基于短时分形维数的汉语语音自动分段技术研究

本文根据汉语语音的构成特点，提出了一种新的基于短时分形维数的汉语语音自动分段方法。该方法首先用等差尺度网格维数替代传统盒维数计算方法来快速计算语音信号的分形维数，然后在统计、分析汉语男女声２１种声母和３８种韵母语间信号的分形特性基础上，利用中心偏离限定算法来实现汉语语音信号的自动分段。仿真实验表明，该方法不但能正确实现不同语速条件下的语音自动分段，而且具有噪声鲁棒性，是一种有效的汉语语音自动分段技

Study on automatic segmentation of mandarin speech based on short-time fractal dimension

In this paper,a new automatic segmentation technology of Mandarin speech,based on short time fractal dimension,is presented according to the characteristics of Mandarin speech signal First,the conventional Box Dimension is substituted by the Arithmetical Progression Scales Gridding Dimension to reduce the computation cost of short time fractal dimension;Second,based on the statistical analysis of fractal properties of 21 consonants and 38 vowels of Mandarin speech,the automatic segmentation of Mandarin speech was implemented by using the RCDSA(Restricting Centroid Deviation Segment Algorithm) Simulation experiment results show that this method is an efficient automatic segmentation technology of Mandarin speech that not only can segment speech signals with variously articulating velocity correctly,but also has good noisy robustness