A recursive calculating algorithm for higher-order cumulants over sliding window and its application in speech endpoint detection

In order to resolve the problem that the performance of traditional endpoint detection algorithms gets worse under noisy environments, a recursive calculating algorithm for higher-order cumulants over sliding window is proposed which has been applied to the speech endpoint detection. By using the proposed calculating algorithm, the higher-order cumulants of noisy speech signals are estimated. Furthermore, endpoint detection is carried out combined with the feature of energy. Experimental results reveal that the computational efficiency of the recursive calculating algorithm has been promoted remarkably compared with traditional one. Besides, in a variety of noise and signal-noise ratios (SNR) environments, the probability of correct point detection (Pc-point) of the proposed voice activity detection (VAD) obtains the relative increasing of 6.07% on average compared with the VAD of G.729b. The speech endpoint detection method based on higher-order cumulants over sliding window has higher computational efficiency and better robustness.