Resynthesis of Sound Scenes on Wave-Field Synthesis from Stereo Mixtures Using Sound Source Separation Algorithms

Spatial sound reproduction systems are capable of enhancing the perceived sound quality of reproduced scenes by using specific loudspeaker setups. Of all the spatial reproduction systems, wave-field synthesis (WFS) has been shown to be the one that comes closest to creating realistic sound fields in an extended area. Despite some practical constraints, WFS noticeably enhances stereophony in terms of the spatial reproduction of sound scenes. However, most of the recorded material is intended for stereo reproduction and, therefore, adapting this material to WFS is a challenging issue. Blind audio source separation techniques can be used for extracting the sources present in a stereo mixture, obtaining separated tracks which can be used for reconstructing spatially enhanced scenes. However, timbral distortion and intersource residuals are limiting factors that can degrade the perceived quality. The quality of several acoustic scenes (music and speech) is evaluated in three different situations: stereo image, WFS scene with original sources, and WFS scene with separated sources from stereo mixtures. The change of perceptual attributes between the different cases is discussed using listening tests.