On the Informed Source Separation Approach for Interactive Remixing in Stereo
Informed source separation (ISS) has become a popular trend in the audio signal processing community over the past few years. Its purpose is to decompose a mixture signal into its constituent parts at the desired or the best possible quality level given some metadata. In this paper we present a comparison between two ISS systems and relate the ISS approach in various configurations with conventional coding of separate tracks for interactive remixing in stereo. The compared systems are Underdetermined Source Signal Recovery (USSR) and Enhanced Audio Object Separation (EAOS). The latter forms a part of MPEG’s Spatial Audio Object Coding technology. The performance is evaluated using objective difference grades computed with PEMO-Q. The results suggest that USSR performs perceptually better than EOAS and has a lower computational complexity.
This paper costs $20 for non-members, $5 for AES members and is free for E-Library subscribers.