Method and device for decomposing, recombining and playing audio data

Abstract

A system processes mixed input data using a neural network trained to separate audio data of predetermined timbres from mixed audio data and to obtain a group of decomposed tracks comprising at least first, second, and third decomposed audio tracks representing audio signals of first, second, and third predetermined timbres, respectively. The system reads a control input representing a setting of a first volume level and of a second volume level. The system recombines at least a first selected track and a second selected track selected from the group of decomposed tracks to generate a first recombined track. The system recombines the first recombined track at the first volume level with at least a third track selected from the group of decomposed tracks, at the second volume level, to obtain a second recombined track. The system plays the audio data based on the second recombined track.