多信息融合的汉语方言辨识  

Fusing Multi- information for Chinese Dialect Identification

作　　者：夏玉果[1] 戴红霞[1] 顾明亮[2,3] 徐敏[1] 

机构地区：[1]江苏信息学院电子信息工程系,无锡214153 [2]徐州师范大学物理与电子工程学院,徐州221116 [3]江苏省神经与认知工程重点实验室,江苏徐州221116

出　　处：《电脑与信息技术》2009年第5期32-36,共5页Computer and Information Technology

摘　　要：文章提出了一种融合声学、音素配位和韵律特征等多信息融合的汉语方言辨识系统,分析了将语言信息转化为这些特征的实验方法,在此基础上,根据汉语方言辨识的特点,提出了一种基于概率模型的多信息辨识机制,实验结果表明,韵律特征对于短时语音具有很好的辨识效果,而音位配列特征对于长时语音更加有效。对于汉语三种方言的辨识,融合这三种特征的辨识率达95%。This paper presents a novel Chinese dialect identification method fusing multi-information including acoustic, phonotactic and prosody feature.We firstly analyses the way to translate language information into such features, then introduces a new information fusing model method based on the character of Chinese dialect.The results show that the prosodic features are more effective for shorter speech while the phonotactic features work better for longer speech. For three Chinese dialect, the system with fusion of three feature achieves a high accuracy of 95%.

关 键 词：韵律特征 信息融合 汉语方言辨识 

分 类 号：TN912.34[电子电信—通信与信息系统]