【作者】 周杨；

【导师】 尉迟治平；

【作者基本信息】 华中科技大学 ， 语言学及应用语言学， 2008， 博士

【副题名】以黄孝片方言为例

【摘要】 计算机汉语方言辨识是计算机自动判别说话人所讲汉语方言语音片段的方言属性,并据此辨别说话人所属地域的一项技术。这项技术在言语工程、公共安全、语言研究等众多领域均具有十分重要的价值。这是一项跨学科的综合研究,融合了语言学、心理学、数学、计算机技术、模式识别和语音信号处理等诸多学科的相关内容。言语工程领域的汉语方言辨识研究主要基于言语产生模型,采用概率统计方法进行系统的设计和实现,在研究对象、方法及功能等方面都存在一定缺陷和问题。“基于语言的汉语方言辨识”建立在汉语方言的语言学研究之上,将感知模式和语言规律作为方言辨识的重要依据,能够较好地克服上述缺点。基于语言的汉语方言辨识研究的研究重点是为汉语方言建立能反映其语言学性质的数学模型,而研究难点在于如何选取用于方言辨识的方言语音特征以及如何实现语音特征物理属性、心理属性和语言学属性的统一。通过深入分析汉语方言辨识所涉及的相关语言理论问题,特别是汉语方言语音系统性、结构性的特点,首次提出了一种基于“音类—音值”结构的汉语方言数学模型,并在此基础上构造了方言语音特征“音类—音值—区别性特征”的数学模型,从而为方言辨识系统的构建打下良好的理论基础。针对方言语音特征的相关问题,通过对汉语方言辨识认知模式的分析以及对语音区别性特征的探讨,提出方言语音特征的内部必须包含最大区别性差异,通过这种区别性差异来克服声学性质与感知存在的矛盾。同时针对不同类型的方言语音特征,提出处理方言语音特征的求值型策略和比较型策略;并提出利用区别特征检测器法和频域比较法来提取方言语音特征中的区别性特征。在此基础上,提出一个汉语方言辨识系统的基本框架。该系统利用具有语言学意义的方言语音特征进行方言辨识。这些方言语音特征在汉语方言辨识系统中依据一定的层级和顺序运作,从而逐级确定方言的属性;同时依据方言语音特征在地域的分布特点,辨别方言所属县一级行政区域;在方言特征混杂的情况下,则利用核心特征和共同特征确定方言的地域归属。在上述模型和框架的基础上,通过田野调查和录音制作了黄孝片20余个方言代表点的语音数据库和语音音档,并利用黄孝片方言作为汉语方言辨识理论研究的主要材料。通过对黄孝片方言语音进行了全面、系统的分析和研究,从中提取了30余组语音特征,逐一探讨这些语音特征的性质、来源以及在汉语方言辨识中所起的功能与作用;并从这些语音特征中归纳出10余对区别性特征,通过实验语音分析指出这些区别性特征的声学对应物和具体的处理方法。在考察黄孝片方言语音特征时,注重将这些语音特征置于汉语方言的宏观背景下,细致考察语音特征在汉语方言中的分布及其所包含的类型意义。针对黄孝片方言语音特征中的一些具体问题,利用多种方法进行了全面细致的研究。如对黄孝片方言?韵系的音系特点、语音性质及在汉语方言的分布规律进行了全面的考察,提出?来源于^*u(<^*iu<^*iu)的看法;利用声学语音实验的方法对黄孝片方言入声韵尾、松紧元音、鼻韵尾等问题进行了深入的分析。同时从行政区划沿革、移民史等几个方面较全面地分析了黄孝片方言的来源、性质和归属;并根据全浊入声的归并规律等特征将黄孝片方言划分为三个小片。从理论上看,利用本文提出的基于语言的汉语方言辨识系统及其一整套的解决方案,能够较好地实现以非连续语音、特定词汇为对象,以确定方言关系及归属地为目标的汉语方言辨识。但是汉语方言辨识系统的构建是一个复杂的系统工程,目前由于各方面条件的限制,未能最终完成方言辨识系统的实现,这有待于各领域专家的进一步探索。更多还原

