摘要
首先介绍了大规模语音语料库以及基于大规模语音语料库的文语转换技术的研究现状,接着介绍了一个大规模连续汉语语音语料库的实例Slib的结构和内容;在此基础上,讨论了面向大规模语音语料库的索引技术,提出了语料库检索中的集合运算和最小包容问题,证明了最小包容问题是NP完全的,给出了求解该问题的贪婪算法以及算法的近似比;最后,讨论了基于集合运算的大规模语音语料库的检索技术在文语转换系统中的应用,特别是在基本语言单位实例的选取问题上实现了一种基于最小包容的优化方法,对提高文语转换系统的自然度有实用价值.
The recent advances of large-scale speech corpus （LSSC） and text-to-speech （TTS） technologies are briefly reviewed,then the architecture and annotation information of a large-scale speech corpus Slib are introduced.Based on Slib,the LSSC-oriented indexing methods is discussed,the set operations and the minimum cover problem related to information retrieval in LSSC are presented.The minimum cover problem is a NP-complete problem,and a greedy algorithm is proposed to obtain an approximation solution.The approximation ratio of the proposed algorithm is analyzed.The application and realization of set operations in TTS are presented,and an approach for choosing proper speech instances of linguistic units based on minimum cover is developed,which can improve the naturalness of the synthesized speech of TTS system.
出处
《计算机学报》
EI
CSCD
北大核心
2010年第4期687-696,共10页
Chinese Journal of Computers
基金
国家自然科学基金(60572125)资助~~
关键词
语音语料库
集合运算
文语转换
最小包容
信息检索
speech corpus
set operation
text to speech
minimum cover
information retrieval
