Perceptual Single-Channel Audio Source Separation by Non-negative Matrix Factorization

Kirbiz S., Günsel Kalyoncu B.

IEEE 17th Signal Processing and Communications Applications Conference, Antalya, Turkey, 9 - 11 April 2009, pp.654-657 identifier

  • Publication Type: Conference Paper / Full Text
  • City: Antalya
  • Country: Turkey
  • Page Numbers: pp.654-657
  • Istanbul Technical University Affiliated: Yes


This paper proposes a single-channel audio source decomposition method that integrates perceptual quality criteria into source separation. Unlike the existing methods, the proposed method applies a perceptually weighted non-negative matrix factorization on log-frequency spectrogram of the mixed signal. The weights are adaptively calculated for each critical band based on a perceptual model described by ITU-R BS. 138 7 perceptual quality standard. It is shown that the proposed adaptive weighting scheme significantly improves the quality of audio sources estimated by minimizing the weighted divergence between the observed log-frequency spectrogram and the model.