当前位置:文档之家› 噪声估计的算法及MATLAB实现

噪声估计的算法及MATLAB实现

太原理工大学毕业设计(论文)任务书噪声估计算法的研究及MATLAB仿真摘要日常的通信过程中,语音会常常受到环境噪声的干扰而使通话质量下降,严重时使得语音处理系统不能正常工作。

因此,必须采用信号处理方法通过语音增强来抑制背景噪声从而提高语音通信质量,而噪声估计的准确性又直接影响语音的增强效果。

可见,噪声估计是语音增强的一个非常重要的部分,所以研究噪声估计算法有很好的实用价值。

本文主要研究两种噪声估计算法:基于最小统计和最优平滑的噪声估计算法和最小值控制递归平均法的噪声估计算法,通过实验仿真比较最终研究了一种改进的最小值统计量控制递归平均噪声估计算法。

本文的主要工作总结归纳为以下几方面:首先,本文对几种经典的噪声估计算法进行研究,了解它们的各自优缺点,在此基础上选定两种较好的算法进行具体分析。

其次,了解最小统计和最优平滑和噪声功率谱统计跟踪的噪声估计算法的原理,它的基本思路是先用最优平滑滤波器对带噪语音的功率谱滤波,得到一个噪声的粗略估计,然后找出粗略估计噪声中的在一定时间窗内的最小值,对这个最小值进行一些偏差修正,即得到所要估计的噪声的方差。

通过MATLAB仿真看其特征。

再次,本文研究了一种改进的最小统计法。

算法采用递归平均进行噪声估计,其递归平均的平滑量控制递归平均噪声估计算因子受语音存在概率控制,而语音存在概率的计算采用了两次平滑和最小统计量跟踪。

与I.Cohen提出的IMCRA 算法相比,本文采用了一种快速有效的最小统计量跟踪算法。

仿真结果表明:在非平稳噪声条件下,该算法具有较好的噪声跟踪能力和较小的噪声估计误差,可以有效地提高语音增强系统的性能。

最后,对整体论文总结,通过研究发现改进的最小统计量控制递归平均噪声算法在IMCRA算法的基础上,采用了一种简单有效地最小统计量估计算法,在保证噪声估计准确性的同时,减小了算法的复杂度。

同时,基于这种噪声估计的语音增强系统能有效地提高增强语音的信噪比,并且能有效地消除增强语音中的音乐噪声。

关键词: 噪声估计,谱减法,语音检测,最小递归统计量NOISE ESTIMATION ALGORITHM RESEARCHAND MATLAB SIMULATIONABSTRACTThe Daily communication process, speech will often affected by environmental noise interference and make calls the quality descend, serious when make speech processing system didn't work properly. Therefore, must use signal processing methods through speech enhancement to curb background noise so as to improve the quality of voice communication, and the accuracy of the noise estimates directly affected speech enhancement effect. Visible, the speech enhancement noise estimation is a very important part, so the noise estimation algorithm has very good practical value. This paper makes a study of the two kinds of noise estimation algorithm based on least statistics and: the optimal smooth noise estimation algorithm and minimum control recursion average method noise estimation algorithm through experiment comparative simulation, finally puts forward an improved minimum statistic control recursion average noise estimation algorithm.This paper sums up the main work for the following aspects:First of all, the paper on the noise estimation algorithm several classic study, understand their respective advantages and disadvantages, based on selected two good concrete analysis algorithm.Second, understand the smallest statistics and optimal smoothing and noise power spectrum statistical tracking noise estimation algorithm of principle, it is to use the basic ideas of the optimal smoothing filter belt chirp voice power spectral filtering, get a noise a rough estimate, and then find out roughly in certain time window of noise to a minimum, within the minimum deviation correction, some of which is estimated to have the variance of noise. Through the simulation of MATLAB see its characteristics. noise recursively, estimates that the average recursive average smooth quantity cont Again, this paper proposes an improved minimum statistics. Algorithm for recursion average noise estimates by speech exist probability is factor control, and calculation of speech exist probability by two smooth and minimum statistic tracking. And I. Cohen proposed IMCRA algorithms, this paper adopts a kind of fast and effective minimum statistic tracking algorithm. The simulation results show that the non-stationary noise conditions, the algorithm has good noise tracking ability and smaller noise estimation error, can effectively improve the performance of the system speech enhancement.Finally, the whole thesis summed up, through the research found that improved least statistic control recursion algorithm in IMCRA average noise based on the algorithm of a simple and effective minimum statistic estimation algorithm, noise estimation accuracy in guarantee, while reducing the algorithm complexity. Meanwhile, based on this kindof noise estimates speech enhancement system can effectively improve the signal-to-noise ratio of the voice enhanced, and can effectively eliminate the music noise voice enhanced. Analytical papers deficiency and future development direction.Key words: noise estimation, the spectral subtraction, voice detection, recursive least statistic目录1 绪论 (1)1.1 噪声估计算法研究的目的和意义 (1)1.2 国内外研究的现状 (3)1.3 论文的整体安排 (5)2 几种经典的噪声估计的算法 (7)2.1 几种噪声估计算法的优点 (7)2.2 噪声估计算法 (7)3 基于语音活动性检测的噪声估计算法及MATLAB实现 (10)3.1 基于语音活动性检测的噪声估计算法 (11)3.1.1 短时能量 (11)3.1.2 短时平均过零率 (12)3.1.3 基于短时能量和短时平均过零率的语音活动性检测 (12)3.1.4 实验仿真 (15)4 最小统计递归平均的噪声估计算法及MATLAB仿真 (18)4.1 最小值统计法 (18)4.1.1 最优平滑 (18)4.1.2 最小功率谱统计跟踪 (19)4.1.3 实验仿真 (21)4.2 基于统计信息的非平稳噪声自适应算法 (23)4.2.1 概述 (23)4.2.2 非平稳噪声自适应算法 (23)4.3 最小值控制递归平均算法 (28)4.3.1 计算局部能量最小值 (29)4.3.2 估计语音存在的概率 (30)4.3.3 更新噪声谱的估计 (30)4.4 一种改进的最小统计量控制递归平均噪声估计算法 (31)4.4.1 改进的噪声估计算法 (31)4.4.2 实验仿真 (33)5 总结与展望 (35)5.1 论文的主要工作 (35)5.2 目前存在的问题及今后的发展方向 (36)参考文献 (37)致谢 (39)附录:外文文献 (40)1 绪论1.1 噪声估计算法研究的目的和意义语音作为语言的声学表现,是人类特有的也是最重要的思想和情感交流段,也是人机交互最自然的方式。

相关主题