Separation of Audio Signals Into Direct and Diffuse Soundfields for Surround Sound

This paper presents a new method for reproduction of music in multichannel audio systems. The proposed method separates signals from individual channels into their direct and diffuse components which are then sent to different speaker elements. The direct components are sent directly to the listener, while the diffuse components are additionally scattered. The purpose of this scattering of diffuse components is twofold: first, it eliminates spurious localization cues which may be created by reproducing the sound using a small number of speakers (three to five), and second, it provides additional diffusion which improves the envelopment experience. We investigate two methods to separate direct and diffuse soundfield components, both of which assume the knowledge of the impulse response of the performance auditorium to the corresponding microphones. One method is based on techniques for multichannel equalization, while the other uses techniques for signal reconstruction after oversampled filter-bank processing. The latter method turns out to be computationally more manageable. In listening tests, subjects preferred music reproduction which separates direct and diffuse soundfields to reproduction in which both soundfields are sent to the same speaker elements