基于维吾尔语单词清、浊音组成结构特征的连续语音单词切分算法

Research on word segmentation algorithm based on Uyghur phoneme acoustic feature

提出了面向关键词检索的连续语音单词切分算法.算法的核心是一组多元高斯分布,其5个参数分别由语音信号的短时能量、短时过零率、短时自相关系数、第一个预测系数、预测误差归一化能量等声学特征获取.证明了基于单词声学特征的连续语音单词切分比等宽单词切分具有更好的单词切分及检索效率;讨论了优化算法的构思.

This paper introduces continues speech word segmentation algorithm used for key-word spotting. The core of the algorithm is based on a multivariate Gaussian distribution, five parameters can be extracted by the short-time log energy, the short-time zero-crossing rate, the short-time autocorrelation coefficient, the first predictor coefficient of LP and the normalized energy of the prediction error of a LP. The test outcomes show that the word segmentation algorithm based on Uyghur phoneme acoustic feature is better than mono-spaced word segmentation. Finally, some ideas to optimize algorithm are given.