On the use of Kalman Filter for Enhancing Speech Corrupted by Colored Noise

摘要

Kalman filtering is a powerful technique for the estimation of the speech signal observed in additive background noise. This paper presents a contribution in the enhancement of noisy speech with white and colored noise assumption. Some tests were performed with ideal filter parameters, others using the Expectation Maximization (EM) algorithm to iteratively estimate the spectral parameters of the speech and noise. Simulation results show that the application has the best performance evaluated with objective quality scores, observation of the waveforms, as well as informal listening tests in the case of Noizeus database.