Evaluation and alternative regularization of audio equalization techniques

Abstract

Inverse-filtering techniques are used in audio equalization of rooms and loudspeakers to correct or enhance the listening environment and to create virtual reality audio systems. The inverse-filtering concept is based on the concept that one can undo a system (loudspeaker, room, or some audio/acoustic path) completely or modify it to some desired/target response. Regularization is commonly used in the calculation of inverse filters to reduce the effort done by the inverse filter. The regularization is needed when the system is ill-conditioned and/or when zeros are present in the transfer function. Current regularization methods applied can create an audible pre-response to the corrected time-domain impulse responses, thus are not always suitable to high-quality audio systems. This thesis reviews some known FIR single and multichannel inverse filtering methods, and evaluates them objectively and subjectively, highlighting problems with relating to audible artifacts. New attempts are investigated for reducing the pre-response caused by regularization in single and multichannel inverse filtering systems, by adjusting the phase of the regularization terms to be minimum or partially minimum phase. The minimum-phase regularization method is shown to reduce the pre-response and therefore improve the audio quality of the inverse filters.