【摘要】 为面向语音评测的研究构建韵母分类器,计算了汉语韵母之间语音参数的DTW距离,使用层次聚类的方法构建韵母识别路径的树状图。在此基础上,通过树状图把韵母的识别过程转换为在树状图上的二元搜索过程。使用支持向量机对每个搜索节点构建两类分类器,实现汉语韵母的分类,大部分韵母的识别率在90%以上。更多还原

【Abstract】 The classification of Mandarin finals is studied for pronunciation scoring.Decision trees of Mandarin finals are constructed using multi-layer clustering methods based on DTW distances of acoustic features.Two-class classifiers are constructed by support vector machine to check the nodes.Mandarin finals are identified according to a specific binary search process in decision trees.The identification accuracy of most finals is above 90%.更多还原