Subband Kalman filtering incorporating masking properties for noisy speech signal

This paper considers a subband Kalman filtering scheme that incorporates the auditory masking properties for single channel speech enhancement. It attempts to achieve high quality enhanced speech by optimizing the trade-off between speech distortion and noise reduction. The use of Kalman filtering in the subband instead of the full-band domain leads to considerable complexity reduction and performance improvement. We propose a novel approach to incorporate the masking threshold with subband Kalman filtering, whereby the estimate of the noise variance that is used in the Kalman filtering process in each subband is modified according to the masking threshold. We adopt an iterative scheme for the estimation of autoregressive (AR) parameters. We investigate, through simulations, the proposed approach by studying the contributions from different functions including Kalman filtering, subband decomposition and perceptual effect based on masking threshold. At the same time, we examine the optimal configuration for the proposed scheme. The proposed approach leads to better enhancement results as compared to the full-band and the conventional subband Kalman filtering methods. Through intensive simulations, we show that our proposed enhancement scheme outperforms various existing well known schemes in terms of objective as well as subjective measures.