摘要
引言文语转换系统具有将输人的文字自动地转换成语音的功能,广泛地应用于文稿校对、语音应答系统、电子邮件的语音服务等领域。根据合成模型的不同,可以将文语转换技术划分为三类川,即基于声学模型、发声模型和对自然语音进行编码模型的文语转换技术。基于声学模型的文语转换技术以语音产生的源一滤波器理论为基础,最典型的是共振峰合成方法;基于发声模型的文语转换技术是通过模拟人的发声过程来合成语音;
A method and principles of building voice bank used in text-to-speech system (TTS), which based on TDPSOLA method and made use of Chinese syllable with tone,were described,The quality of voice bank has direct relation to the clearness and naturalness of TTS system,specially to the system based on the syllable concatenation. In this TTS system,some method were adoptted to ensure the quality of syllable voice bank,such as making detail classifications to some synthesis u-nit, abstracting recording example sentences, recording the voice bank in the quasi-naturalness speech and abstracting the valid synthesis unit,etc. The result showed that the naturalness of the TTS system had been improved greatly.
出处
《计算机科学》
CSCD
北大核心
1998年第6期66-69,89,共5页
Computer Science
