A robust voice activity detection technique based on combined framework of lacunarity and empirical mode decomposition

In this paper, we have proposed a new technique for voice activity detection (VAD) using lacunarity index combined with empirical mode decomposition (EMD) technique. In the preprocessing stage of the proposed framework, the noisy speech signal is decomposed into several intrinsic mode functions (IMFs) based on EMD technique. After that more informative IMFs are selected using spectral flatness measurement (SFM) approach. In the decision stage, the speech signal frames are identified by a threshold limit. The threshold value is calculated using statistical parameter of the lacunarity index of the reconstructed speech signal. The proposed technique gives superior results over existing techniques and is more effective at moderate noise levels.