Abstract
In order to capture and reproduce sound all around the listener in a perceptually realistic way, a recording technique (microphone layout) and a reproduction system (loudspeaker layout) capable of accurate 360° imaging are required. In order to develop the reproduction system, binaural and loudspeaker-based localisation cues were investigated in a literature review. The necessary loudspeaker-based localisation cues to manipulate the location of a sound image were analysed, and these showed that the common 5.1 surround sound loudspeaker arrangement could produce stable images to the front of the listener but not to the side. An 8-channel loudspeaker configuration arranged as a regular octagon was therefore chosen based on attempting to recreate plausible interaural cues for all locations. The performance of this was compared to a 5.1 loudspeaker system in terms of localisation accuracy. It was shown that the octagonal loudspeaker configuration produces more stable and more accurate images all around the listener. The octagonal loudspeaker arrangement required new microphone techniques for optimum recording and reproduction. An overview of the current literature showed how localisation curves can be used to design microphone arrays, and the other physical and perceptual parameters that need to be considered in microphone array design. This showed that it is useful to derive localisation curves for the new loudspeaker configuration, as these can be used to optimise the microphone array's sound imaging accuracy. An experiment was therefore conducted to measure the localisation curves, and it showed that localisation curves are different for the loudspeaker pairs to the front, side or the rear of the listener. It also showed that to the side of the listener, interchannel time differences have little influence on the perceived location of the sound event. The derivation of these localisation curves enabled the development of a microphone array design tool that predicts the localisation profile of a microphone array as a function of its geometrical set-up and the directivities of the microphones. The predictions of this tool were evaluated by comparing sound image positions for recordings of ten sound sources with ten different microphone arrays. This showed that the error of prediction is small when the crosstalk due to the microphone array is low. A linear regression model was created and shown to reasonably model the error of prediction as a function of the crosstalk level and crosstalk delay.