蒙古语语音识别相关问题研究

发布时间：2018-03-11 08:03

本文选题：蒙古语　切入点：自动注音器　出处：《内蒙古大学》2013年博士论文　论文类型：学位论文

【摘要】：本文将在前人研究的基础上,用实验语音学的理论和方法,以及计算语言学的方法,分析在蒙古语标准音语音识别研究中所遇到的相关问题以及朗读文本语料库设计、蒙古语标准音语音库建立、注音词典的建立、自动注音器的建立、语音库的注音、建立蒙古语语音识别系统等的相关问题。本论文的研究成果主要有：一、建立了蒙古语语音识别的语音数据库和相应的标注文本库。该库包含3299条句子,共30多个小时的语音以及相应的标注读音文本。二、建立了蒙古语注音词典。该词典包含2万5千多个词条,是由蒙古语书面语转写为相应蒙古语标准音的自动注音器的核心之一,并且是蒙古语语音识别系统的核心资源基础。三、研制了蒙古语自动注音器。其功能为完成由蒙古语书面语到音标的实时动态转换。该研究也填补了无蒙古语自动注音的空白。四、建立了基础的蒙古语标准音语音识别系统。五、从蒙古语自身的语音学特点出发分析和研究了影响识别率的原因。针对蒙古语语音识别错误造成的原因,提出了相应的解决方法和建议。六、提出了蒙古语大词汇量连续语音识别系统的难点问题。
[Abstract]:On the basis of previous studies, this paper uses the theories and methods of experimental phonetics and computational linguistics to analyze the problems encountered in the study of Mongolian standard phonetic speech recognition and the design of a corpus of reading texts. The establishment of Mongolian standard phonetic corpus, the establishment of phonetic dictionary, the construction of automatic phonetic device, the construction of phonetic corpus, the establishment of Mongolian speech recognition system and so on. The main research results of this thesis are as follows:. The main contents are as follows: (1) A phonetic database of Mongolian speech recognition and a corresponding annotated text library are established, which contains 3299 sentences with a total of more than 30 hours of speech and corresponding tagged phonetic texts. Second, a Mongolian phonetic dictionary was established. The dictionary contains more than 25,000 entries and is one of the cores of the automatic phonetic device, which is translated from Mongolian written language into the corresponding Mongolian standard pronunciation. And it is the core resource foundation of Mongolian speech recognition system. Third, a Mongolian automatic phonetic phonetic device is developed. Its function is to complete the real-time dynamic conversion from Mongolian written language to phonetic alphabet. This research also fills up the blank of non-Mongolian automatic phonetic phonetic system. Fourth, established the basic Mongolian standard phonetic speech recognition system. Fifthly, this paper analyzes and studies the reasons that affect the recognition rate based on the phonological characteristics of Mongolian language, and puts forward the corresponding solutions and suggestions for the causes of Mongolian speech recognition errors. Six, put forward the difficult problem of Mongolian large vocabulary continuous speech recognition system.
【学位授予单位】：内蒙古大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：H212

【参考文献】