基于DCT变换的变参数ZA-LMS算法的语音增强研究

发布时间：2018-02-08 13:46

本文关键词： 油田噪声语音增强自适应滤波离散余弦变换稀疏信号　出处：《北京工业大学》2016年硕士论文　论文类型：学位论文

【摘要】：现在,油田井下作业时常会遇到很多危险,及时有效的通讯措施则对安全保障起着重要的作用,但是目前油田工作环境中的语音通信质量大都达不到要求,是由于油田工作环境非常复杂,且大功率设备较多,包括压缩机、水泵、钻井机、风机等等,在这些设备运行时则会产生分贝数极高的强噪音,从而产生噪声污染,最终将会严重影响工人之间的通讯质量。目前,消除噪声的主要方法就是语音增强技术,语音增强技术就是要将尽可能纯净的语音信号从带噪语音中提取出来,提高语音的可懂度,要完成此工作则必须要依靠高质量的语音增强算法。先前的学者们已经对语音增强技术做了大量的研究,并提出了大量的算法,但是在油田环境中的语音增强效果并不是很好,因为在此环境中的噪声类型多,强度高,且与人的语音频率近乎在一个范围里,所以如何提高在特定环境中的语音质量是一个难题,本文根据此问题做了如下工作:首先,本文针对井下噪声特点,展开了对语音降噪的算法研究,根据噪声特性,选择了自适应滤波算法,该算法无需得到输入信号的先验知识,理论上只需要收集噪声以及带噪语音的信号,便可很好的完成降噪工作。经典的LMS(Least Mean Square)自适应算法虽然优点很多,但是还存在着收敛速度与稳态误差相矛盾的缺点。然后,本文对自适应算法的缺点进行了研究,提出了改进的可变零吸引自适应滤波算法。通过仿真实验的验证,本文的算法在收敛速度提升的同时,又保证了稳态误差没有增大。然后又根据油田井下噪声信号稀疏性不高的特点,提出了基于离散余弦变换的可变零吸引自适应算法,离散余弦变换可使信号能量集中,并且降低信号的自相关矩阵的相关性,将信号通过离散余弦变换基可使信号稀疏性增大,从而提高收敛速度和稳态误差。之后,本文经过了大量的实验仿真,并且通过分析四种语音质量评价指标的结果,包括时域图、信噪比、分段信噪比和PESQ评价方法,与其他同类算法相比,本文的算法在各性能参数上获得了较大的提高,语音增强效果明显。最后,本文在实验室中模拟了实际油田噪声环境,开发了语音降噪软件,将本文算法转化为C语言,并在实际场景下进行了实验测试,测试结果显示本文的算法在实际应用中也有较好的效果。
[Abstract]:At present, oil field downhole operations often encounter a lot of risks, and timely and effective communication measures play an important role in safety and security. However, at present, the quality of voice communication in the working environment of oil fields mostly fails to meet the requirements. It is because the working environment in the oil fields is very complex and there are more high-power equipment, including compressors, pumps, drilling rigs, fans and so on. When these equipments are in operation, they will produce very high decibels of strong noise, which will result in noise pollution. Eventually, it will seriously affect the quality of communication between workers. At present, the main way to eliminate noise is speech enhancement, which is to extract the purest speech signal from noisy speech as far as possible. In order to improve speech intelligibility, we must rely on high quality speech enhancement algorithms. Previous scholars have done a lot of research on speech enhancement technology and put forward a large number of algorithms. But the effect of speech enhancement in the oil field environment is not very good, because there are many kinds of noise in this environment, the intensity is high, and the speech frequency is almost the same as the human speech frequency. Therefore, how to improve the quality of speech in a specific environment is a difficult problem. This paper has done the following work according to this problem: firstly, this paper, aiming at the characteristics of downhole noise, has carried out a research on the algorithm of speech noise reduction, according to the characteristics of noise. The adaptive filtering algorithm is chosen. The algorithm does not need to obtain prior knowledge of input signal. In theory, it only needs to collect noise and noisy speech signals. The classical LMS(Least Mean Square-adaptive algorithm has many advantages, but it still has the disadvantage of the contradiction between the convergence speed and the steady-state error. Then, the shortcomings of the adaptive algorithm are studied in this paper. An improved adaptive filtering algorithm with variable zero attraction is proposed. The simulation results show that the proposed algorithm can improve the convergence speed. The steady-state error is not increased. Then, according to the characteristics of low sparsity of downhole noise signal, an adaptive algorithm of variable zero attraction based on discrete cosine transform is proposed. The discrete cosine transform can concentrate the signal energy. And reduce the correlation of autocorrelation matrix, signal through the discrete cosine transform basis can make the signal sparsity increase, thus improve the convergence rate and steady state error. And by analyzing the results of four kinds of speech quality evaluation indexes, including time domain map, signal-to-noise ratio, segmental signal-to-noise ratio and PESQ evaluation method, compared with other similar algorithms, the performance parameters of this algorithm have been greatly improved. The speech enhancement effect is obvious. Finally, the paper simulates the actual oilfield noise environment in the laboratory, develops the speech noise reduction software, transforms the algorithm into C language, and carries on the experiment test in the actual scene. The test results show that the algorithm in this paper also has a good effect in practical application.
【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN912.35

【相似文献】