Sixth European Conference on Speech Communication and Technology
(EUROSPEECH'99)

Budapest, Hungary
September 5-9, 1999

Co-Channel Speech Separation in the Presence of Correlated and Uncorrelated Noises

Kuan-Chieh Yen (1), Jun Huang (1), Yunxin Zhao (2)

(1) Beckman Institute and Dept. of ECE, University of Illinois, Urbana, IL, USA
(2) Department of CECS, University of Missouri, Columbia, MO, USA

The noise robustness of an adaptive decorrelation filtering (ADF)-based co-channel speech separation system is addressed. While ADF algorithm has been shown as an effective method in speech-alone scenarios, its performance deteriorates in the presence of background noise. In this work, it is shown that the performance of the ADF-based system deteriorates when the speech-to-noise ratio (SNR) worsens, especially when the the noise components in the acquired signals are highly correlated. To reduce the negative impact of background noise, an energy-constrained signal subspace (ECSS) speech enhancement method was integrated as a processing front-end. Experimental results show that with the integration of ECSS method, the separation performance of the ADF-based system against background noise of various spectral and spatial characteristics was improved in most SNR conditions. The improvement was especially significant when the SNR was low and the noises were highly correlated.


Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Yen, Kuan-Chieh / Huang, Jun / Zhao, Yunxin (1999): "Co-channel speech separation in the presence of correlated and uncorrelated noises", In EUROSPEECH'99, 2587-2590.