Frame-based subband Kalman filtering for speech enhancement

Kalman filtering is an effective speech enhancement technique, in which speech and noise signals are usually modeled as autoregressive (AR) processes and represented in the state-space domain. Since AR coefficient identification and Kalman filtering require extensive computations, practical implementation of this approach is difficult. This paper proposes a simple and practical scheme that overcomes these problems. Speech signals are first decomposed into subbands. Subband speech signals are then modeled as low-order AR processes, such that low-order Kalman filters can be applied. Enhanced fullband speech signals are finally obtained by combining the enhanced subband speech signals. Using a frame-based algorithm, autocorrelation functions of subband speech are calculated and the Yuler-Walker equations are solved to obtain the AR parameters. Simulation results show that Kalman filtering in the subband domain not only greatly reduces the computational complexity, but also achieves better performance compared to that in the fullband domain.