Voice activity detection using higher-order statistics in the teager energy domain
Zhe Song, Tianqi Zhang, Demin Zhang, Tiecheng Song · 2009
In this paper, an effective algorithm employing higher-order statistics (HOS) with Teager energy operator (TEO) for voice activity detection (VAD) in noisy environments is proposed. The presented VAD utilizes the kurtosis of a speech signal in the Teager energy domain and is shown to be efficient and robust in detecting speech in low signal-to-noise ratio (SNR) conditions without being adversely affected by types of colored noises. The use of TEO significantly enhances the discriminability between speech and noise via a nonlinear operator, and overcomes the inability of the HOS in classifying speech and non-Gaussian noise. The results of computer simulations prove that the proposed approach has an overall better performance than the standard ITU-T G.729B VAD, and other recently reported VADs.