位方法,这种方法的优点是抗混响能力较强,计算量较小,但其抗噪声能力较弱。本文提出一种基于SVM的声源定位算法,通过提取鉴别互相关函数的特征,选取其合适的参数,并对SVM的核函数进行优化。在含有噪声的混响环境情况下,这种算法有着显著的优越性。3.为了进一步改进SVM的性能,提出了一种构造多个分类器进行组合的构想,构建并对改进的Adaboost与SVM组合的分类器分析。通过比较单个SVM分类器模型和基于改进的Adaboost与SVM组合分类器模型这两者之间的准确率以及性能,更加能够证明这种改进算法的实用价值。
[Abstract]:With the further development of multimedia technology, microphone array technology has been widely used in sound source location. However, there are some defects in many current positioning techniques based on microphone arrays, such as poor positioning performance in noisy and reverberated environments, high cost of system implementation and complex computation. In order to solve the above problems, this paper presents a new sound source location algorithm based on support vector machine support Vector machine. SVM is a machine learning method based on statistical learning theory and structural risk minimization principle. The choice of many parameters directly affects the performance of SVM. For this reason, this article has done the following work mainly: 1. The preprocessing of speech signal. It is very necessary to preprocess the speech signal received by the microphone. Only after the corresponding preprocessing of the speech signal received by the microphone is completed can the accuracy of the sound source location algorithm be improved on the basis of the improved algorithm. By transforming the non-stationary, wideband speech signal into a stationary, narrowband speech signal, this can facilitate the processing of the subsequent localization algorithm. 2. Nowadays, many microphone array positioning systems use the localization method of GCC-PHAT time delay estimation. The advantage of this method is that the anti-reverberation ability is strong, the computation is small, but the anti-noise ability is weak. In this paper, a sound source location algorithm based on SVM is proposed. By extracting the features of the discriminant cross-correlation function, the appropriate parameters are selected, and the kernel function of SVM is optimized. In the reverberation environment with noise, this algorithm has significant advantages. 3. In order to further improve the performance of SVM, a concept of constructing multiple classifiers for combination is proposed, and an improved combination of Adaboost and SVM is constructed and analyzed. By comparing the accuracy and performance between the single SVM classifier model and the improved Adaboost and SVM combined classifier model, the practical value of the improved algorithm can be more proved.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3;TP18

【参考文献】