These demonstrations are related to the paper Separation of Drums From Polyphonic Music Using Non-Negative Matrix Factorization and Support Vector Machine by M. Helen and T. Virtanen, published in the proc. of 13th European Signal Processing Conference Antalaya, Turkey, 2005.

These signals demostrate the performance of drum separation system. The signals are first separated into components using Non-negative Matrix Factorization(NMF) and then these components are classified into drums and harmonic instruments using Support Vector Machine(SVM). First four signals are real music samples and "generated signals" are samples where harmonic instruments, singing and drums are manually mixed together as a polyphonic signal. "Original signal" is an input for the system. "Drums" and "Harmonic" are the sum of components classified into those classes. "Original - Drums" also represents the harmonic part but it is generated by subtracting "Drums" from the "Original signal".

Signal name
Original signal
Original minus drums
Hotel California

Like a Virgin


Money for Nothing

Generated Signal 1

Generated Signal 2

Generated Signal 3

Generated Signal 4

Contact: Tuomas Virtanen, tuomas.virtanen@tut.fi