Noise variance speech estimation for Kalman filtering of noisy speech

this paper proposes an algorithm that adaptively estimates time-varying noi se variance used in Kalman filtering fot real-time speech signal enhancemen t. In the speech signal contaminated by white noise, the spectral component s except dominant ones in high frequency band are expected to reflect the n oise energy. Our approach if first to find the dominant energy bands over s peech spectrum using LPC. We then calculate the average value of the actual spectral components over the high frequency region excluding the dominant energy bands and use it as noise variance. The resulting noise variance est imate is then applied to Kalman filtering to supress the background noise. Experimental results indicate that the proposed approach achieves a signifi cant improvement in terms of speech enhancement over those of the conventio nal Kalman filtering that uses the average noise power over silence interva l only. As a refinement of our results, we employ multiple-Kalman filtering with multiple noise models and improve the intelligibility.