Convolutive BSS with multichannel NMF

This page demonstrates sound examples for convolutive blind source separation (BSS) with multichannel nonnegative matrix factorization (NMF).

Room configuration

Sound examples (16 kHz)

Bearlin, Roads, snip(85-99)

Stereo mixture bearlin-roads
Separation with IS-NMF piano 6.98 ambient+windchimes 0.75 vocals 8.36
Separation with EU-NMF piano 3.34 ambient+windchimes -2.09 vocals 5.29

Another dreamer, The ones we love, snip(69-94)

Stereo mixture another_dreamer-the_ones_we_love
Separation with IS-NMF drums 4.67 vocals 5.00 guitar 3.21
Separation with EU-NMF drums 2.05 vocals 4.06 guitar 1.92

Fort minor, Remember the name, snip(54-78)

Stereo mixture fort_minor-remember_the_name
Separation with IS-NMF drums 8.95 vocals 9.17 violins_synth 3.88
Separation with EU-NMF drums 5.68 vocals 5.94 violins_synth 0.58

Ultimate_nz_tour, snip(43-61)

Stereo mixture ultimate_nz_tour
Separation with IS-NMF drums 12.08 guitar 5.36 synth 4.90
Separation with EU-NMF drums 8.52 guitar 1.56 synth 0.88


  1. H. Sawada, H. Kameoka, S. Araki, N. Ueda, "Multichannel Extensions of Non-negative Matrix Factorization with Complex-valued Data," IEEE Trans. Audio, Speech, and Language Processing, vol.21, no.5, pp.971-982, May 2013. (Paper: PDF)