摘要
在语音合成的研究中，韵律表示和韵律学习是改善合成语音质量的关键。本文作者研制了汉语文字－语音转换系统Ｓｏｎｉｃ。在研究韵律描述和模型的基础上，设计了韵律控制符号和韵律模拟算法；通过对文本进行置标，实现了重音和语调的模拟。为了进一步改善合成语音的自然度和表现力，作者改造了Ｓｏｎｉｃ系统，采用神经网络学习算法，按词语和语句两级进行韵律学习，建立了具有韵律学习能力的文语转换（ＴＴＳ）系统Ｓｏｎｉｃ－Ｌ。本文分析了汉语ＴＴＳ技术研究的现状，介绍了作者在韵律学习、描述、模拟方面的研究工作和实验结果。
In the research of speech synthesis, prosody expression and study are the kernels of improving performance of speech synthesis. The authors established a Chinese text to speech (TTS) system, Sonic. The researchers have designed prosody symbols and algorithm, basing on studying prosody descriptions and models, then marked up the text with prosody symbols, and implemented simulation of stress and intonation. To get the better naturalness and expressivity of the output speech, the system have been improved and a prosody learning TTS system, Sonic L, have been developed. In sonic L, learning with neural network is splitted two levels: based on words and based on sentences. The article introduces the actuality of technology of text to speech, describe the author's research and result on prosody studying, description and stimulation.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
1998年第S1期95-98,共4页
Journal of Tsinghua University(Science and Technology)
基金
国家"八六三"高技术项目
