基于改进GCC-PHAT算法的麦克风阵列声源定位技术

发布时间：2018-06-05 01:21

本文选题：时延估计 + 广义互相关　；参考：《辽宁工程技术大学》2017年硕士论文

【摘要】：随着人民生活水平的日益提高和科学技术的的进步,服务型机器人的应用也在逐渐的推广。针对比较贴近人类日常生活的服务型机器人来说,更好的实现人机交互就显得十分必要。语音识别已经在机器人上得到了应用,声源定位功能的开发也将是一个具有重要意义的研究方向。正是在这样一个背景下,本文对服务型机器人上的声源定位方法进行了研究。提出了相应的解决方法,有望提高其智能化水平。本文的主要研究内容有以下几个方面。(1)分析比较了常用的几种声源定位方法,最后选择了易于实现的基于声信号到达时间差的声源定位方法作为本文的研究主线。(2)对信号预处理阶段开展了细致的研究。基本信号模型进行了分类。针对语音段与非语音段的划分,采用了行之有效的语音端点检测方法-频带方差法,可以有效去除非语音段,减小了计算复杂度。由于一般的带通滤波只能滤除设定频带范围外的噪声,对于频带内叠加的噪声无可奈何,所以提出了基于改进谱减法的语音增强方法。(3)系统分析了几种时延估计方法,对它们的工作原理进行了比较细致的研究,并进行了仿真对比。最后选定GCC-PHAT作为本文的时延估计方法,并改进了它的时延估计性能,使它在信噪比较低的情况下时延估计的准确性以及可靠性得到提高,为下一步的位置估计提供可靠的时延值。(4)介绍了两大类定位估计方法:几何定位法和目标函数搜索法。在传统几何定位法的基础上提出了一种基于七元麦克风立体十字阵列的定位方法,快速的缩小了定位的范围,有效排除了模糊解,提高了定位的成功率。(5)对基于声信号到达时间差的声源定位方法进行理论分析以后,又进行了实验验证。通过麦克风阵列采集语音信号,经处理器对数据打包处理后通过串口送入电脑端,运用MATLAB软件对采集到的数据进行分析处理,最终得出声源的估计位置。结果表明本文所采用的方法能够定位出声源的位置,且精度较高,能够满足一定的实际需要。
[Abstract]:With the improvement of the people's living standard and the progress of science and technology, the application of the service robot is also gradually popularized. For the service robot which is close to human daily life, it is necessary to realize the human-computer interaction better. The speech recognition has been applied to the robot and the sound source positioning function is used. Development will also be an important research direction. Under such a background, this paper studies the sound source localization method on the service robot. It puts forward the corresponding solutions and is expected to improve its level of intelligence. The main contents of this paper are as follows. (1) analysis and comparison of several common methods are used. The sound source localization method is selected as the main line of this paper. (2) a detailed study of the signal preprocessing stage is carried out. The basic signal model is classified. The effective voice endpoint detection side is adopted for the division of the speech and non speech segments. The method of frequency band variance can effectively remove the non speech segment and reduce the computational complexity. Because the general bandpass filter can only filter the noise outside the range of the frequency band, the noise overlay in the frequency band is helpless, so a speech enhancement method based on the improved spectral subtraction is proposed. (3) several time delay estimation methods are analyzed systematically. The working principle is studied carefully and the simulation comparison is carried out. Finally, GCC-PHAT is selected as the time delay estimation method in this paper, and its delay estimation performance is improved, which makes it improve the accuracy and reliability of time delay estimation under low signal noise comparison, and provides reliable delay for the next position estimation. (4) (4) two kinds of location estimation methods are introduced: geometric positioning method and target function search method. Based on the traditional geometric positioning method, a positioning method based on seven yuan microphone stereoscopic array is proposed, which reduces the range of positioning quickly, effectively eliminates the fuzzy solution, and improves the success rate of positioning. (5) sound signals based on sound signals. After the theoretical analysis of the sound source location method of the arrival time difference, the experimental verification is carried out. The speech signal is collected through the microphone array. After the processing of the data, the data is sent to the computer side through the serial port through the processor. The data collected by the MATLAB software is analyzed and processed, and the estimation position of the sound source is finally obtained. The method adopted in this paper can locate the location of the sound source, and the accuracy is high, which can meet certain practical needs.
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242;TN912.3

【参考文献】