基于预分类的声母客观评测方法  

Study of initial objective assessment based on Pro-classification

作　　者：汤霖[1] 彭土有[1] 尹俊勋[2] 

机构地区：[1]江门职业技术学院电子与信息技术系,广东江门529090 [2]华南理工大学电信学院,广东广州510641

出　　处：《湖南科技大学学报（自然科学版）》2012年第1期88-93,共6页Journal of Hunan University of Science And Technology：Natural Science Edition

基　　金：国家自然科学基金资助项目(50875086);广东省自然科学基金资助项目(8152900006000002)

摘　　要：详细分析了语音评测中的基本策略——对数后验概率评价策略,提出利用语音自身的特征相似性组织判决音素集的方法,既提高了评测精度又减少了计算量.在分析普通话声母显著性特征的基础上,提出了基于预分类的评测方法.同时,根据音征互载的特点,提出了声韵结合和韵声韵结合的音素模型.声母评测系统取得83.56%的评测准确率,基本达到人工评测水平.The details of basic strategy of speech assessment（strategy of log posterior probability） was carefully analyzed,thus,the method of obtaining phonemes decision set was proposed by utilizing the character similarities of speech itself.The assessment accurate was promoted accompanying with the decreasing of calculated amount.Based on the distinct characters analysis of the Initial of Putonghua,the assessment method based on pro-classification was delivered.Meanwhile,the joint models of Intial-Final and Final-Initial-Final were established depending on the features of Coarticulation.83.56% assessment accurate was achieved in the Initial assessment system,and it reached the assessment level of human.

关 键 词：语音评测 客观评测 语音处理 计算机辅助语言学习 普通话水平测试 

分 类 号：TN912.3[电子电信—通信与信息系统] TP391.7[电子电信—信息与通信工程]