Digital signal processing methods and algorithms for audio conferencing systems

Today, we are interconnected almost all over the planet. Large multinational companies operate worldwide, but also an increasing number of small and medium sized companies do business overseas. As people travel to meet and do businesses, the already exposed earth is subject to even more strain. Audio conferencing is an attractive alternative to travel, which is becoming more and more appreciated. Audio conferences can of course not replace all types of meetings, but can help companies to cut travel costs and to reduce the environmental strain. The continuously growing market for audio conferencing systems proves that audio conferencing will play an important part in future communication solutions. This thesis treats digital signal processing methods and algorithms for single microphone audio conferencing systems. Concrete real problems, all in relation to audio conferencing systems, are discussed. An intrinsic problem to an audio conferencing system is the acoustic echoes picked up by the microphone. Acoustic echoes are generally cancelled using adaptive fi ltering. In such adaptive filter systems, a major difficulty is to achieve robustness in situations where both participants in a conversation are talking simultaneously. This thesis presents methods and solutions, focusing on the use of parallel adaptive fi lters, which provides the desired robustness. Audio conferencing systems are consumer electronic products and the manufacturing cost is a constant issue. Therefore, it is desirable to implement solutions on low-cost fi nite precision processors. A method to reduce fi nite precision effects in parallel filter implementations is presented in he thesis. In order to run algorithms on low-cost processors it is necessary to keep the computational complexity low. The thesis proposes a number of different methods to reduce complexity,including specific methods targeted for wideband solutions and systems equipped with extension microphones. A high quality audio conferencing system should be equipped with some sort of noise reduction feature. In the end of the thesis a method for integrating such noise reduction with the acoustic echo cancellation is presented. The performance of the proposed methods and algorithms are demonstrated through simulations as well as on real acoustic systems