一种基于段间距离测度的语言自动分割方法

根据连续语音中不同类语音段之间内在的相异性，提出了段间距离的概念，并基于特征矢量距离给出了段间距离的一种测度。在此基础上，提出一种新的语音自动分割方法，它根据语音段间距离曲线的峰值点对语音音素进行分割，而不需要关于语音单元的先验知识或参考模型，在该方法中，语音特征矢量按帧计算，通过采用适当的帧重迭度，可以使语音分割具有较高的时间分辨率，实验结果表明，该方法具有较高的分割精度和速度。

Automatic Speech Segmentation Method Based on Measurement of the Distance between Segments

Based on the inherent discrimination among speech segments belonging to various classes, this paper proposed a concept of the distance between segments, presented a measurement of the distance between segments based on the distance of feature vector. Based on it, a new method of automatic speech segmentation was proposed. By this method, the phonemes in a piece of continuous speech are segmented according to the peaks of its curve of the distance between segments, and no a priori knowledge or reference models of speech units are required. The feature vectors of speech are calculated in time frames. Adopting the appropriate overlap of frames, the segmentation has high time resolution. The results of experiment indicate that the method has high precision and speed of segmentation.