Abstract: In order to improve the ability of anti-noise and fractal detection of short time TEO energy, a new algorithm about voice activity detection is proposed which can adapt to a variety of noise environments. The algorithm use the feature of energy distribution characterization of the fractal dimension in the frequency domain and TEO characteristics of zero mean noise elimination and characterization of formant characteristics, add the calculation of fractal dimension to the twice judge noise and voice part based on short-time energy method TEO. Compared with the traditional voice activity detection algorithm, this algorithm is easy to realize and with a high accuracy. The experiments show that in different background noises, the algorithm has a better robust performance compare to the single voice activity detection algorithm, and improve the performance of single algorithm for anti-noise. it can adapt to a variety of noise under the low SNR environment.