Monaural Sound Source Separation by Non-Negative Matrix Factorization with Temporal Continuity and Sparseness Criteria - Demonstrations

The following audio examples demonstrate the test signals used in the evaluation of unsupervised one-channel sound source separation algorithms. The mixture signals are generated by mixing samples of pitched musical instruments and drums. The separated signals are generated by separating each mixture signal into component using the proposed methods.

mixture signal separated component 1 separated component 2 separated component 3 separated component 4 separated component 5 separated component 6 separated component 7 separated component 8 separated component 9 separated component 10
orig4.wav separated4-1.wav separated4-2.wav separated4-3.wav separated4-4.wav separated4-5.wav separated4-6.wav separated4-7.wav
orig6.wav separated6-1.wav separated6-2.wav separated6-3.wav separated6-4.wav separated6-5.wav separated6-6.wav separated6-7.wav separated6-8.wav separated6-9.wav separated6-10.wav
orig17.wav separated17-1.wav separated17-2.wav separated17-3.wav separated17-4.wav separated17-5.wav separated17-6.wav separated17-7.wav separated17-8.wav separated17-9.wav
orig25.wav separated25-1.wav separated25-2.wav separated25-3.wav separated25-4.wav separated25-5.wav separated25-6.wav separated25-7.wav

Demonstrations main page

- Tuomas Virtanen,