Fast active noise control for robust speech acquisition

Active noise control (ANC) can be a valuable resort for robust speech acquisition. Furthermore, this technique can have the added benefit of minimising the Lombard effect. In the context of broadband active noise cancellation, the adaptive algorithm most widely used is the well-known Filtered-x Least Mean Squares (FxLMS). In the present work we study an alternative to FxLMS algorithm that tries to overcome its sometimes slow convergence without loss of cancellation capability. The alternative presented here is the ALE + FxLMS system, where an Adaptive Line Enhancer (ALE) is used as decorrelating stage for the FxLMS algorithm. The single-channel case (one reference signal, one actuator and one error sensor) and three different extensions of the system to the multiple channel case are presented and evaluated. The proposed system has proven to be able to achieve faster convergence with reference to the single FxLMS, without decorrelating pre-processing.