Study on adaptive microphone array system for concurrent speech segregation

提出了两种算法，一种只使用两个传声器单元，另一种使用多个单元，提出了频域双耳模型。该算法基于双耳信息、耳间相位和电平差异来执行分离。该算法的优点是只需要两个元素就可以同时检测到方位和仰角。此外，由于计算成本低，使用笔记本电脑进行实时处理。多元素(超过2个)：该模型是为计算机用户提出的。八个麦克风元件固定在计算机显示器的框架上。将基于AMUSE方法的盲信号分离作为信号的前处理。在后续处理中，根据每个到达方向为每个声源设计自适应陷波滤波器，以便同时分离信号。特别是，前一种方法FDBM有望应用于考虑双耳信息的助听器系统，未来将讨论算法的改进和安排，以便将该算法应用于日常生活。

Two algorithms are proposed here ; one uses only two microphone elements, other uses multiple elements.Two elements :Frequency Domain Binaural Model (FDBM) is proposed. The algorithm performs segregation based on binaural information ; inter-aural phase and level differences. Advantage of the algorithm is archive to detect both azimuth and elevation for direction of arrival with only two elements. Moreover, real time processing using a laptop computer is performed due to low calculation cost.Multiple elements (over 2) :This model is proposed for computer users. Eight microphone elements are attached to frames of a computer display. Blind signal separation based on AMUSE method is performed as pre-processing. In post-processing, adaptive notch filter is designed for each sound source based on each direction of arrival in order to segregate signals simultaneously.In this research project, two microphone array algorithms are proposed for the future human interface. In particular, the former method, FDBM, is expected to apply to hearing-aid system that takes into account of binaural information.In future, improvement and arrangement of the algorithms will be discussed to use the algorithm in a daily life.