Fractal characteristic-based endpoint detection for whispered speech

In this paper, a fractal based approach is proposed to detect endpoints in whispered speech. The underlying principle is based on the fact that whispered speech is sufficiently chaotic and thus can be analyzed using fractal theory. Due to the different scope of fractal dimensions of silence, noise and speech segment, speech/non-speech segment could be determined from that with a simple decision scheme. The results of experiments, which have been performed on two databases and different methods, show the efficiency of the proposed whispered speech segmentation algorithm.