Statistical Model-Based Voice Activity Detection Algorithm in DCT Transformation Domain

摘要

A statistical model-based voice activity detection (VAD) algorithm in discrete cosine transformed (DCT) domain is proposed. This algorithm uses a Hidden Markov Model (HMM) with two states to estimate the probability of voice activity and employs the decision-directed (DD) parameter estimation method for the likelihood ratio test. Tiicu develops a smoothed likelihood ratio (SLR) instead of conventional likelihood ratio (LR) in order to alleviate the delayed term of LR. Simulation results show that the proposed VAD outperforms the G.729 B VAD and the traditional discrete Fourier transformation (DFT) based VAD in various noise environments.