【Abstract】 Chinese Dialect Identification is a technique that the computer can distinguish automatically the category of the dialect of the voice fragments of Chinese dialect pronounced by the speaker, and according to this, we can identify the geographical origin. In the Speech Project, Public Safety, Language Study and other fields, the technology is of great value. This is a comprehensive interdisciplinary research, the convergence of relevant contents of linguistics, psychology, mathematics, computer technology, pattern recognition and voice signal processing, and many other subjects.In the field of speech project, the research on Chinese Dialect Identification is mainly based on Speech Generation Model, and to design and accomplish the system according to the method of probability and statistic, but there are still some deficiencies and problems in the object of research, method and function, etc. Chinese Dialect Identification based on language can better overcome these shortcomings because it is based on the language study of the Chinese dialects and makes the mode of perception and the rule of language an important basis for identification dialect.The research core of Chinese Dialect Identification based on language is to establish a mathematics model for Chinese dialect which can show its quality of linguistics, and the research difficulty is how to choose the phonology feature for dialect identification and how to accomplish the unification of the physical category, linguistic category and psychological category of the phonology feature.Through deep analysis on relevant language theory problems related to Chinese Dialect Identification, especially the voice systematic, structural characteristics of Chinese dialect, we put forward a mathematics model of Chinese dialect based on the phonology feature of“sound kind-sound value”for the first time. And on the basis of it we proposed a“sound kind-sound value-distinction feature”model for the Chinese Dialect phonology features.These models build the theoretical foundation for the Chinese Dialect Identification System.About the relevant problems of phonology feature of dialect, we put forward the biggest distinctive differences must be included in sound characteristics of dialect by analyzing the cognitive model of Chinese Dialect Identification and discussing the distinctive characteristics. We overcome the inconsistency between the acoustics character and apperception by the distinctive differences.At the same time, about different types of sound characteristics of dialect, we put forward value-seeking strategies and comparison of strategies to deal with dialect characteristics for dialect identification system, and we can get the distinctive feature of sound characteristics of dialect by the method of distinctive characteristics detector and comparative frequency.Based on this, we put forward a basic framework of Chinese Dialect Identification system. The system identifies dialects according to the phonology features of quality of linguistics. In the Chinese Dialect Identification system, these phonology features of dialect work according to a certain order and sequence, and so we can get the category of the dialect step by step. At the same time, according to the geographical features of the distribution of sound characteristics of dialect, we identify their respective dialects in administrative region of the county level about the dialect features. If the dialect phonology features were mixed, we use the core features and common features to know the dialect geographical attribution.Based on the above models and framework, through field research, we collect, record and make phonology databases and phonetic databases of more than 20 kinds of voice and voice-file database of Huang-xiao Dialects. We use Huang-xiao Dialects as the key materials to study the theory of Chinese Dialect Identification. Through a comprehensive, systematic analysis and research on voices of Huang-xiao Dialects, we extract more than 30 groups of voice features, and study these voice characteristics of the nature, source and the function and role in Chinese Dialect Identification one by one. In addition, we sum up more than 10 distinctive features from the voice features, and point out the acoustic counterparts and specific treatment of the distinctive features through experiments and voice analysis.In seeing about the phonology features of Huang-xiao Dialects, we put the phonology features under the background of Chinese dialects and see about the distributing and typological significance of the phonology features in Chinese dialects. About the detailed problems of phonology features of Huang-xiao Dialects, we make a comprehensive and detailed study through many methods. For example, we see about thoroughly the sound quality, phonology characteristic and distribution rule of ? finals system in? Huang-xiao Dialects, and put forward the opinion of ?? origins? from?^*u(<^*iu<^*iu). In addition, we analyze deeply the problems about ru-sheng finals, Tense/lax vowel and nasal finals in Huang-xiao Dialects through the method of acoustic sound experiment. At the same time, we analyze thoroughly the origin, quality and ascription from the aspects of evolution of districts, the history of emigration and so on. Then we divide the Huang-xiao Dialects into three scraps according to the characteristics such as merger of quan-zhuo ru-sheng.Theoretically, through Chinese Dialect Identification system based on language and the whole scheme advanced in this article, we can better accomplish the dialect identification of non-continuous sound, specific words, and determining the relation of dialects. However, the construction of Chinese Dialects Identification System is a complex systematic project. At present, due to various conditions of the restrictions, Chinese Dialect Identification system can not come true, and it still depends on further study of the experts in various fields.更多还原