DReaM: A Novel System for Joint Source Separation and Multi-Track Coding

Active listening consists in interacting with the music playing, has numerous applications from pedagogy to gaming, and involves advanced remixing processes such as generalized karaoke or respatialization. To get this new freedom, one might use the individual tracks that compose the mix. While multi-track formats loose backward compatibility with popular stereo formats and increase the file size, classic source separation from the stereo mix is not of sufficient quality. We propose a coder/decoder scheme for informed source separation. The coder determines the information necessary to recover the tracks and embeds it inaudibly in the mix, which is stereo and has a size comparable to the original. The decoder enhances the source separation with this information, enabling active listening.