Robust Representation of Spatial Sound in Stereo-to-Multichannel Upmix

This paper presents a stereo-to-multichannel upmix algorithm based on source separation method. In the conventional upmix algorithms, panning source and ambient components are decomposed or separated by adaptive algorithm, i.e. least-squares (LS) or least-mean-square (LMS). Separation performance of those algorithms is easily influenced by primary to ambient energy ratio (PAR). Since PAR is time-varying, it causes the energy fluctuation of separated sound sources. To prevent this problem, we propose a robust separation algorithm using pseudo inverse matrix. And we propose a novel post-scaling algorithm to compensate for the influence of interference with considering desired multichannel format. Performance of the proposed upmix algorithm is confirmed by subjective listening test in ITU 3/2 format.