A fractal based voice activity detector for Internet telephone

The fractal dimension is useful in discriminating different textures of images. Also, the distinction between the ´texture´ of speech signals and that of noise should exist because speech signals are regular ´man-made´ processes and noises are random ´natural´ processes. Through experiments, we found that the fractal dimension of speech signals is distinctly different from that of noise. Thus, a fractal based voice activity detector (VAD) is proposed. The experimental results show that the proposed VAD outperforms two traditional approaches based on energy and zero-crossing rate at low SNR. Applied to the Internet telephone developed by us, this VAD improves the quality of service.