Télécom Paris

Blind sparse source separation with spatially smoothed time-frequency masking
  • Shoko Araki, (NTT Communication Science Laboratories)
  • Hiroshi Sawada, (NTT Communication Science Laboratories)
  • Ryo Mukai, (NTT Communication Science Laboratories)
  • Shoji Makino, (NTT Communication Science Laboratories)
  • Sound enhancement and sound separation
  • Microphone arrays and array signal processing
Get the paper in PDF format
Acrobat Reader (version 5 minimum) is necessary to read this document.

In this paper, we handle the blind sparse source separation problem with time-frequency masks. Some existing methods employ time-frequency binary masks to extract the signals, and therefore, the extracted signals tend to contain loud musical noise. It causes from the winner-take-all property of the binary mask. To overcome the problem, this paper proposes some non-binary time-frequency masks that allow each time-frequency component to belong to several output signals. Experimental results show that our proposed method can separate signals with little distortion.

©2006 Télécom Paris/TSI
Edition : Télécom Paris -- 2006