Decomposition of two-channel audio into primary sources and ambient or diffuse components has been described in the literature for applications such as stereo-to-multichannel upmix, wherein the primary and ambient components estimated from a two-channel input are used to synthesize output signals for reproduction over a multichannel loudspeaker system. In this paper, we review several approaches to two-channel primary-ambient decomposition including binary masking, soft masking, and PCA-based methods; for each two-channel approach, we consider extension to the general multichannel case. In primary-ambient decompositions, since the ambience ideally includes reverberation and diffuse noise, the estimation of the primary component can be thought of as a dereverberation and denoising process; since dereverberation and denoising are key elements in microphone array processing, we also discuss the relationships between microphone array processing and our scenario of processing multichannel music recordings for spatial enhancement.
展开▼
