连续语流中重音标注和大规模重音语料库建设  

Stress annotation in continuous speech and large-scale stress corpus construction

作　　者：卢颖超[1,2] 李雅[2] 陶建华[2] 许小颖[1,2] 

机构地区：[1]北京师范大学文学院,普通话水平培训测试研究中心,北京100875 [2]中国科学院自动化研究所,模式识别国家重点实验室,北京100190

出　　处：《清华大学学报（自然科学版）》2011年第9期1271-1275,共5页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金面上项目(60873160,61011140075);国家自然科学基金重大研究计划项目重点课题(90820303);北京师范大学教改课题(004-127028);中央高校基本科研业务费专项资金(2010105565004)

摘　　要：规范的重音标注将对重音研究和语音合成技术产生积极影响。该文通过对大规模语料进行语流中词重音和句重音标注,总结出了重音标注的规则。重音标注包括标注训练、正式标注、语料库建立3个步骤。通过标注训练明确重音概念和标注方法,词重音采取词内对比法,所谓"重"主要指突显度和饱满度;句重音采取呼吸句群单位标注法,注重标注的层级性,不跨句群进行重音知觉对比。重音标注完成后,重音语料库也随之建立起来。这个大规模的重音语料库具备词重音和句重音信息,适用于语音合成领域,能较好地提高合成语音的自然度。Standard stress annotation is needed for stress studies and speech processing systems.This study provides word stress and sentence stress annotations of a large-scale speech synthesis corpus and summarizes the annotation rules.The stress annotation uses annotation training,normal annotation and corpus construction.The annotation training helps the annotators understand the stress concepts and annotation rules.The word stresses were annotated by comparison of each syllable in the prosodic word,with the ＂stress＂ referred to as the prominence degree.The sentence stress used the breath group as an annotation unit with emphasis on the annotation hierarchy.The annotation work gave a two-level stress annotated corpus which can be used in speech and language processing systems to improve the naturalness of synthesized speech.

关 键 词：词重音标注 句重音标注 重音语料库 

分 类 号：H11[语言文字—汉语]