Robust voice activity detection algorithm using spectrum estimation and dynamic thresholding
Efrain Zenteno, Manuel Sotomayor · 2009
An algorithm for voice activity detection (VAD) is presented in this paper, it uses the classical short time Fourier transform (STFT) followed by dynamic thresholding over the noise floor which allows accurate determination for voiced/unvoiced frame. This algorithm has also been implemented in a DSP plataform to evaluate its performance in real time, where results have shown that such method presents accurate voiced/unvoiced determination under unknown signal to noise ratio (SNR) conditions.