Abstract

This thesis aims at constructing a system, being able to separate the signals of simultaneously speaking persons in a room. The desired source signal is supposed to be extracted from the actual mix of appearing sources. It shall be recovered from any noise or interfering sources, such as other speakers.

The approach is based on a microphone array whereas the received data are processed in two successive stages. Firstly, a beamforming network, consisting of several fixed beamformers steering in different directions, scans the room. The second stage, a Blind Source Separation algorithm, controls the individual beams in order to separate the desired signal from any interferences as well as possible.

Furthermore, it was required to construct twenty analogue amplifiers in order to complete the available hardware setup. The final result is a fully functional system consisting of a Matlab Graphical User Interface utilizing the associated hardware. It enables the user to listen separately to the individual speakers in a room.

Full Text & Additional Material

The full thesis can be downloaded here.