摘要
从张家騄提出的韵母区别特征树出发,研究区别特征的声学参数。将韵母分为无介音和介音韵母两部分,使用支持向量机检测韵母的区别特征。在此基础上,通过区别特征树上的二元搜索过程实现韵母的分类。每个节点对应于一个区别特征,经过特定的搜索路径,韵母就被唯一确定下来。使用上述方法,大部分韵母的识别率在90%以上。
Acoustic parameters of distinctive feature are studied based on distinctive features framework proposed by Zhang Jialu. Finals are divided into beginning with and without glides. In both cases, two-class classifiers are constructed by support vector machine to check the distinctive features. On this basis, finals are classified through binary search in distinctive features trees. Each node corresponds to a kind of distinctive feature and finals are identified according to a specific search process. The accuracy of most mandarin finals is above 90%.
出处
《电声技术》
2011年第9期38-41,共4页
Audio Engineering
关键词
语音评测
韵母
介音
区别特征
支持向量机
speech evaluation
finals
glides
distinctive features
support vector machine
