The development and analysis of an adaptive echo cancellation microphone system
In teleconferencing, the use of a speaker-phone system has a troublesome problem associated with it — acoustic echoes. Traditionally, acoustic echoes have been reduced by a very high order adaptive transversal filter. The filter utilizes the correlation between the far-end speech and the echo received by the microphone. A sophisticated system then has to be used to implement this high order filter. The situation becomes worse when multiple microphones are used simultaneously. All these problems make the real-time application very difficult and limited. In this thesis, new approaches to these problems are proposed. A new acoustic echo cancellation microphone system is first proposed. This system uses the high correlation between two closely positioned microphone signals. In this case, a low order adaptive filter is sufficient to reduce the echo. The algorithm is then derived and the steady-state analysis of this new system shows significant improvement over the traditional system. By incorporating the unique structure of this system, a new mode detector and a double-talk adaptable acoustic echo cancellation microphone system are developed. Finally, a microphone selection and mixing system is presented which is designed to overcome the undesired effects caused by the linear superposition of multiple microphone signals. All systems, algorithms, and methods proposed in this thesis have been tested using real speech data or by real-time experiments. The experimental and simulation results reinforce that the new system and algorithms developed in this work are both practical and successful.