Accurate speech segmentation via the improved short-time fractal dimension / 西安交通大学学报·英文版

ABSTRACT

Objective: To improve the accuracy of speech segmentation through the improved short-time fractal dimension. Methods: An equation was established for window size selection of speech analysis. Dynamic Window Step (DWS), a novel method to determine the sliding window steps adaptively in agreement with the local properties of signals, was proposed. Results: The influence of the window step on the short-time fractal dimension was discussed. Compared with fixed window steps, more accurate and efficient fractal dimension trajectories were obtained with dynamic window steps. Conclusion: The proposed method was applied to a number of speech signals. It shows promise in speech segmentation, speech recognition and other transient signal analysis.