A single microphone noise canceller based on an adaptive Kalman filter

This paper deals with the problem of adaptive noise cancellation (ANC) when only corrupted speech signal by an additive noise is available for processing. A new method for analysis of speech signal based on a representation in term of an autoregressive (AR) process with variable order described by a state-space model is presented. An algorithm for the minimal realization of this model is proposed. The state transition matrix that characterizes the speech signal is determined by minimizing the mean squared error in the estimation of the speech signal by Kalman filtering. This paper presents an alternative solution that does not require the explicit estimation of the noise and the driving process variances using a new formulation of the approach proposed within a control literature framework by Mehra.