摘要
友好语音是指表达友好态度的语音。本研究是在口语对话中友好语音与中性语音的对比声学分析基础上,进一步分析疑问句和陈述句两种功能语句在表达友好态度时的声学表现,得到不同重音位置下基频 FO 和时长的变化模式,以及由中性语音合成友好语音的合成参数。然后用 MOMEL 语调生成模型和 PSOLA 语音合成方法进行友好语音合成并进行语音感知验证,分析韵律特征基频和时长对友好态度语音哪个贡献大。结果表明:1)表达友好态度的陈述句和疑问句与对应的中性语音比较,基频和时长两个韵律声学参数变化模式不同,而且这种模式还受到句重音的影响;2)可以通过调整合成参数实现友好态度的陈述和疑问句的语音合成;3)基频和时长这两个韵律参数中,基频对友好态度语音合成贡献更大,仅仅调整时长或语速不能实现友好语音的合成;4)表达友好态度时,合成的疑问句比陈述句的感知结果好;5)发音人在有疑问语助词(如"吗")的疑问句中经常使用句末高边界调来表达友好语音。
Friendly speech means the speech expressing friendly attitude.Based on the acoustic analysis on friendly speech and the corresponding neutral speech for an expressive dialogue corpus,this paper reports the analysis results on declarative and interrogative sentences.Pitch and duration patterns of prosodic words were statistically analyzed and compared concerning factors of their positions and stresses.The conversion rules from neutral speech to friendly speech on tonal pitch and prosodic duration parameters were got and the synthesized stimuli produced by using MOMEL model and PSOLA method were subjected to perception test.It was found that:(1)The acoustic patterns of pitch and duration of friendly declarative and interrogative utterances are quite different from those of the neutral utterances,which are varied with the patterns of sentences stress;(2)Friendliness of synthesized speech can be achieved via adjusting the perceptually distinctive acoustic parameters;(3)Tonal pitch is the most important means for abetter expression of friendliness;Only adjusting duration is no use for friendly speech synthesis;(4) Interrogative sentences get higher perceptual results than declarative sentences;(5)A high boundary tone for interrogative sentence is usually used by speakers to express friendly attitude.
出处
《中国语文》
CSSCI
北大核心
2005年第5期418-431,共14页
Studies of the Chinese Language
基金
本课题受到国家自然科学基金支持(项目号60275015) IBM 中国研究中心合作项目基金支持。
