基于双态维纳滤波的语音增强算法研究

发布时间：2018-03-09 13:26

本文选题：语音增强　切入点：变换域　出处：《烟台大学》2017年硕士论文　论文类型：学位论文

【摘要】：现实生活和工作中,人们的学习、工作及娱乐活动多以语音的传输为桥梁,但语音在传输过程中不可避免地会受到外界噪声的干扰,从而导致最终接收到的语音信号质量与可懂度下降,严重影响人们的工作和生活品质。而要使接收到的语音信号质量和可懂度得到提高、更符合人耳的主观感受,就必须对被污染的语音进行去噪处理。语音增强即是对带噪语音信号进行处理,以尽可能地消除背景噪声、恢复出原始语音信号的过程。目前,按照语音信号处理方式的不同可以将语音增强算法分为两大类,即时域语音增强算法和变换域语音增强算法。由于变换域中语音和噪声能量更集中、两者之间的特征信息相对时域更便于处理,故在变换域中实现语音增强的算法已成为目前国内外学者研究的重点。本文研究工作主要在离散余弦变换(DCT,Discrete Cosine Transform)域内围绕基于维纳滤波的语音增强技术展开,通过分析和研究现有语音增强算法中增益因子的不足和缺陷,基于语音信号的统计模型,采用双态维纳滤波技术对其进行改进,提出了新型的语音增强算法,并从理论和实验仿真两个方面验证了提出算法的性能。本文的主要研究和创新工作具体如下:首先,对现有变换域中语音增强算法中增益因子的取值问题进行了研究,利用实际的语音及噪声数据重点分析了DCT域中纯净语音系数和噪声系数之间的符号关系,发现了传统维纳滤波语音增强技术中存在的不足;其次,基于前文的实验分析结果,采用双态维纳滤波对原有算法进行了有效改进,同时结合DCT域中常用的高斯及拉普拉斯模型分别对语音和噪声信号进行建模,设计出了三种统计模型下的双态维纳滤波语音增强算法;最后,通过理论分析和仿真实验验证了本文研究算法的有效性和优越性,同时对全文进行了总结,并对所研究语音增强技术的未来发展状况做出了展望。
[Abstract]:In real life and work, people's study, work and entertainment mostly take the voice transmission as the bridge, but in the transmission process, the voice will inevitably be interfered by the external noise. As a result, the final received speech signal quality and intelligibility decline, seriously affecting the quality of people's work and life. To improve the quality and intelligibility of the received speech signal, it is more in line with the subjective feelings of the human ear. Speech enhancement is the process of processing noisy speech signal in order to eliminate background noise as much as possible and recover the original speech signal. According to different speech signal processing methods, speech enhancement algorithms can be divided into two categories, instant speech enhancement algorithm and transform domain speech enhancement algorithm. The characteristic information between the two is easier to process than time domain. Therefore, the implementation of speech enhancement in transform domain has become the focus of scholars at home and abroad. The research work in this paper is mainly focused on Wiener filter based speech enhancement technology in discrete cosine transform (DCT / discrete Cosine transform) domain. By analyzing and studying the deficiency and defect of gain factor in existing speech enhancement algorithms, based on the statistical model of speech signal, a novel speech enhancement algorithm is proposed, which is improved by using two-state Wiener filter technology. The performance of the proposed algorithm is verified by theoretical and experimental simulation. The main research and innovation work of this paper is as follows: firstly, the gain factor in the existing speech enhancement algorithm in transform domain is studied. The symbolic relationship between pure speech coefficients and noise coefficients in DCT domain is analyzed by using actual speech and noise data, and the shortcomings of traditional Wiener filter speech enhancement techniques are found. The two-state Wiener filter is used to improve the original algorithm effectively. At the same time, combined with Gao Si and Laplace models in DCT domain, the speech and noise signals are modeled, respectively. This paper designs three kinds of two-state Wiener filter speech enhancement algorithm under three statistical models. Finally, through theoretical analysis and simulation experiments, the validity and superiority of this algorithm are verified, and the full text is summarized. The future development of the speech enhancement technology is prospected.
【学位授予单位】：烟台大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.35

【参考文献】