A robust multichannel speech enhancement method based on decorrelation

The paper introduces a scheme that combines blind signal separation (BSS) and adaptive noise cancellation (ANC) to enhance a speech signal under the influence of noise. The main idea is to exploit the "blindness" (no array geometry or source localisation information) of BSS whilst boosting its suppression performance temporally. Initially, the scheme spatially decorrelates the received signal, i.e. BSS separates the target signal from interference using the L observed inputs. Then, the target signal is further temporally decorrelated through the ANC to remove any residual noise. It is observed that the proposed method compensates for the separation limitation of BSS to remove non-directive components. Evaluations in a real room show that the scheme offers an impressive noise suppression up to 20 dB with only a few microphones.

[1]  Lucas C. Parra,et al.  On-line Convolutive Blind Source Separation of Non-Stationary Signals , 2000, J. VLSI Signal Process..

[2]  Lucas C. Parra,et al.  The generalized sidelobe decorrelator , 2001, Proceedings of the 2001 IEEE Workshop on the Applications of Signal Processing to Audio and Acoustics (Cat. No.01TH8575).

[3]  Shoko Araki,et al.  The fundamental limitation of frequency domain blind source separation for convolutive mixtures of speech , 2003, IEEE Trans. Speech Audio Process..

[4]  Paris Smaragdis,et al.  Efficient blind separation of convolved sound mixtures , 1997, Proceedings of 1997 Workshop on Applications of Signal Processing to Audio and Acoustics.

[5]  Julie E. Greenberg,et al.  Modified LMS algorithms for speech processing with an adaptive noise canceller , 1998, IEEE Trans. Speech Audio Process..

[6]  Sven Nordholm,et al.  Convolutive blind signal separation with post-processing , 2004, IEEE Transactions on Speech and Audio Processing.

[7]  Jean-Francois Cardoso,et al.  Blind signal separation: statistical principles , 1998, Proc. IEEE.