Improved the Efficiency of Generalized Sidelobe Canceller Algorithm by using Speech Presence Probability

Abstract

Speech signal processing application always encounter certain difficulties in real complex environment. The captured signal on microphones often interfered by coherent, incoherent, stationary, non-stationary noise and self acoustic mismatch. To solve this problem, the necessary requirement is speech enhancement to extract target speaker from observed signals in condition minimum speech distortion, while removing background noise. The author proposed a speech enhancement generalized sidelobe canceller based on an estimation of speech presence probability. Main ideal of the algorithm is accuracy estimation of auto and cross power spectral densities of main and reference signal, which used in process of filtering. The experimental result ensures the effectiveness of the proposal algorithm, the background noise is suppressed while the quality of speech is improved in compared with the conventional generalized sidelobe canceller. The proposed algorithm can be evaluated as a frontend for automatic speech application.