Efficient voice activity detection algorithm based on sub-band temporal envelope and sub-band long-term signal variability

Voice activity detection (VAD) is widely used for various speech-based systems which is an important pre-processing step. This paper proposes a robust voice activity detection algorithm. In the proposed algorithm, the sub-band temporal envelope and the sub-band long-term signal variability are considered to distinguish the speech from all kinds of non-speech which include stationary noise and non-stationary noise. The two features are combined to make a robust VAD decision according to the fusion decision. The proposed algorithm also is an unsupervised low-complexity algorithm and can operate without pre-train models. The experiments results show that the proposed algorithm is prior to the different baseline algorithms and can handle a variety of noise environments over a wide range of signal-to-noise ratios. The proposed algorithm could apply to speech-based systems.