社科网首页|客户端|官方微博|报刊投稿|邮箱 中国社会科学网

您现在的位置:今日语言学 → 语言学系 → 学位论文摘要 → 硕士

梁霄:基于声学参数的汉语方言声调系统研究【语言学及应用语言学】

作者:梁霄 导师:熊子瑜  来源:科研处  时间:2016-10-27

摘 要

  随着语言学研究方法的日益进步与完善,关于汉语方言声调的研究越来越多地采用新的思路和手段。本研究依据《方言调查字表》以及现代汉语常用字字表整理出的《常用汉字记音字表》,尝试针对中原官话区、粤语区、北方官话区、西南官话区的五个方言点数据进行声调系统自动判别分析,试图用精简、科学、合理的方式为汉语方言声调系统研究提供一种新的思路。同时,通过实验数据,探讨方言字音数据的处理方法问题。例如,不同的音高换算方法和不同的音高规整办法是否对声调判别结果产生显著影响等。

  本文正文分为五个章节。本文第一章主要介绍语料设计,第一节介绍字表设计,包括记音字表的设计、录音字表的制作以及本文实验选取的五个方言点情况等;第二节主要介绍录音工作,包括录音工具的介绍、录音工作的大体流程等;第三小节介绍数据的标注与检查整理等。本文第二章为音高数据处理办法的讨论,主要讨论以下几个问题:未校准音高数据与校准后的音高数据对声调判别的影响有多大差距?哪种或哪些声学参数对声调判别的作用最为突出?不同的音高换算方法和规整办法对声调判别结果会有多大程度地影响?从而为实验中选取最为有效的分析参数和分析方法提供依据。实验结果得出,未校准的音高数据与校准后的音高数据在进行声调类型判别分析时所得到的结果未见显著性差异;10个音高点数据和音高段时长是判别调类最有成效的声学参数,对单个发音人的字音数据而言,无需进行音高换算和规整,用原始的基频数据即可得到较高的判别正确率。本文的第三章主要是利用SPSS的决策树分析,并从古今字音演变的角度,讨论汉语方言古今声调系统的演变规律.并基于决策树模型,分析讨论不同方言点各类声调的演化。结果表明,采用古调类与古声清浊这两类因素可以较好地建立起各方言的声调系统演变规律。本文第四章根据古音信息、音高数据和音高段时长等声学参数,训练建立FFN方言声调分析模型和系统,并细致比较感知调类(根据实际读音感知的调类称之为感知调类)和判别调类(根据声调模型训练结果,对每个汉字条目的判别结果称之为判别调类)的差异程度,并在此基础上探讨开展声调系统自动分析的可行性。本文第五章为结论和讨论。

  根据实验结果分析得出基于FFN模型的声调系统分析是基本可行的,但也还存在一定的改进空间。

  关键字:声调系统、古今声调演变、决策树、声调判别模型

 

ABSTRACT

  Along with the increasing of the research methods of linguistics, more and more research about the tone of Chinese dialects based on new ideas and methods. This study is on the basis of “Character Table of Dialect Investigation" and the Basic Vocabulary Table of Modern Chinese Characters, trying to analysis the data from Zhongyuan district, Cantonese district, northern Mandarin Chinese and Southwestern Mandarin in automatically way, and trying to provide a new train of thought by a scientific and reasonable way for the study of the tone system of Chinese dialect .At the same time, this paper discusses the way for dealing with the dialect data. For example, whether different pitches conversion method and different sort has different impact on tone distinguish or not.

  This text is divided into five chapters. The first chapter introduces the design of corpus, the first section introduces the design of the vocabulary table, including the design of the table, making recording-table and the situation of the five selected dialects etc.; the second section mainly introduces the recording work, including recording instruments, the whole process of recording, the third section describes the data. The second chapter discusses the processing method of pitch data, about the following problems: the difference between none-calibration data and the calibration data of pitch, what kind of parameters play the most prominent role in tone recognition? Different affects between different conversion method and structured method of pitch. The answer is data without calibration as same as the data with calibration;10 points of pitch data and duration paly the most important role in the process;initial data can use directly.The third chapter analysis by using SPSS tree, and discusses the evolution of various tones of different dialects. The results show that the answers are ancient tone and QingZhuo.In the fourth chapter, according to the sound information, training to establish the model dialect tone system called FFN, and comparing the feel tone (according to the actual pronunciation) and the distinguish tone (according to the result of tone model training) . In the fifth chapter is for the conclusion and discussion.

  According to the experimental results, it is concluded that the analysis of the tone system based on the FFN model is feasible, but there is still some problems for improvement.

  KEY WORDS: tone system, the evolution of the ancient and modern tones, SPSS tree, the discrimination model of tone

友情链接

COPYRIGHT © 2017

中国社会科学院语言研究所版权所有

京ICP备17005063号-1

邮编:100732

地址:北京市东城区建国门内大街5号

电话:010-85195379

Mail:lingcass@yeah.net