A Kalman Smoothing Algorithm for Speech Enhancement Based on the Properties of Vocal Tract Varying Slowly
Hui Li, Xin Wang, Beiqian Dai, Wei Lu · 2007
The linear prediction coefficients obtained from noisy speech have an important impact on improving the quality of the enhanced speech in the speech enhancement algorithm based on the Kalman Smoother. According to the properties of the slow changes of the vocal tract, this paper proposes a novel Kalman smoothing algorithm for speech enhancement based on vocal tract parameters smoother. Firstly, the linear prediction coefficients are converted into the line spectrum frequency parameters. Then, these parameters of the adjacent frames are smoothed before they transform into state transition matrix. Experimental results indicate that the proposed Kalman smoothing algorithm for speech enhancement based on vocal tract parameters smoother can suppress the sudden changes of residual noise energy and improve the quality of enhanced speech. The quality of the enhanced speech is evaluated by means of segmental SNR and ITU-PESQ scores. Experimental results indicate that the proposed algorithm achieves obvious improvements compared with conventional Wiener filter.