These demonstrations are related to the paper "Sound Source Separation Using Sparse Coding with Temporal Continuity Objective" (Virtanen).
The following signals demonstrate the separation ability of the system. 20 second excerpts were used to analyze the components, the first 10 seconds of which available for listening. Only the most interesting separated components are selected, and some components (i.e. bass and horns in signal 2) are mixed to save space and listening time.
| original signal | separated component 1 | separated component 2 |
| signal1orig.wav | signal1sep1.wav (chord 1) | signal1sep2.wav (drums) |
| signal2orig.wav | signal1sep1.wav (bass and horns) | |
| signal3orig.wav | signal3sep1.wav (bass) | |
| signal4orig.wav | signal4sep1.wav (drums) | |
| signal5orig.wav | signal5sep1.wav (drums) | |
| signal6orig.wav | signal6sep1.wav (drums & backround) | |
| signal7orig.wav | signal7sep1.wav (drums) |
The following signals demonstrate the separated components that were used in the drum transcription system. Please note that these signals have been optimized for the transcription, not for perceptual signal quality.
| original signal | separated bass and snare drums |
| midi1orig.wav | midi1drums.wav |
| midi2orig.wav | midi2drums.wav |
Last modified: Tue Mar 18 13:37:13 EET 2003 - Tuomas Virtanen, tuomas.virtanen@tut.fi