基于NLMS算法回波消除的研究与实现

发布时间：2018-06-25 19:14

本文选题：回波消除 + 改进NLMS　；参考：《哈尔滨工业大学》2015年硕士论文

【摘要】：大多数听力受损患者依靠佩戴数字助听器补偿听力,然而市场上各种助听器良莠不齐,体现性能差异的一个主要方面是助听器中回波的处理。回波对语音的清晰度和舒适度都有很大的影响。为了解决助听器中回波造成的问题,本文针对自适应算法在数字助听器回波消除中的应用进行研究和实现。最小均方误差(Least Mean Square,LMS)算法常常被用于回波消除,然而由于收敛速度慢,误差较大,不能满足助听器系统实时性和清晰度的要求。因此本文采用具有更高性能的归一化的最小均方误差(Normalized Least Mean Square,NLMS)算法,而且在传统的NLMS算法上进行改进,对系数更新频率、步长更新方式进行调整,更好地解决了数字助听器中回波消除问题。本文中,数字助听器回波消除模块设计包括:回波检测算法、改进的NLMS算法、回波时延估计算法以及非线性处理(Non-Linear Processing,NLP)算法,以实现语音信号处理的准确性和实时性。在算法模块入口,加入回波检测装置,当回波能量低于阈值时不对输入语音处理,装置可以避免滤波器的频繁跳转,保证处理效果的同时节省系统开销。对NLMS算法的改进包括:分批次更新滤波器系数,同时对步长迭代加入控制因子。因此,改进算法具有较小的运算量和较快的收敛速度。回波时延估计算法用来估计已处理语音信号与期望信号之间的延时。利用这一延时,两个缓冲数组能够对齐以保证输入数据的正确处理。时延估计算法对期望信号和已处理信号进行互相关,并将结果与期望信号自相关结果再次进行互相关。延时估计通过两次互相关方式对延时进行估计,具有更强抗噪能力和稳定性。NLP算法利用一个滤波器依据幅值大小对信号进行滤波,阻止低幅值信号并让高幅值信号通过。通过这种处理,可以实现回波残留的进一步消除,使语音输出具有更好的舒适度。在MATLAB上对改进的NLMS算法进行仿真,分析该算法与传统算法的性能对比。之后利用Cool EditPro软件对处理后的语音同其他算法处理语音进行分析比较。从仿真结果上看,改进算法要强于传统算法。改进算法拥有较快的收敛速度、更小的回波残留和较好的舒适度。
[Abstract]:Most hearing loss patients rely on digital hearing aids to compensate for their hearing. However, there are different kinds of hearing aids in the market. One of the main aspects of performance difference is the processing of echo in hearing aids. Echo plays an important role in articulation and comfort of speech. In order to solve the problem caused by echo in hearing aid, the application of adaptive algorithm in digital hearing aid echo cancellation is studied and implemented in this paper. Least mean square error (LMS) algorithm is often used for echo cancellation. However, due to the slow convergence rate and large error, it can not meet the requirements of real-time and clarity of hearing aid system. Therefore, the Normalized least mean Square-NLMS (NLMS) algorithm, which has higher performance, is adopted in this paper, and the traditional NLMS algorithm is improved to adjust the updating frequency and step size of the coefficients. The problem of echo cancellation in digital hearing aid is better solved. In this paper, digital hearing aid echo cancellation module design includes: echo detection algorithm, improved NLMS algorithm, echo time delay estimation algorithm and Non-Linear processing (NLP) algorithm to achieve the accuracy and real-time speech signal processing. At the entrance of the algorithm module, the echo detection device is added, when the echo energy is lower than the threshold value, the input voice is not processed correctly, so the device can avoid the frequent jump of the filter and save the system cost while ensuring the processing effect. The improvements to the NLMS algorithm include updating the filter coefficients in batches and adding control factors to step size iterations. Therefore, the improved algorithm has less computation and faster convergence speed. The echo delay estimation algorithm is used to estimate the delay between the processed speech signal and the desired signal. With this delay, the two buffering arrays can be aligned to ensure the correct processing of the input data. The time delay estimation algorithm cross-correlates the desired signal with the processed signal, and the result is cross-correlated with the autocorrelation result of the desired signal. The delay estimation can estimate the delay by two cross-correlation methods. It has stronger anti-noise ability and stability. NLP algorithm uses a filter to filter the signal according to the amplitude to prevent the low-amplitude signal and let the high-amplitude signal pass through. Through this treatment, the echo residue can be further eliminated and the speech output has better comfort. The improved NLMS algorithm is simulated on MATLAB, and its performance is compared with that of the traditional algorithm. Then, Cool EditPro software is used to analyze and compare the processed speech with other algorithms. From the simulation results, the improved algorithm is better than the traditional algorithm. The improved algorithm has faster convergence speed, smaller echo residue and better comfort.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN912.3

【参考文献】