摘要
针对高表现力情感语音合成的需要,设计并建立了一个具有四种情感状态(正常、喜悦、忧伤、愤怒)的高表现力情感语料库。该数据库包含1000个词组、700条语句,共30分钟的语料。论文分析并讨论了设计和建立情感语料库的几个重要准则:语料的选择、制定标注规范、安静的录音环境、情感的情景准备以及情感语料库经主观辨听的后期筛选。为达到高表现力,论文着重强调了情境诱发的重要性,主观听觉评价表明,本数据库具有明显的情感特征。
In view of the high performance emotional speech synthesis need,a high performance emotion corpus which has four kinds of emotional state(normal,happiness,sadness,anger) is designed and built.The database contains 1000 phrases,700 statements,a total of 30 minutes of corpus.This paper analyses and discusses some important guidelines for the design and construction of emotional corpus:corpus selection,standard,quiet recording environment,emotional scene preparation and emotion corpus by subjective listening post screening.In order to achieve high performance,this paper emphasizes the importance of context induced.The subjective evaluation shows that the database has the obvious characteristics of emotion.
出处
《计算机与数字工程》
2014年第8期1383-1385,1453,共4页
Computer & Digital Engineering
基金
江苏省自然科学基金项目(编号:BK20131196)
国家级大学生创新项目(编号:201210285032)资助
关键词
情感语料库
情感分类
主观辨听
语料标注
emotional corpus
sentiment classification
subjective hear
corpus tagging