汉语重音的凸显度分析与合成

The prominence analysis and synthesis of emphasis in Putonghua

重音是重要的语调特征,重音合成技术可以提高语音的自然度和表现力。针对重音的局部凸显性,该文提出了声学特征凸显度的表示方法,分析了不同韵律位置(韵律词首、中、尾,韵律短语首、中、尾等)重音音节的声学特征凸显度,发现在韵律单元末(韵律词末音节和韵律短语末韵律词)的重音其基频最大值凸显度要低于非韵律单元末重音,提出了基于声学特征凸显度的非线性的重音声学参数生成算法,解决了传统重音声学参数线性修改算法的修改幅度不足或过大的问题。采用该算法建立了基于隐Markov模型的支持重音合成的语音合成系统。实验表明,该系统可以有效合成带有重音的语音,提高了合成语音的自然度和表现力。

Emphasis is an important feature of intonation. The technology of emphatic speech synthesis can improve the naturalness and the expressiveness of the synthesized speech. This paper defined the prominences of acoustic features, analyzed the prominences of the acoustic features of the emphasized syllables in different prosody locations, e.g. the head, the body, and found that the prominences the maximum F0 of the emphasized syllables at the end of prosody units（prosody words or prosody phrases） are lower than other emphasized syllables. A parameter generation algorithm of emphasized syllables based on the prominences of the acoustic features was proposed based on the analysis. It avoided the problem of oversize modification of the traditional linear modification algorithm. An emphatic speech synthesis system based on hidden Markov model （HMM） was built with the proposed algorithm. Experiments demonstrated that the system could synthesize emphasized speeches and improve the naturalness and the expressive of the synthesized speeches.