摘要
为了研究耳语音情感识别,制定了耳语音情感数据库的制作流程和制作规范,设计并建立了一个包含4000个单词、5000条语句和600个段落的耳语音情感数据库。利用模糊数学中的隶属度函数和层次分析法对语句的有效性进行了检验,得到一个满足研究要求的耳语音情感数据库。利用库中的语料进行初步的耳语音情感识别,得知正常音语音情感识别所用的音长、能量等参数,在耳语音中仍然可以使用;而正常音的其它涉及基音的一些参数,不适用于耳语音,需要进行改进或者提取新的特征参数。
In order to study the new issue of emotion recognition of whispered speech, the construction flow and standards are developed, and the whispered speech emotion database, which includes 4000 words, 5000 sentences and 600 paragraphs, is designed and established. With the concepts of Membership Function and Analytic Hierarchy Process in fuzzy math, the validity of each sentence has been checked out to obtain a qualified database with which a preliminary experiment of whispered speech emotion recognition is carried out. It is concluded that the parameters such as sound length and sound power, which are suitable for normal speech emotion recognition, are still useful in whispered speech, while the other parameters related to pitch are not suitable for whispered speech emotion recognition. So the old parameters should be improved or new parameters would be extracted.
出处
《声学技术》
CSCD
2010年第1期63-68,共6页
Technical Acoustics
基金
国家自然科学基金项目(60872073
60472058)
江苏省自然科学基金项目(BK2008291)
作者简介
通讯作者：金赟（1979-），男，浙江乐清人，博士研究生，研究方向为语音信号处理。E-mail：jinyun9999@yahoo.com.cn
