【作者】 包叶波； 胡郁； 刘聪； 江辉； 戴礼荣； 刘庆峰；

【Author】 BAO Yebo~1,HU Yu~2,LIU Cong~2,JIANG Hui~3, DAI Lirong~1,LIU Qingfeng~(1,2) (1.Department of Electronic Engineering and Information Science, University of Science and Technology of China,Hefei 230027,China; 2.Anhui USTC iFLYTEK Company Ltd,Hefei 230088,China; 3.Department of Computer Science and Engineering, York University,Toronto M3J 1P3,Canada)

【机构】 中国科学技术大学电子工程与信息科学系； 安徽科大讯飞信息科技股份有限公司； 约克大学计算机科学与工程系；

【摘要】 在识别系统中,建模单元能够勾画一种语言的声学和语音学特性,因此对系统性能起到至关重要的作用。该文参照一些已在大词汇量连续语音识别系统(LVCSR)中取得较好效果的建模单元集,构建了新的音素建模单元集(NewPS)。另外,根据NewPS中元音及其变体对前后接音素协同发音的影响,提出了基于扩展的元音三角图设计问题集(NewQS)的方法。实验表明:NewPS和NewQS结合的识别性能超越了传统的声韵母建模单元集;并且,建模单元数目大幅度的减少给系统后续模块的处理带来了便利。更多还原

【Abstract】 Modeling units can be used to describe the salient acoustic and phonetic information for a language in speech recognition systems.Thus,they play a very important role in the system.This paper describes a phoneme set using several modeling units,which has good performance in large vocabulary continuous speech recognition(LVCSR) systems.A question set design method is given based on the extended vowel triangle.Tests show that the combination of the new phoneme set and the new question set surpasses the initial/final in performance.AJso,the number of modeling units is greatly reduced which is more convenient for processing succeeding system modules.更多还原