论文部分内容阅读
针对汉语发音特点,基于对大量自然汉语语句基频轮廓数据的统计和分析,提出一种用于数据驱动生成汉语韵律特征的数学模型。该模型以基频参数为主,辅以时长和增益参数,能表现汉语的语气、短语节奏、韵律词声调及轻重音多层韵律信息,各层参数可按语言知识分类训练和标注。给出了模型的各种归一化“调素”函数和变调规则。仿真实验表明了该模型的有效性。
Aiming at the characteristics of Chinese pronunciation, based on the statistics and analysis of the fundamental frequency contour data of a large number of natural Chinese sentences, a mathematical model for data-driven generation of Chinese prosodic features is proposed. The model is mainly based on the fundamental frequency parameters, supplemented by duration and gain parameters, which can display Chinese tone, phrase rhythm, prosodic word tone and multi-level prosody information of light and heavy tones. The parameters of each layer can be trained and marked according to the language knowledge. Various normalized “prime factors” functions and rules of variation are given. Simulation experiments show the effectiveness of the model.