Separation of Moving Sound Sources Using Multichannel NMF and Acoustic Tracking

Separation of Moving Sound Sources Using Multichannel NMF and Acoustic Tracking

Article published in IEEE/ACM Transactions on Audio Speech and Language Processing, "Separation of Moving Sound Sources Using Multichannel NMF and Acoustic Tracking," authors: Joonas Nikunen, Aleksdr Diment and Tuomas Virtanen

Video demonstrations of algorithm output

  • Videos recorded using screen + audio capture while operating a demo GUI implemented for listening and visualizing of the algorithm output
  • Videos may be laggy because of screen capture performance issues

Example 1, operation in regular meeting room

Link to video (webM format): A person speaking while moving and stationary loudspeaker playing music.

The 10 second signal is played altogether three times:

  1. The entire mixture is played once
  2. Separated source corresponding to speech is activated in the middle
  3. Separated source corresponding to music is activated in the middle

Example 2, operation in outdoors

Link to video (webM format): A person speaking while moving outdoors, noise from the traffic.

The 10 second signal is played altogether three times:

  1. The entire mixture is played once
  2. Separated source corresponding to speech is activated in the middle
  3. All other sources than speech acticated in the middle

Page last updated on 14.6.2017

Audio Research Team / Tampere University of Technology