A reduced-rank approach to single-channel noise reduction

This paper investigates the application of the so-called reduced-rank principle to the problem of single-channel noise reduction. Under the reduced-rank framework, we develop reduced-rank Wiener and tradeoff filters. In comparison to the classical Wiener filter, the reduced-rank Wiener one can provide a degree of freedom to control the compromise between noise reduction and speech distortion. Similarly, the reduced-rank tradeoff filter offers one more degree of flexibility than the traditional tradeoff filter to optimize the noise reduction performance. Experiments are carried out and preliminary results indicate that the reduced-rank optimal filters are more interesting than their conventional counterparts in many aspects.