作 者: ;
机构地区: 西安外国语大学汉学院
出 处: 《语言科学》 2011年第6期574-585,共12页
摘 要: 提高汉语合成语音自然度和可懂度的最有效、最关键途径在于更准确地预测语句的韵律结构。目前文语转换系统预测语句韵律结构所能依据的内容通常只有语句的词类-音节数序列,文章主要讨论从词类-音节数序列到语句韵律结构预测间的关联,并据此对如何提高韵律结构预测准确率提出了一些建议:词类-音节数序列中包含的"韵律上最可能共存与最不可能共存的组合"以及"能够帮助有效预测停延段信息的虚词"可以提高预测准确率;对序列中某些片段语法结构的确定,可以提高预测准确率;最后,根据韵律分布特征细分词类,也可以提高预测准确率。 How to predict the prosodic structure automatically is one of very important tasks in Chinese Text-to-Speech, and it will make a direct impact on the naturalness of the synthesized speech. The sequence of part-of-speech and number-of-syllable is the base of prediction, and the paper presents the information in the POS-NOS which are effective and useful in achieving the goal. The four suggestions are: the possibility-to-merging of the POS-NOS, functional words available to predict the grammar structure, some mappings between the POS-NOS and structure, and the refining of the criterions to indentify the category according to prosodic distribution.
关 键 词: 词类 音节数序列 文语转换韵律结构预测
领 域: [语言文字]