Voice Activity Detection Based on Distance Entropy in Noisy Environment
Huan Zhao, Zhao Li-Xia, Kai Zhao, Gang‐Jin Wang · 2009
Voice activity detection is considered as a crucial part of the speech signal processing. In order to improve the accuracy of voice activity detection under the high-noisy environment, an algorithm named distance entropy is proposed. The algorithm firstly enhances speech with short time spectral amplitude, and then utilizes the robustness of cepstral distance and spectral entropy. The experimental results show that this method performs well on anti-noise, and is more accurate to detect the endpoint in low SNR environment.