Semi-automatic mono-to-stereo upmixing via separation of note events

Abstract

A novel method for upmixing mono recordings into stereo is presented. The proposed approach uses a source separation strategy to extract a number of note events from within the original mono mixture, which are clustered into individual sources by exploiting a user-interactive interface. The isolated sources can then be panned in different parts of the stereo image to create a wider spatial experience in the final version. In this work, the quality of the stereo sound is evaluated by conducting a listening test, and the results are compared with a similar process based on a different source separation strategy. The proposed system is shown to be able to deliver stereo versions with higher audio quality and naturalness, suitable for music containing harmonic instruments or singing voices.