小波域回声消除算法研究

发布时间：2018-02-28 02:59

本文关键词： 回声消除小波分析双端通话检测短时平均能量基音周期自适应滤波算法混合共轭梯度算法　出处：《江西理工大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着时代的进步,科技的发展为各大通信设备商带来机遇的同时也带来了挑战。随着通信设备种类的增多,用户对设备中语音质量要求也越来越高。一个通信设备能够拥有清晰的声音也是作为语音通信产品最基本的条件,然而由于人们的生活环境的复杂化,在通话过程中,大多数的语音信号都会受到不同程度不同类型的回声信号的干扰,影响语音通话的质量,如何有效进行通信设备的回声消除就显得尤为重要。文章通过对常见声学回声消除方法的研究,分别对自适应回声消除系统中的两个主要模块:自适应回声消除模块与双端检测模块,分别做了研究与改进。(1)双端通话检测算法的研究:首先列举了常见的双端通话检测算法,针对所列举算法的不足,本文在传统的双端通话检测算法的基础上提出了一种改进的小波域多特征值双端通话检测算法。改进算法将自适应滤波器的输入信号与参考信号进行小波分解,对分解后的小波子带信号进行小波阈值去噪处理,再对去噪后的低频小波子带进行双阈值能量判断,判断出语音信号所处的不同状态:静音态,活跃态与通话态。当判定为静音态与通话态时,自适应滤波器根据其判定的状态做出相应的处理;而活跃态的语音信号由于能量的增加较少,无法一次性判定出其所处的状态,需要对活跃态语音信号的基因周期进行比较,进而判断出活跃态的通话状态,以便自适应滤波器做出相应的处理。(2)针对自适应回声消除算法:本文首先列举了常见的自适应滤波算法,针对所列举算法的不足,提出了一种改进的自适应混合共轭梯度迭代算法,该算法将均值算法与混合共轭梯度迭代算法相结合,并应用到自适应滤波器的权值更新中。最后对提出的改进的小波域多特征值双端检测算法与改进的自适应混合共轭梯度迭代算法分别进行实验仿真。实验结果表明,改进后的算法表现良好,且与理论相符。
[Abstract]:With the progress of the times, the development of science and technology brings opportunities as well as challenges for the major communication equipment manufacturers. It is also the most basic condition for a communication device to have a clear voice. However, due to the complexity of people's living environment, in the process of communication, Most of the speech signals will be disturbed by different types of echo signals, which will affect the quality of voice calls. How to effectively carry out echo cancellation of communication equipment is particularly important. Two main modules of adaptive echo cancellation system: adaptive echo cancellation module and two-terminal detection module are studied, and the algorithms of two-terminal call detection are studied and improved respectively: firstly, the common two-terminal call detection algorithms are listed. In view of the shortcomings of the listed algorithms, Based on the traditional two-terminal call detection algorithm, an improved multi-eigenvalue two-terminal call detection algorithm in wavelet domain is proposed, which decomposes the input signal and reference signal of adaptive filter by wavelet transform. The decomposed wavelet sub-band signal is de-noised by wavelet threshold, and the low-frequency wavelet sub-band after de-noising is judged by double threshold energy, and the different states of speech signal are determined: silent state. When the active state and the call state are determined, the adaptive filter makes corresponding processing according to the state determined by the active state, and the active state speech signal can not be determined in one time because of the small increase in energy. We need to compare the gene cycle of active speech signal, and then judge the active state of speech. So that the adaptive filter to do the corresponding processing. 2) for adaptive echo cancellation algorithm: firstly, this paper lists the common adaptive filtering algorithm, aiming at the shortcomings of the listed algorithm, An improved adaptive hybrid conjugate gradient iterative algorithm is proposed, which combines the mean algorithm with the hybrid conjugate gradient iterative algorithm. Finally, the improved multi-eigenvalue two-terminal detection algorithm in wavelet domain and the improved adaptive hybrid conjugate gradient iterative algorithm are simulated, respectively. The experimental results show that, The improved algorithm performs well and agrees with the theory.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN912.3

【参考文献】