盲源分离算法及应用研究-计算机应用技术专业毕业论文.docxVIP

摘 摘 要 I万方数据 I 万方数据 摘 要 在现代信号处理中，很多情况下，源信号以及传递通道信息都是未知的，怎样由 传感器接收的混合信号来估计出各个源信号，成为实际工作中信号处理研究与应用面 临的一大困难。 盲源分离是近二十年来信号处理领域中发展起来的一种前沿技术，它正是分离上 述混合信号以得到各个源信号的新方法。盲源分离在语音信号处理、阵列信号处理和 说话人识别等众多领域都有广泛和极具价值的应用前景。 在以上背景下，本文首先利用时域的卷积混合对应于频域复数信号的瞬时混合这 一思想，对瞬时混合信号的时域盲分离算法进行扩展，得到了一种卷积混合信号的频 域盲分离算法。通过实验对时域和频域算法的性能作了对比，在此基础上提出了一种 时-频域综合盲源分离算法。与频域算法相比，时-频域综合算法输出的信号-干扰比提 高了 5.5dB，表明了时-频域综合算法的有效性。 其次，对固定波束形成技术进行改进，得到一种针对语音信号的自适应宽带波束 形成技术，提出了一种基于麦克风阵的盲源分离算法。该算法既能有效减少麦克风阵 的阵元数，也能提高盲源分离算法的稳健性。真实环境下，对噪声的抑制达到 22dB。 最后，是广西区研究生教育创新计划项目的内容。利用语音的独立分量基函数在 说话人识别中的良好分类性能，使用独立分量分析提取说话人特征，提出了一个基于 盲源分离的说话人识别系统。实验结果表明，在噪声环境下此系统具有较高的识别率。 关键词：盲源分离，独立分量分析，麦克风阵列，自适应波束形成，说话人识别 Abst Abstract II万方数据 II 万方数据 Abstract In modern signal processing, the source signals and the information of the transimission channels are unkown in many cases.Then how to recover the source signals is becoming a difficult problem in practical work. Blind Source Separation (BSS) is a kind of signal processing technique developed in the latest 20 years, which is a new method to separate the observed mixtures so as to estimate the independent components.There are a lot of protential and very valued applications of blind source separation in many areas, especially in speech signal processing, array signal processing and speaker recognition and so on. Based on this background, this paper firstly extends blind source separation algorithm of instantaneous mixing case to that of convolutive mixing case and obtains a kind of frequency domain algorithm for convolutive mixtures. Experiments are conducted to compare the performance between time domain and frequency domain algorithms. A hybrid blind source separation algorithm based on time-frequency domain is proposed which has manages to achieve a SNR improvement of more than 5.5dB. Secondly, the Delay and Sum Beamformer is improved in order to get an adaptive wideband beamforming technique for speech signal.Then we proposed a blind source separation algorithm based on adaptive beamforming